Physically Based Evaporative Demand as a Drought Metric ... · Physically Based Evaporative Demand...
Transcript of Physically Based Evaporative Demand as a Drought Metric ... · Physically Based Evaporative Demand...
University of Nevada, Reno
Physically Based Evaporative Demand as a Drought Metric: Historical Analysis and
Seasonal Prediction
A dissertation submitted in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Atmospheric Science
by
Daniel J. McEvoy
Dr. John Mejia/Dissertation Advisor
August, 2015
We recommend that the dissertation
prepared under our supervision by
DANIEL J. MCEVOY
Entitled
Physically Based Evaporative Demand As A Drought Metric: Historical Analysis
And Seasonal Prediction
be accepted in partial fulfillment of the
requirements for the degree of
DOCTOR OF PHILOSOPHY
Dr. John Mejia, Advisor
Dr. Timothy Brown, Committee Member
Dr. Eric Wilcox, Committee Member
Dr. Mike Hobbins, Committee Member
Dr. Justin Huntington, Graduate School Representative
David W. Zeh, Ph. D., Dean, Graduate School
August, 2015
THE GRADUATE SCHOOL
i
Abstract
Lack of sufficient early warning from drought monitoring and prediction tools
that rely heavily on precipitation and soil moisture has prompted the need for
development of new drought metrics that can account for evaporative dynamics and
interactions between the land surface-atmosphere interface. Previous studies have shown
that using anomalies in actual evapotranspiration (ET) estimated from satellite imagery
can provide some drought early warning during the growing season, but there are a
number of limitations to satellite monitoring that encourage more research on easily
accessible and real-time (or forecasted) evaporative demand (E0) drought tools. The focus
of this study is the development of a novel drought metric that relies only physically
based E0 driven by temperature, wind speed, solar radiation, and humidity, which can all
be obtained from gridded weather data and dynamical forecast model output. An
evaluation of several gridded data products was first carried out using the Nevada
Climate-ecohydrological Assessment Network (NevCAN) in a remote part of the Great
Basin to investigate biases and deficiencies that are inherent to regions with sparse
observations. It was determined that the University of Idaho Gridded Meteorological
Data (METDATA) was most suitable to drive a new drought index, the Evaporative
Demand Drought Index (EDDI). EDDI was computed over CONUS for the period of
1979-2013 and compared against other commonly used drought indices. During rapid
onset drought, or flash drought (i.e., 2011-2012 in the Midwest) EDDI was found to lead
other indicators by as much as 1-3 months. Given that E0 contains no precipitation input,
the potential exists to improve seasonal drought predictions, which currently suffer from
a lack of skillful precipitation forecasts. Skill of seasonal E0 anomaly forecasts were
ii
assessed over CONUS using the Climate Forecast Version 2 (CFSv2) hindcasts for 1982-
2009 and METDATA was used as baseline observations. E0 forecast skill during drought
events was consistently greater than precipitation, with much improved skill over parts of
the central and northeast U.S. during the growing season. Results from this study suggest
that continued efforts should be put towards incorporating physically based E0 in
operational drought monitoring and prediction frameworks.
iii
Acknowledgements
I would like to thank my family for continued support throughout this long
journey, especially my mother, Ilene McEvoy, and my wife, Heather McEvoy. Many
thanks to my committee members; I could not have made it this far without their expert
knowledge and guidance. I would especially like to thank Dr. John Mejia and Dr. Justin
Huntington for having confidence in me to largely choose my own path and take charge
of my dissertation topics. There is a long list of DRI graduate students who I would like
to thank for help and support along the way, particularly with the steep learning curve
associated with scientific programming. I’m extremely thankful that I got the opportunity
to experience this challenging task, and the experience will stick with me for the rest of
my life. I would like to dedicate this work to my two year daughter, Sierra McEvoy, and
my two week old son, Sage McEvoy.
iv
Table of Contents
Abstract…………………………………………………………………………………….i
Acknowledgements………………………………………………………………………iii
Table of Contents…………………………………………………………………………iv
List of Tables…………………………………………………………………………….vii
List of Figures…………………………………………………………………………...viii
Introduction………………………………………………………………………………..1
Use of an Observation Network in the Great Basin to Evaluate Gridded Climate Data….5
Abstract……………………………………………………………………………………6
Introduction………………………………………………………………………………..7
Data and Methodology…………………………………………………………………...10
NevCAN data…………………………………………………………………………….10
Additional observations………………………………………………………………….18
Gridded data……………………………………………………………………………...19
GDP/observation comparison and statistical methods…………………………………...23
Results……………………………………………………………………………………25
Snake Range SN5 and Wheeler Peak SNOTEL intercomparison……………………….25
Snake Range comparisons of observations and GDPs…………………………………..28
Sheep Range comparisons of observations and GDPs…………………………………..36
Summary and Conclusions………………………………………………………………43
Acknowledgements………………………………………………………………………46
References………………………………………………………………………………..47
The Evaporative Demand Drought Index: CONUS-wide Assessment Against Common
Drought Indicators……………………………………………………………………….55
v
Abstract…………………………………………………………………………………..56
Introduction………………………………………………………………………………57
Data and methods………………………………………………………………………...60
Evaporative demand……………………………………………………………………...60
Evaporative Demand Drought Index…………………………………………………….61
NLDAS-based drought metrics…………………………………………………………..62
Evaporative Stress Index…………………………………………………………………63
United States Drought Monitor…………………………………………………………..64
Results and Discussion…………………………………………………………………..65
NLDAS-2 drought index correlations with EDDI……………………………………….65
ESI correlations with EDDI……………………………………………………………...72
Flash drought over the central US……………………………………………………….74
Extended drought in arid to semi-arid regions…………………………………………...80
Summary and Conclusions………………………………………………………………84
Acknowledgements………………………………………………………………………87
References………………………………………………………………………………..87
Exploring the use of Physically Based Evaporative Demand Anomalies to Improve
Seasonal Drought Forecasts……………………………………………………………...96
Abstract…………………………………………………………………………………..97
Introduction………………………………………………………………………………98
Data and methodology………………………………………………………………….101
Results…………………………………………………………………………………105
Deterministic skill………………………………………………………………………105
Categorical skill of probability forecasts in drought events……………………………115
vi
ENSO as a source of predictability……………………………………………………..118
Discussion and conclusions…………………………………………………………….119
Acknowledgements……………………………………………………………………..123
References………………………………………………………………………………123
Summary and Conclusions……………………………………………………………..127
vii
List of Tables
Use of an Observation Network in the Great Basin to Evaluate Gridded Climate Data
Table 1. Summary of station locations, elevations, measured variables (where T indicates
temperature), measurement frequency, and precipitation gauge information (where TB
indicates tipping bucket)………………………………………………………………….12
Table 2. Geonor rain gauge and tipping bucket rain gauge comparison
statistics…………………………………………………………………………………..16
The Evaporative Demand Drought Index: CONUS-wide Assessment Against Common
Drought Indicators
Table1. Drought classes for comparing USDM to SPI, SSI, ESI, and EDDI……………64
Exploring the Use of Physically Based Evaporative Demand Anomalies to Improve
Seasonal Drought Forecasts
Table 1. CFSv2 monthly ensembles are listed and each initial day consists of four
members initialized at 00Z, 06Z, 12Z, and 18Z………………………………………102
viii
List of Figures
Use of an Observation Network in the Great Basin to Evaluate Gridded Climate Data
Figure 1. (A) Study area with insets indicating locations of the Snake and Sheep Ranges.
Close-up of the (B) Snake Range and (C) Sheep Range with red dots indicating station
locations. Zoomed-in frame (B) highlights the close proximity of Wheeler Peak SNOTEL
(WPS) to NevCAN. Station summaries can be found in Table 1………………………..13
Figure 2. Time series of Geonor rain gauge (black line) and tipping bucket rain gauge
(green line) precipitation at SN2 during the (a) cold season and (b) warm season.
Abscissa tick marks indicate date (MM/DD)…………………………………………….17
Figure 3. Water year 2012 total precipitation [mm] at the Snake Range for (a) PRISM 4-
km, (b) PRISM 800-m, (c) JA 4-km, and (d) Daymet 1-km…………………………….22
Figure 4. Elevation of each station and nearest GDP pixel (a) Snake and (b) Sheep
Ranges. Difference between grid point and station elevation (GDP-station) are shown in
(c) and (d), respectively. X-axis follows the dominant alignment, in the west-to-east
direction for the Snake Range transect and nearly south-to-north for the Sheep Range...24
Figure 5. NevCAN SN5 (black) and WPS (magenta) daily precipitation totals (a) and
accumulated precipitation (b) throughout the water year. Abscissa tick marks indicate
date (month-year)………………………………………………………………………...26
Figure 6. Photographs of the SN5 Geonor gauge and the surrounding vegetation (a, b).
Photograph (a) was taken looking to the east, and (b) was taken looking to the west.
Photographs of the WPS weighing gauge and surrounding vegetation (c, d). Rain gauges
are highlighted by yellow rectangles. The orientation of (c) and (d) is unknown.
Photographs courtesy of WRCC and NRCS……………………………………………..28
Figure 7. Snake Range seasonal precipitation totals (a, b) and seasonal mean Tmax (c, d),
Tmin (e, f), and Tdew (g, h). Cold season is shown on the left (a, c, e, g) and warm season
on the right (b, d, f, h). X-axis is aligned west to east (left to right)……………………..32
Figure 8. Snake Range seasonal bias (GDP - obs) for cold season (left) and warm season
(right) precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h). Variables needed to
calculate Tdew are not measured at WPS, therefore no Tdew values are shown…………..33
Figure 9. Snake Range cold season R2 (left) and MAE (right) computed at the daily time
step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and JA.
For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and Tdew
is expressed in °C………………………………………………………………………...35
Figure 10. Snake Range warm season R2 (left) and MAE (right) computed at the daily
time step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and
JA. For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and
Tdew is expressed in °C…………………………………………………………………...36
Figure 11. Sheep Range seasonal precipitation totals (a, b) and seasonal mean Tmax (c, d),
Tmin (e, f), and Tdew (g, h). Cold season is shown on the left (a, c, e, g) and warm season
on the right (b, d, f, h). X-axis is aligned west to east (left to right). For precipitation
observations, filled circles represent tipping bucket gauges, and filled upside down
triangles represent weighing gauges……………………………………………………..39
ix
Figure 12. Sheep Range seasonal bias (GDP - obs) for cold season (left) and warm season
(right) precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h)……………………….40
Figure 13. Sheep Range cold season R2 (left) and MAE (right) computed at the daily
time step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and
JA. For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin,
and Tdew is expressed in °C……………………………………………………………..42
Figure 14. Sheep Range warm season R2 (left) and MAE (right) computed at the daily
time step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and
JA. For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and
Tdew is expressed in °C…………………………………………………………………...43
The Evaporative Demand Drought Index: CONUS-wide Assessment Against Common
Drought Indicators
Figure 1. Correlation coefficient between EDDI and SPI at (a) 1-month, (c) 6-month, (e)
12-month, and SSI (b) 1-month, (d) 6-month, and (f) 12-month time scales……………66
Figure 2. Shading indicates METDATA terrain height (m) and red boxes indicate area-
averaging domains for Figures 3 and 4. IA, TX, and PA boxes are 50 x 100 4-km
METDATA pixels (200 km x 400 km), and CA box is 25 x 25 pixels (100 km x 100
km)……………………………………………………………………………………….67
Figure 3. Monthly correlations between EDDI and SPI (top row) and SSI (bottom row) at
all time scales for (a, e) TX, (b, f) CA, (c, g) IA, and (d, h) PA. Y-axis indicates ending
month of each time scale, and x-axis shows time scale (months). Shading indicates
correlation coefficients…………………………………………………………………...70
Figure 4. Lagged correlation between 3-month SSI ending in August and EDDI for (a)
CA, (b) TX, (c) IA, and (d) PA. Y-axis indicates EDDI ending months and x-axis
indicate EDDI time scale. Green dots are placed in the ending month containing the
strongest correlation for each time scale, and blue dots are used as a reference to show
SSI time scale and ending month………………………………………………………...72
Figure 5. Seasonal correlation coefficient (left column spring and right column summer)
between ESI and EDDI at (a, b) 4-week, (c, d) 8-week, and (e, f) 12-week time scales.
Areas shaded in white indicate an insufficient amount of ESI data……………………..74
Figure 6. EDDI under sustained and flash drought conditions. (a) Monthly time series of
1-month EDDI, SSI, and SPI area averaged over the IA domain. (b) Monthly time series
of 1-month EDDI and EDDI constrained by climatology Tair (EDDI-T), q (EDDI-q), Rd
(EDDI-Rd), and U2 (EDDI- U2). Black box highlights time period shown in (c). (c) Daily
time series of 1-month EDDI, EDDI-T, EDDI-q, EDDI-Rd and EDDI-U2 for May and
June 2011 shown to highlight details of flash drought initiation. Note that the vertical axis
of EDDI is reversed to clearly visualize drought onset and duration when compared to
SPI and SSI. Light green reference line indicates start of moderate drought classification
(-0.78)……………………….............................................................................................77
Figure 7. Evolution of the 1-month EDDI (top row), USDM (second row), 1-month ESI
(third row), 1-month SSI (fourth row), and 1-month SPI (fifth row) through spring and
summer of 2012. USDM data are from 1 May, 2012 (April column), 5 June, 2012 (May
x
column), 3 July, 2012 (June column), and 31 July, 2012 (July column). EDDI, ESI, SSI,
and SPI are at 1-month time scales at the end of each month: April, May, June, and July.
All drought metrics have been converted to USDM categories according to Table 1…..79
Figure 8. USDM from 02 October, 2007 (a) and 25 June, 2002 (b), 12-month (October-
September) EDDI (c), SSI (e), and SPI (g) ending September, 2007, and 6-month
(January-June) EDDI (d), SSI (f), and SPI (h) ending June, 2002………………………82
Figure 9. Area-averaged time series of EDDI over the northern Sierra Nevada from 1979
to 2013 aggregated at 2-week (a), 1-month (b), 3-month (c), 6-month (d), and 12-month
time scales………………………………………………………………………………..84
Exploring the Use of Physically Based Evaporative Demand Anomalies to Improve
Seasonal Drought Forecasts
Figure 1. Accumulated E0 (a) and Prcp (b) anomaly percentiles from METDATA for
AMJ 2002. Note that upper E0 and lower Prcp percentiles indicate drought (brown
shading). NCDC climate regions (described in Section 2) used as area averaging domains
for Section 3 results are shown in the bottom panel (c). Regions are named as follows:
Northwest (NW), West (We), Southwest (SW), West North Central (WNC), South (So),
East North Central (ENC), Central (Ce), Southeast (SE), and Northeast (NE)………...100
Figure 2. CONUS average percent area in drought based on 3-month accumulated E0 (a)
and Prcp (b) percentiles………………………………………………………………...101
Figure 3. Comparison between 1982-2009 CONUS-average annual E0 from the
METDATA native grid of 4-km (x-axis) and the re-gridded 1° spatial resolution (y-
axis)……………………………………………………………………………………..103
Figure 4. Average ET0 anomaly correlation between METDATA and CFSRF over each
region (refer to Figure 1c for full region names and locations). Labels on the x-axis
indicate lead time (months) and labels on the y-axis indicate target month……………106
Figure 5: Average precipitation anomaly correlation between METDATA and CFSRF
over each region (refer to Figure 1c in main manuscript for full region names and
locations). Labels on the x-axis indicate lead time (months) and labels on the y-axis
indicate target month……………………………………………………………………107
Figure 6. As in Figure 4, but for maximum temperature……………………………….108
Figure 7. As in Figure 4, but for minimum temperature………………………………..109
Figure 8. As in Figure 4, but for specific humidity…………………………………….110
Figure 9. As in Figure 4, but for downwelling shortwave radiation at the surface…….111
Figure 10. As in Figure 4, but for wind speed………………………………………….113
Figure 11. Season-1 anomaly correlation area-averaged over CONUS and individual
climate regions. The black reference line is anomaly correlation of 0.3, which indicates
the start of moderate skill……………………………………………………………….115
Figure 12. The HSS for lead one-month, season-1 forecasts for cases when both E0 and
Prcp indicate drought (>80th
percentile for E0 and <20th
percentile for Prcp). Labels inside
of each panel indicate region and mean HSS. Red circles show notable drought events of
JFM, FMA, and MAM 1992 in the NW, JJA 1988 in the WNC, ENC, and Ce, JAS, ASO,
and SON 1999 in the Ce, and MJJ and JJA 1999 in the NE. These events are described in
further detail in the text…………………………………………………………………117
xi
Figure 13. The difference in AC (ENSO conditional forecasts – All forecasts) and
regionally averaged AC for E0 (a and c) and Prcp (b and d) forecasts…………………119
1
Introduction Drought is a costly and devastating type of natural disaster that leads to adverse
effects in many groups and sectors throughout the United States (US) including
agriculture, water resources, public health, outdoor recreation, and the national and global
economy. The 20th
century saw a number of severe droughts in the US, most notably the
Dust Bowl era of the 1930’s, but some of the most extreme events have occurred in the
first 15 years of the 21st century: 2000-2003 in the Southwest, 2011 in Texas, 2011-2012
in the Midwest, and 2012-present in California. Recent drought severity has been
exacerbated by high temperatures and increased evaporative demand (E0), and future
projections of a warming climate indicate that extreme drought events (like those of the
21st century) are likely to become more frequent and longer in duration. Improving
drought monitoring and prediction capabilities is therefore of utmost importance for the
US, and could reduce the damaging environmental and societal effects of drought by
providing much needed early warning. Improved early warning allows for reactive
emergency responses, implementation of actions within statewide and local drought
plans, and can provide essential information for decision support.
Precipitation, temperature, and soil moisture have been the most commonly used
variables in drought monitoring and prediction in the past. However, recent advances
have been made that primarily utilize the wealth of data provided by satellites and land
surface models (LSMs) to explore the utility of evapotranspiration (ET) and E0 as
drought indicators. Promising new research has shown that through interactions between
the land surface-atmosphere interface these variables (ET and E0) can often times provide
early warning over traditional drought metrics. Satellite data has been an invaluable
2
resource to geophysical scientists, but there a number of limitations to satellite
monitoring that encourage more research and development of easily accessible and real-
time (or forecasted) ET and E0 drought tools. Even with advanced satellites accurate
spatial estimates of ET continue to be a major challenge, making E0 estimates driven by
more commonly measured meteorological variables a more attractive option for drought
monitoring.
Over the past several decades a number of options have become available to
obtain data for estimation of physically based E0 (driven by temperature, humidity, wind
speed, and solar radiation) including LSMs, reanalysis of atmospheric models, and purely
statistical models where observations are spatially interpolated. A major challenge in
today’s data-rich world is determining which data source is best suited for a specific
application, such as E0 estimation, and unfortunately the consequences of these choices
are often overlooked.
The research in this dissertation is focused on improving drought monitoring and
predictions using physically based E0, and evaluating gridded climate data that may be
used for the application of estimating E0. The hypothesis and research questions for each
chapter are as follows:
1. It is hypothesized in Chapter 1 that in the complex terrain of the western US,
specifically the Great Basin, significant differences can be found in the variables
of precipitation, temperature, and humidity between several gridded data products
(GDPs) of varying spatial resolution, with higher resolution not always leading to
greater skill. Of particular interest are the impacts of low station density and
3
station siting on GDP precipitation estimates, and methodology and accuracy of
GDP humidity estimates, which are a critical component of physically based E0
estimates. In this research four GDPs are evaluated (ranging from 4-km to 800-m
spatial resolution) against a new observing network: the Nevada Climate-
ecohydrological Assessment Network (NevCAN).
2. Development and assessment of the first drought index based solely on physically
based E0, the Evaporative Demand Drought Index (EDDI), is presented in
Chapter 2. Based partially on several findings from Chapter 1, the 4-km
METDATA was chosen to calculate E0 using the physically based American
Society for Civil Engineers Standardized Reference ET (ET0) approach. A
calculation procedure for EDDI is presented, and EDDI is calculated over
CONUS for several aggregation time scales for the period of 1979-2013. EDDI is
then compared against several commonly used drought indices and the US
Drought Monitor. It is further hypothesized that EDDI can be a leading indicator
during rapid onset, or “flash” drought—conditions when ample moisture may still
be available at the surface (i.e., energy limited ET conditions) both due to both
advective and radiative meteorological forcings, ET and ET0 are driven in the
same direction, thus leading to a drought signal from EDDI and a wetting signal
from ET-based drought metrics. The hypothesis that EDDI can also serve as an
effective indicator of hydrologic drought in water-limited regions (based on the
classic complementary relationship between ET and ET0) using longer time scales
is also tested.
4
3. Seasonal drought prediction remains a challenge primarily due to poor skill in
long-term precipitation forecasts. Skill in seasonal temperature predictions is
significantly better than precipitation. However, temperature alone is not always
an accurate drought indicator. In Chapter 3, the hypothesis that predictions of
seasonal ET0 anomalies contain improved skill over precipitation anomaly
forecasts due largely to the sensitivity of ET0 to temperature is tested. Reforecast
data from the National Center for Environmental Prediction Climate Forecast
System Version 2 (CFSv2) is used to compute ET0 and evaluate both ensemble
(deterministic) and probabilistic forecasts against precipitation for 1982-2009
focusing on drought events. METDATA is used as observations against which to
evaluate CFSv2. The drivers of ET0 are also evaluated providing novel
information on seasonal predictions of the often-overlooked variables of specific
humidity, solar radiation, and wind speed.
5
Use of an Observation Network in the Great Basin to Evaluate Gridded Climate
Data
Daniel J. McEvoy1, 2
, John F. Mejia1, and Justin L. Huntington
1
1Desert Research Institute, Reno, Nevada
2Atmospheric Science Graduate Program, University of Nevada, Reno, Nevada
6
Abstract
Predicting sharp hydro-climatic gradients in the complex terrain of the Great
Basin can be challenging due to the lack of climate observations that are gradient
focused. Furthermore, evaluating gridded data products (GDPs) of climate in such
environments for use in local hydro-climatic assessments is also challenging and
typically ignored due to the lack of observations. In this study, we use independent
Nevada Climate-ecohydrological Assessment Network (NevCAN) observations
temperature, relative humidity, and precipitation collected along large elevational
gradients of the Snake and Sheep Mountain ranges from water year 2012 (October, 2011,
to September, 2012) to evaluate four GDPs of different spatial resolutions: PRISM
(Parameter Regression on Independent Slopes Model) 4-km, PRISM 800-m, Daymet 1-
km, and a North American Land Data Assimilation System (NLDAS)/PRSIM hybrid 4-
km product. Inconsistencies and biases in precipitation measurements due to station
siting and gauge type proved to be problematic with respect to comparisons to GDPs.
This study highlights a weakness of GDPs in complex terrain: an underestimation of
inversion strength and resulting minimum temperature in foothill regions, where cold air
regularly drains into neighboring valleys. Results also clearly indicate that for semi-arid
regions, the assumption that daily average dew point temperature (Tdew) can be estimated
as the daily minimum temperature does not hold, and therefore should not be used to
interpolate Tdew spatially. Comparisons of GDPs to observations varied depending on
the climate variable and grid spatial resolution, highlighting the importance of conducting
local evaluations for hydro-climatic assessments.
7
Introduction
Weather over complex terrain is particularly sensitive to small changes in climatic
forcings (Loarie et al., 2009; Rangwala and Miller, 2012). Therefore, weather observation
networks in complex terrain are useful for studying the local effects of potential changes
in regional temperature and precipitation. Globally, mountainous regions serve as the
primary source of water for about 50 percent of the population (Bandyopadhyay et al.,
1997) and nearly all of the perennial surface and groundwater resources in the Great
Basin (Eakin, 1966; Flint et al., 2004), where the accumulation of wet season (October to
March) precipitation in the form of snow comprises roughly 90 percent of the annual
precipitation. Diffuse snowmelt during the spring provides nearly all of the annual runoff
and groundwater recharge, which makes the Great Basin particularly sensitive to climatic
changes under a warming climate (Barnett et al., 2005; Rauscher et al., 2008). With
development continuing to increase in metropolitan and rural areas of the Great Basin
and pending inter-basin groundwater transfers planned from eastern to southern Nevada
(Burns and Drici, 2011; Nevada Bureau of Land Management, 2012), detailed analyses
of hydro-climatic variability across elevational gradients in the Great Basin are needed.
Weather stations in the Great Basin are predominately located in valleys, which
presents a unique challenge for studying elevational climatic gradients and their effects
on the environment. Because climate observations are particularly sparse, gridded data
products (GDPs) are used extensively by researchers and practitioners to make estimates
of temperature, precipitation, and humidity distributions across space and time, despite
potential large uncertainties. In the Great Basin and surrounding regions, Parameter
Regression on Independent Slopes Model (PRISM; Daly et al., 1994) products are
8
commonly used for research and applied studies related to ecology (Ackerly et al., 2010),
biology (Bradley, 2009; Leger, 2013), hydrology (Welch et al., 2007; Burns and Drici,
2011; Huntington and McEvoy, 2011; Huntington and Niswonger, 2012; McEvoy et al.,
2012; Feld et al., 2013), climatology (Porinchu et al., 2010), and meteorology (Lundquist
et al., 2010). Of particular importance is the fact that over the last 10 years, PRISM
precipitation products have also been used in most expert witness studies and reports
associated with major water rights hearings in Nevada, where uncertainty of PRISM is
commonly a central focus for assessing uncertainty in the perennial yield of groundwater
(Jeton et al., 2006; Lundmark et al., 2007; Zhu and Yong, 2009; Epstein et al., 2010;
Burns and Drici, 2011; NSEO, 2012). Gridded data products are often used without
comparing estimates to independent or dependent observations (Bradley, 2009; Ackerly
et al., 2010; Porinchu et al., 2010; Leger, 2013). Therefore, studies that use independent
observations collected along mountain transects can be invaluable for validation of
GDPs, as well as revealing physical phenomena related to elevational gradients, such as
location of maximum precipitation, orographic processes, temperature and vapor lapse
rates, and spatial variability related to wind and topographic characteristics such as slope
and aspect.
The Nevada Climate-ecohydrology Assessment Network (NevCAN), located in
eastern and southern Nevada (Figure 1), is a new observation network designed to assess
climate variability and change and associated impacts on the surrounding ecology and
hydrology (Mensing et al., 2013). The network consists of one west-to-east transect in
eastern Nevada (Snake Range) and one south-to-north transect in southern Nevada
(Sheep Range) (Figure 1). With records beginning in June 2010, observations from
9
NevCAN have not been assimilated into the generation of GDPs, so a novel GDP
validation can be conducted with independent observations. In describing guidelines for
assessing modeled spatial climate data sets, Daly (2006) notes that using data
independent of the model will provide the least-biased evaluation. In this study, we use
acquired NevCAN data as the independent data set to evaluate four GDPs with different
spatial resolutions. Using different spatial resolutions of 4 km, 1 km, and 800 m provides
beneficial insight into disparities among the different GDPs, observations, and the ability
of GDPs to resolve local-scale precipitation, temperature, and humidity features.
An important, yet often overlooked, aspect of comparing any estimated weather
data to observations is the acknowledgement of measurement uncertainties. Measuring
solid precipitation remains particularly challenging and automated systems have been
found to under measure by as much as 20 percent to 50 percent mostly due to gauge
under-catch from strong winds (Rasmussen et al., 2011). Weather station siting and
gauge type can also impact measured precipitation totals, especially during snowfall
events (Goodison et al., 1998; Yang et al., 1998; Fassnacht, 2004). Therefore, biases in
observed precipitation should be established and taken into consideration before
analyzing differences between GDPs and measurements.
As highlighted above, an abundance of dynamically and statistically derived
precipitation and temperature GDPs is available to help overcome observational
limitations (Daly et al., 1994; Thornton et al., 1997; Abatzoglou, 2011). Our first
objective is to understand the degree to which these products can satisfactorily resolve
elevational climatic gradients in complex terrain and at what resolution. The second
10
objective of this study is to assess the uncertainties associated with precipitation
measurements and the impacts on the comparisons to estimated GDPs.
In the following sections, we describe the NevCAN transects, additional
observations, and GDPs, as well as the analyses and statistics used for the comparisons
and quality assured/quality controlled (QA/QC) protocols used to assess observational
uncertainty and error (Section 2). The results of the measurement uncertainty analysis
and comparisons between GDPs and observations are presented (Section 3) and discussed
with respect to elevational gradients and systematic biases found in estimated and
measured temperature, precipitation, and humidity. Lastly, we summarize and discuss our
results and provide concluding remarks on the differences between GDPs and
observations, and how the differences vary with grid size, parameter, and elevation
(Section 4).
Data and Methodology
NevCAN data
NevCAN meteorological data were obtained from the Western Regional Climate
Center (WRCC; http://www.wrcc.dri.edu/SRtransect/,
http://www.wrcc.dri.edu/GBtransect/) for the 2012 water year (October 1, 2011 through
September 30, 2012) and site descriptions are shown in Table 1. Alternatively, NevCAN
data can be obtained from the Nevada Climate Change Portal:
(http://sensor.nevada.edu/NCCP/Climate%20Monitoring/Network.aspx). The orientation
of the Snake Range transect is east/west; north/south for the Sheep Range (Figure 1). For
each of the two transects, daily maximum and minimum temperature (Tmax and Tmin), and
10-minute averaged relative humidity (RH) and temperature were obtained. Measurement
11
of near-surface vapor pressure and dew point temperature (Tdew) are often neglected in
mountain observing networks (e.g., SNOTEL [SNOpack TELemetry]), but crucial for
estimating evapotranspiration, atmospheric water demand, and land surface and boundary
layer feedbacks, which are often required for hydrologic and ecological modeling (Crago
et al., 2010; Huntington et al., 2011; Feld et al., 2013). Here, we compute vapor pressure
(ea) and Tdew, from 10-minute RH and temperature data, which is then averaged to daily
and monthly time steps to compare against GDPs. Tdew was calculated from ea following
the Murray (1967) equation. Actual vapor pressure was derived from saturation vapor
pressure (es; a function of air temperature) and RH as follows: [ea = es * RH/100].
12
Table1. Summary of station locations, elevations, measured variables, measurement frequency, and precipitation gauge
information.
Snake Range Network Latitude Longitude Elevation (m) Variables Analyzed Sampling Frequency P gauge type Orificie size (mm) Alter Shield
Sagebrush west (SN 1) NevCAN 38.9256 -114.4078 1768 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Pinyon Juniper west (SN 2) NevCAN 38.8922 -114.35 2202 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Montane west (SN 3) NevCAN 38.89 -114.3314 2819 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Subalpine west (SN 4) NevCAN 38.9061 -114.3089 3554 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Subalpine east (SN 5) NevCAN 39.01 -114.3094 3081 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Sagebrush east (SN 6) NevCAN 39.0206 -114.1764 1840 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Salt Desert Shrub east (SN 7) NevCAN 39.0369 -114.0572 1580 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (150)
weighing (160)
tipping (no)
weighing (yes)
Wheeler Peak (WPS) SNOTEL 39.00995 -114.31 3085 PPT, temperature hourly weighing 305 yes
Sheep Range
Desert Shrub (SH 1) NevCAN 36.4353 -115.3558 893 PPT, temperature, RH 10-minute tipping bucket 200 yes
Blackbrush (SH 2) NevCAN 36.5197 -115.1633 1680 PPT, temperature, RH 10-minute tipping bucket 200 yes
Pinyon Juniper (SH 3) NevCAN 36.5728 -115.2042 2065 PPT, temperature, RH 10-minute tipping bucket 200 yes
Montane (SH 4) NevCAN 36.5903 -115.2142 2272 PPT, temperature, RH 10-minute
tipping bucket
and weighing
tipping bucket (200)
weighing (160)
tipping (no)
weighing (yes)
Yucca Gap (YG) RAWS 36.4367 -115.3314 969 PPT, temperature, RH hourly tipping bucket 200 no
Hayford Peak (HP) SCAN 36.6581 -115.201 3013 PPT, temperature, RH hourly tipping bucket 200 no
13
Figure 1: (A) Study area with insets indicating locations of the Snake and Sheep Ranges.
Close up of the (B) Snake Range and (C) Sheep Range with red dots indicating station
locations. Zoomed-in frame (B) highlights the close proximity of Wheeler Peak SNOTEL
(WPS) to NevCAN. Station summaries can be found in Table 1.
All observations were QA/QCed by manual inspection to check for erroneous outliers,
and then aggregated to monthly time steps for monthly comparisons of GDP data.
Precipitation can be highly variable over short temporal scales, therefore raw 10-
minute data were summed to the day instead of using the WRCC pre-computed daily
precipitation, as an additional QA/QC measure. At the Snake Range, each station is
equipped with two precipitation gauge systems: (1) a weighing gauge with a 160-mm
14
diameter orifice (Geonor T-200B), and (2) a tipping bucket with a 150-mm diameter
orifice (TE 525). At the Sheep Range, all stations are equipped with tipping buckets
(TB4, 200-mm diameter orifice) except for SH4, which is the only station to have both
types of gauges. At locations with both types of gauges, only the Geonor gauges are
equipped with Alter shields (Alter, 1937) to reduce gauge under-catch, while the tipping
buckets were left unshielded. Alter shields were installed at tipping-bucket-only sites in
the Sheep Range.
Tipping buckets are known to underestimate precipitation, especially during
heavy rainfall or light drizzle (e.g., Humphrey et al., 1997) and have been shown to
collect much less frozen precipitation than standard weighing gauges (e.g., Rasmussen et
al., 2011). Daily tipping bucket precipitation measurements were compared to coincident
Geonor measurements of precipitation and the coefficient of determination (R2) and
season total differences were computed at each station for cold and warm seasons (Table
2). During the cold season, tipping buckets consistently underestimated precipitation
totals with differences exceeding 100 mm at SN3 and SN5 , and R2 was found to
decrease (R2
range of 0.01-0.87) with an exceptionally weak relationship found between
the two gauge types at high elevation (R2 of 0.01 at SN5). As expected, correlations of
daily precipitation were much higher during the warm season (R2 from 0.78 to 0.98), but
decreased with elevation due to more days with frozen precipitation. The lower
correlations of daily precipitation during the cold season are primarily caused by a delay
in timing of tip counts due to frozen precipitation events. For example, snow or ice in the
tipping buckets may take several hours to several days to melt and the event is then offset
from the Geonor data by one to several days (Figure 2). Because of the well-known
15
limitations of tipping buckets (e.g. Humphrey et al., 1997, Rasmussen et al., 2011), and
as highlighted in this analysis, weighing gauge precipitation measurements were used for
evaluating the skill of GDP precipitation estimates when available.
16
Table2. Geonor rain gauge and tipping bucket rain gauge comparison statistics.
Tipping bucket (TB) and Geonor comparison*
SN1 SN2 SN3 SN4 SN5*** SN6** SN7*** SH4****
Cold season
(Oct-Mar) R
2 0.67 0.25 0.03 no TB data 0.01 0.24 0.87 0.12
difference (mm) 13.91 49.3 118.13 no TB data 186.32 8.36 12.92 12.63
% of total 16% 32% 51% no TB data 64% 11% 17% 12%
Warm season
(Apr-Sep)
R2 0.98 0.97 0.78 no TB data no TB data no TB data no TB data 0.91
difference (mm) 23.26 19.73 9.98 no TB data no TB data no TB data no TB data 2.78
% of total 21% 12% 5% no TB data no TB data no TB data no TB data 1%
*difference given as Geonor – TB and percent of total shown with respect to Geonor seasonal total.
** complete data for October 1 - February 17 of cold season only
*** complete data for October 1 - February 28 of cold season only
**** complete data for all of cold season and April 1 - August 15 of warm season
17
Figure 2: Time series of Geonor rain gauge (black line) and tipping bucket rain gauge
(green line) precipitation at SN2 during the (a) cold season and (b) warm season.
Abscissa tick marks indicate date (MM/DD).
18
Additional observations
We use a Natural Resources Conservation Service (NRCS) SNOTEL station
(Tmax, Tmin, and precipitation) to compare to a nearby (~50 m) NevCAN Snake Range
station located in the high-elevation subalpine region and to all GDPs. The Wheeler Peak
SNOTEL (WPS) precipitation gauge is a weighing-type gauge; however the orifice
diameter is approximately twice the size (~305 mm) of the NevCAN Geonors (~160
mm). Daily Tmax, Tmin, and precipitation SNOTEL data were obtained from the NRCS
(http://www.wcc.nrcs.usda.gov/nwcc/site?sitenum=1147&state=nv) and aggregated to
monthly time steps and QA/QCed.
The WPS station is the only high-elevation station in the Snake Range being used
as a control point for PRISM and Daymet spatial distribution algorithms (M. Halbleib,
Oregon State University, electronic communication, http://daymet.ornl.gov/
overview). Therefore, the effect of dependent versus independent observations compared
to GDPs is examined. In this portion of the study, we highlight the importance of
thoroughly understanding GDP control point when using GDP estimates for local climate
assessments. Important assumptions related to weather station and precipitation gauge
footprint, siting and exposure, and sensor limitations/deficiencies are also explored.
Two additional stations were used in the Sheep Range in order to develop a more
complete south-north transect with one station from the NRCS Soil Climate Analysis
Network (SCAN) (Figure 1, Table 1). Hourly RH, temperature, and precipitation data
were downloaded (http://www.wcc.nrcs.usda.gov/scan/) and QA/QCed. The Hayford
Peak (HP) SCAN station is equipped with an unheated tipping bucket and has an eight-
19
inch (~200 mm) diameter orifice. Therefore, the winter precipitation data contain large
uncertainties due tipping bucket deficiencies described in Section 2.1. The Yucca Gap
(YG) Remote Automated Weather Station (RAWS) was the second additional station
used at the Sheep Range and hourly RH, temperature, and precipitation data were
downloaded from WRCC (http://www.raws.dri.edu/cgi-bin/rawMAIN.pl?nvNYUC) and
QA/QCed. Yucca Gap is also instrumented with an unheated tipping bucket precipitation
gauge, therefore winter precipitation values are highly uncertain and discussed later. Dew
point and vapor pressure were computed following the same methods used for NevCAN
data. All observations used in the study are summarized in Table 1.
Gridded data
In this study, NevCAN data sets are considered to be “baseline measurements” to
evaluate the skill of four GDPs: (1) PRISM 4-km (Daly et al., 1994), (2) PRISM 800-m
(Daly et al., 1994), (3) Daily Surface Weather and Climatological Summaries (hereafter
called Daymet) 1-km (Thornton et al., 1997), and (4) a North American Land Data
Assimilation (NLDAS)/PRISM hybrid 4-km (hereafter called JA; Abatzoglou, 2011). We
address the uncertainties in NevCAN precipitation measurements and highlight how these
biased observations impact the comparisons to GDPs in the results and discussion
section.
Total monthly PRISM precipitation and average monthly Tmax, Tmin, and Tdew
were obtained (acquisition date: January 2013) for both 800-m and 4-km spatial
resolutions from the PRISM website (www.prism.oregonstate.edu). All variables were
interpolated by PRISM using Climatologically-Aided Interpolation (CAI; Willmott and
Robeson, 1995). For Tmax, Tmin, and precipitation, PRISM was used to interpolate 1971-
20
2000 monthly normals, using elevation as the predictor grid, with stations weighted by
vertical and horizontal distance, plus several physiographic factors, such as topographic
orientation, coastal proximity, inversion height, and topographic position (Daly et al.,
2008). Once the normals were interpolated, CAI was used to interpolate data for a given
month and year. Average monthly PRISM Tdew estimates were computed by first taking
monthly dew point depression (Ko) observations and spatially interpolating monthly Ko
using PRISM Tmin as the predictor in the regression function. Dew point was then back
calculated using PRISM Ko and Tmin (Chris Daly, Oregon State University, electronic
communication). To create the monthly dew point time series, CAI was again used;
PRISM assimilated station data in the form of monthly mean dew point, and used the
1971-2000 normal dew point for that month as the predictor grid in its local regression
function. The Murray (1967) equation was rearranged and used to compute vapor
pressure, where [ea = exp[(0.0707 * Tdew - 0.49299)/(0.00421 * Tdew + 1)].
The third GDP evaluated was developed by Abatzoglou (2011) and combines the
spatial attributes of monthly PRISM data with daily temporal resolution of the North
American Land Data Assimilation (NLDAS-2; Mitchell et al., 2004). All of the NLDAS-
2 non-precipitation surface variables are derived from the North American Regional
Reanalysis (NARR; Mesinger et al., 2006), and the native NARR data is spatially
downscaled from 32-km to 12-km and temporally disaggregated from 3-hourly to hourly
(Cosgrove et al., 2003). For NLDAS-2 precipitation, Climate Prediction Center (CPC)
gridded daily gauge data (with a PRISM topographical adjustment) are the primary data
source. Daily CPC data are temporally disaggregated to hourly using radar and satellite-
based estimates (if available), and NARR. The first step in developing JA data is a
21
bilinear interpolation of NLDAS-2 onto the PRISM grid (4-km). CAI is then used to bias
correct the daily temperature, humidity, and precipitation data to a given PRISM month
(Abatzoglou, 2011). Daily Tmax, Tmin, RHmax, RHmin, and total precipitation were
obtained from: http://cloud.insideidaho.org/data/epscor/gridmet/. Dew point from JA was
calculated at the daily time step as a function of actual vapor pressure following the
Murray (1967) equation. Actual vapor pressure was derived from RHmax, RHmin,
saturation vapor pressure at Tmax (estmax), and saturation vapor pressure at Tmin (estmin),
where [ea = (estmax * (RHmin/100) + estmin * (RHmax/100)) / 2], as recommended by Allen
et al. (1998) for daily data.
Daymet was the fourth GDP evaluated, and is available for all of North America
at daily time steps and at 1-km spatial resolution. Daily Tmax, Tmin, precipitation, and
vapor pressure data were acquired online (http://daymet.ornl.gov; Thornton et al., 2012).
To interpolate Tmax, Tmin, and precipitation, Daymet uses a truncated Gaussian filter, and
a weighted least-squares regression is applied to establish the relationship between a
given variable and elevation (Thornton et al., 1997). While both Daymet and PRISM use
local linear regression, Daymet assumes a strictly monotonic relationship between
temperature and elevation, which limits the ability of Daymet to handle temperature
inversions (Daly et al., 2006). Daymet daily average vapor pressure is derived following
the assumption that daily Tmin = daily average Tdew (Thornton et al., 1999, 2000).
However, as we show in the results and discussion, Daymet monthly average Tdew rarely
equals Tmin, especially in semiarid and arid environments. Daily average Tdew was
calculated directly from Daymet daily average vapor pressure following Murray (1967).
22
Figure 3 provides a spatial perspective on grid resolution of different GDPs in
relation to station density and illustrates the 2012 water year total precipitation at the
Snake Range. Although all GDPs indicate maximum precipitation occurring near the
crest of the Snake Range (SN4 and SN5), PRISM 4-km and JA both have a maximum
value of over 100 mm less than PRISM 800-m and Daymet. Coarser grid size leads to
larger areas per pixel being averaged. Therefore, mountain peaks are represented as
lower-elevation areas when compared to 800-m and 1-km DEM values, leading to
precipitation totals to be reduced. A more detailed discussion on the effects of grid
resolution on biases is presented in Section 3.
Figure 3: Water year 2012 total precipitation [mm] at the Snake Range for (a) PRISM 4-
km, (b) PRISM 800-m, (c) JA 4-km, and (d) Daymet 1-km.
23
GDP/observation comparison and statistical methods
For each observing station, the nearest GDP center point was found to conduct
direct comparisons for each meteorological variable, as well as elevation (Figure 4).
Given that GDPs largely rely on elevation to distribute climatic variables, differences
between GDP pixel and station elevations were expected to largely explain GDP biases.
For example, Figure 4 shows the PRISM 4-km pixel at SN2 to be more than 200 m
higher than the station elevation. Based on this alone, PRISM 4-km temperature was
expected to be cooler and precipitation was expected to be greater than SN2 observed
values due to environmental lapse rates and typical mid-latitude precipitation-elevation
relationships, where precipitation increases with elevation (Houghton, 1979; Smith, 1979;
Daly et al., 1994). Biases between station observations and GDPs were computed using
seasonal means for Tmax, Tmin, and Tdew and sums for precipitation. Additionally, R2 and
mean absolute error (MAE) was computed using daily means and sums (for JA and
Daymet GDPs). All biases were computed as GDP – observation.
24
Figure 4: Elevation of each station and nearest GDP pixel (a) Snake and (b) Sheep
Ranges. Difference between grid point and station elevation (GDP-station) are shown in
(c) and (d), respectively. X-axis follows the dominant alignment, in the west-to-east
direction for the Snake Range transect and nearly south-to-north for the Sheep Range.
25
Results
Snake Range SN5 and Wheeler Peak SNOTEL intercomparison
An important aspect of any comparison study between estimated and measured
data is an evaluation, or at least an acknowledgement, of the quality of the measured data.
For this study, we compare measured precipitation at the NevCAN SN5 station to
measured precipitation at the Wheeler Peak SNOTEL station (WPS). The distance
between the two stations is ~50 m with an elevation difference of only 3.7 m. Daily and
water year accumulation of precipitation for SN5 and WPS are shown in Figure 5, which
clearly shows that both stations tend to record the same precipitation events; however
WPS consistently has higher daily totals. It seems unrealistic that WPS would receive
~30% more precipitation in one water year than SN5, considering their close proximity
(~50 m apart) and nearly identical elevation.
26
Figure 5: NevCAN SN5 (black) and WPS (magenta) daily precipitation totals (a) and
accumulated precipitation (b) throughout the water year. Abscissa tick marks indicate
date (month-year).
There are a number of factors that could contribute to these contrasting values that
fall within two general categories: (1) instrumentation differences and (2) station siting.
Both gauges are weighing types and have Alter shields and the orifice heights for WPS
and SN5 are 4.9 m and 3 m, respectively, with the WPS orifice diameter being twice that
of SN5 (diameters of 305 mm and 160 mm, respectively). The higher orifice height at
WPS should experience higher wind speed, and therefore less catch when compared to
SN5, which is in contrast to our findings. However, siting characteristics, such as the
27
height of surrounding vegetation and exposure to wind, could also be affecting
precipitation totals, particularly during snowfall events (Goodison et al., 1998; Yang et
al., 1998; Fassnacht, 2004). Photographs from SN5 reveal large, tightly spaced trees
surrounding the shielded Geonor gauge (Figures 6a and 6b), and the gauge height is low
with respect to surrounding tree height, while the tree spacing around the WPS gauge
appears to be much less dense (Figures 6c and 6d). The prevailing wind direction in the
winter months is from the west-southwest, and the clusters of large trees surrounding
SN5 (specifically the clusters to the west of the gauge; Figure 6b) are likely physically
blocking wind-blown snow from being captured in the gauge, whereas less dense forest
lies directly to the west of WPS (not shown, but can be seen from satellite imagery). It
should also be noted that the SNOTEL gauge reports precipitation to the nearest tenth of
an inch, while the Geonors report to the nearest hundredth of an inch, indicating greater
precision in the precipitation that is caught by the NevCAN gauges. Differences in
gauge calibration could also lead to different precipitation measurements for similar
events (Sieck et al., 2007), which may be yet another factor leading to the discrepancies
identified in this section.
Inconsistencies and biases in measurements due to station siting, calibration, and
design are problematic if used for impact assessments and reports. Through the
intercompariosn of SN5 and WPS, we have shown that great uncertainty remains with
respect to precipitation measurements in this region, and the resulting comparisons to
GDPs will also contain a large degree of uncertainty. Unfortunately, a comparison such
as we have presented here is not possible with other NevCAN stations because WPS is
the only SNOTEL in the Snake Range.
28
Figure 6: Photographs of the SN5 Geonor gauge and the surrounding vegetation (a, b).
Photograph (a) was taken looking to the east, and (b) was taken looking to the west.
Photographs of the WPS weighing gauge and surrounding vegetation (c, d). Rain gauges
are highlighted by yellow rectangles. The orientation of (c) and (d) is unknown.
Photographs courtesy of WRCC and NRCS.
Snake Range comparisons of observations and GDPs
Cold season (October through March) and warm season (April through
September) precipitation totals and mean Tmax, Tmin, and Tdew values for station
observations and GDPs over the Snake Range are shown in Figure 7. Typical valley-to-
mountain precipitation gradients were observed with NevCAN measurements and GDP
estimates, with seasonal totals increasing with elevation from west to east, and then
29
decreasing from the crest to the eastern valley floor (Figure 7a and 7b). The NevCAN
maximum measured precipitation during the cold season occurs at SN4 (297 mm), which
is located on the windward side of the Snake Range at a slightly higher elevation than
SN5, which is located on the lee side. The SN4 site is situated near a ridge line and the
surrounding vegetation is smaller and much less dense when compared to the vegetation
surrounding SN5. The WPS station recorded a much greater amount of cold season
precipitation (389 mm) compared to SN5 (292 mm), which would result in the greatest
cold season precipitation occurring on the lee slope. Based on these observations
(NevCAN and SNOTEL), great uncertainty remains as to where the true precipitation
maximum is occurring in the Snake Range. Similar observed precipitation characteristics
were found during the warm season.
Gridded data seasonal precipitation totals were generally found to be higher than
NevCAN observed totals (Figures 8a and 8b). During the cold season, differences ranged
from 132.8 mm (JA at SN3) to 6.0 mm (PRISM 800-m at SN6), and negative differences
were never observed. When compared to WPS (as opposed to SN5), all GDP differences
(GDP – obs) were negative and the smallest differences were found with PRISM 800-m
and 4-km (-2.3 mm and -9.4 mm, respectively). The positive differences found between
GDPs and NevCAN stations appear to be a result of WPS being the only high-elevation
control point in the area for GDPs in the Snake Range. Differences between station and
grid point elevation could not explain the corresponding precipitation differences. For
example, at SN3, the PRISM 4-km grid cell was approximately 400 m lower than the
station, but precipitation was always greater than observed. Overall for precipitation
30
comparisons, GDP performance was inconsistent, and it was found that finer grid
resolution did not always lead to smaller differences between GDPs and observations.
Maximum temperature biases (Figure 8c and 8d) ranged from 4.9 °C (JA) to -4.6
°C (PRISM 4-km) for cold and warm seasons. Large negative biases (colder than
observed) were found at the low-elevation sites of SN1 and SN2. These biases are
directly related to differences between GDP and station elevations, with SN1 and SN2
grid point elevations being higher than station elevations. However, several GDP
elevations were higher than station elevations leading to positive biases (warmer), and
lower grid point elevations relative to the station elevations with negative biases. For
example, the PRISM 4-km grid point at SN5 is 151 m higher than the SN5 station
elevation and a positive bias of 3.6 °C was found during the cold season.
NevCAN minimum temperature elevational gradients (Figure 7e and 7f) varied
from those of Tmax in that the two alluvial fan “foothill” stations (SN2 and SN6) were
warmer on average when compared to the neighboring valley floor stations (SN1 and
SN7) during both seasons. Previous studies have found that in complex terrain, Tmin can
vary greatly depending on station siting and associated local atmospheric decoupling and
cold-air drainage (Daly et al., 2009; Holden et al., 2011). During the nighttime hours, as
the boundary layer stabilizes (typically during clear sky conditions), cold air sinks and
tends to pool in low-lying areas, leading to temperature inversions near the surface and
warmer conditions in foothill locations (Gustavsson et al., 1998). The ability of GDPs to
capture this feature was variable, with PRISM 800-m being only GDP to capture the
inversions at SN2 and SN6 during both seasons, which is a result of PRISM’s use of
inversion height, topographic position and varying slopes with elevation (Daly et al.,
31
2008). In some instances, PRISM 4-km and JA were able to represent inversions, but the
magnitudes were much smaller than observed. These results highlight the need for
improved methods of interpolating Tmin observations over complex terrain.
Biases in Tmin (Figure 8e and 8f) ranged from 3.1 °C (JA) to -4.7 °C (PRISM 4-
km), with biases being generally slightly larger during the warm season. Although some
of the biases can be attributed to grid point and respective station elevation differences,
the large Tmin biases at SN2 and SN6 are a result of GDPs not being able to replicate the
inversion strength between valley floor and alluvial fan locations. Local lapse rates of
monthly Tmin (not shown) between SN1 and SN2, and between SN7 and SN6 averaged
over the water year were +7.6 °C/km and +9.4 °C/km, respectively, and were largely
underestimated by GDPs (PRISM 800-m was the closest to observations with water year
average Tmin lapse rates of +5.1 °C/km from SN1 to SN2 and +1.9°C/km from SN7 to
SN6).
With a general lack of humidity observations, relatively little is known about the spatial
behavior of near-surface humidity over complex terrain and the skill of GDPs to estimate
humidity. As expected, we found NevCAN station seasonal average Tdew to decrease with
elevation (Figure 7g and 7h). Except for Daymet, differences between GDP and observed
Tdew were generally small during the cold season (Figures 7g and 8g) and ranged from -
2.3 °C to 1.2 °C, whereas warm season biases (Figures 7h and 8h) were larger and
primarily negative, ranging from -4.7 °C to 0.5 °C. All GDPs, except for Daymet,
showed the same trends with elevation. Daymet estimated nearly constant Tdew with
respect to elevation during the cold season (Figure 7g) and increasing Tdew with elevation
during the warm season (Figure 7h). The lack of skill shown by Daymet is primarily due
32
to the underlying assumption that daily average Tdew is equal to Tmin. This assumption is
sometimes reasonable in humid regions, however, for semiarid to arid regions such as the
Great Basin, Daymet’s assumption of Tdew equaling Tmin is largely inaccurate; therefore
Daymet Tdew estimates are compromised.
Figure 7: Snake Range seasonal precipitation totals (a, b) and seasonal mean Tmax (c, d),
Tmin (e, f), and Tdew (g, h). Cold season is shown on the left (a, c, e, g) and warm season
on the right (b, d, f, h). X-axis is aligned west to east (left to right).
33
Figure 8: Snake Range seasonal bias (GDP - obs) for cold season (left) and warm season
(right) precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h). Variables needed to
calculate Tdew are not measured at WPS, therefore no Tdew values are shown.
To examine the ability of daily GDPs to capture measured daily variability of
temperature and precipitation, R2 and MAE were computed for cold and warm seasons
using Daymet and JA GDPs (Figures 9 and 10). Fairly good agreement was found
between measured and estimated precipitation events during the cold season (Figures 9a
and 9b), with JA consistently having higher R2
(0.68-0.80) and smaller MAE (58%-98%)
when compared to Daymet (R2: 0.41-0.71, MAE: 65%-114%). Contrasting results were
found with warm season precipitation (Figure 10a and 10b); with generally much lower
correlations, and higher MAE. Gridded data appear to be generating more daily misses
34
(GDP = 0, and observed > 0) and false alarms (GDP > 0, and observed = 0) during the
late spring and summer months (not shown). The nature of warm season precipitation
events is typically convective and associated with a monsoonal pattern, which leads to a
sporadic and non-uniform spatial distribution and lower correlations between GDPs and
observations.
Daymet showed higher R2 and lower MAE for Tmax (Figures 9c, 9d, 10c, and 10d)
and Tmin (Figures 9e, 9f, 10e, and 10f) at most locations; however, differences between
JA and Daymet error statistics were often times marginal. In general, higher MAE values
were found with Tmin, which is consistent with our seasonal results that highlight the
weakness in GPDs to simulate inversion strength. The downscaling of the 32-km NARR
temperature data to the 12-km NLDAS-grid, and finally to the 4-km JA grid is likely
leading to lager error when compared to Daymet, where observations are interpolated
directly to a 1-km grid. Not surprisingly, JA Tdew correlations were higher, and MAE was
lower than Daymet, especially during the warm season (Figures 9g, 9h, 10g, and 10h).
This is largely a reflection of the assumptions used in the Daymet algorithm (Tmin equals
Tdew). It should be noted that the calculation of Tdew from daily data with JA and Daymet
is a contributing source of error when compared to NevCAN Tdew that was computed
with 10-minute data. Large differences were found at the daily time step when
comparing NevCAN Tdew from 10-minute data to Tdew from daily data, and differences
often times exceeded 3 °C/day (not shown).
35
Figure 9: Snake Range cold season R2 (left) and MAE (right) computed at the daily time
step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and JA.
For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and Tdew
is expressed in °C.
36
Figure 10: Snake Range warm season R2 (left) and MAE (right) computed at the daily
time step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and
JA. For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and
Tdew is expressed in °C.
Sheep Range comparisons of observations and GDPs
As discussed in Section 2.1, tipping buckets largely undermeasure precipitation
during the cold season, and therefore Sheep Range tipping bucket measurements (Table
1, Figure 11a and 11b) must be considered inaccurate at the daily time step (and biased
low). Unfortunately, all Sheep Range stations are equipped with tipping buckets only,
except for SH4. Based on the assumption that frozen precipitation will occur at
temperatures of less than 0 °C, SH1 and YG were the only stations where all precipitation
37
events were classified as liquid during both seasons. When considering the remaining
four stations, a minimum of 38 percent of daily precipitation events were classified as
frozen during the cold season at SH2 and maximum of 91 percent at HP. Therefore, SH2,
SH3, and HP cold season precipitation measurements contain the highest degree of
uncertainty.
Differences between GDPs and observed precipitation were primarily positive
(wet) during the cold season (Figure 12a). Tipping bucket deficiencies are likely causing
undermeasurement at SH2, SH3, and HP, but this does not explain the large differences
found with JA, PRISM 800-m, and Daymet at SH4 (41.69 mm, 59.93 mm, and 71.54
mm, respectively) where precipitation measurements came from the Geonor weighing
gauge. This indicates that the instrumentation is likely not the only source of uncertainty.
An additional source of error in the comparisons may be due to the fact that the nearest
source of input data for GDPs in the region comes from the Spring Mountains (southwest
of the Sheep Range), which are considerably wetter than the Sheep Range. During the
warm season (when less uncertainty in tipping bucket measurements exist), GDP
seasonal precipitation totals were lower than observed at SH1, SH3, and SH4 and higher
than observed at SH2, with mixed results at YG and HP (Figure 12b). Large station-to-
station variability was found with respect to differences between GDP and observed
seasonal totals. For example, during the warm season at SH2, PRISM 800-m was found
to have the greatest difference (75.1 mm) and Daymet had the smallest difference (26.57
mm), whereas the opposite was found at SH3 with PRISM 800-m having the smallest
difference (-2.4 mm) and Daymet the largest difference (-64.8 mm).
38
Maximum temperature biases during the cold season (Figure 12c) ranged from -
1.67 °C (JA) to 5.77 °C (JA) and from -2.11 °C (JA) to 4.33 °C (JA) during the warm
season (Figure 12d). The consistently large warm biases found at HP can be primarily
explained by the large differences found between GDP and station elevations, with GDP
elevations being -88 m to -422 mm lower than the HP station elevation.
Observed seasonal mean Tmin (Figure 11e and 11f) was found to have similar
characteristics as the Snake Range, with the alluvial fan station (YG) being warmer than
the lower-elevation valley floor station (SH1) during both seasons. Daymet was the only
GDP to not capture the cold air drainage feature. This highlights the importance of
accounting for complex, non-monotonic elevational gradients and temperature inversions,
which are common throughout the Great Basin. Seasonal mean Tmin biases during the
cold season (Figure 12e) ranged from
-4.28 °C (JA) to 2.64 °C (PRISM 4-km) and from -4.26 °C (JA) to 0.88 °C (PRISM 800-
m) during the warm season (Figure 12f). As found at the Snake Range, Tmin biases are not
directly related to differences between GDP and station elevations. For example, cold
biases were found at HP with JA and Daymet during both seasons although the GDP
elevations were lower than the station elevation (-335 m and -88 m, respectively).
Both observed and GDP estimated seasonal mean Tdew was found to decrease with
elevation (Figures 11g and 11h). This is in contrast to the Snake Range, where Daymet
Tdew was found to increase with elevation during the warm season. Little consistency was
found in GDP seasonal mean Tdew biases (Figures 12g and 12h), with the exception of
Daymet showing a consistent and large positive (warm) bias during the cold season,
ranging from 3.42 °C to 6.63 °C.
39
Figure 11: Sheep Range seasonal precipitation totals (a, b) and seasonal mean Tmax (c, d),
Tmin (e, f), and Tdew (g, h). Cold season is shown on the left (a, c, e, g) and warm season
on the right (b, d, f, h). X-axis is aligned west to east (left to right). For precipitation
observations, filled circles represent tipping bucket gauges, and filled upside down
triangles represent weighing gauges.
40
Figure 12: Sheep Range seasonal bias (GDP - obs) for cold season (left) and warm season
(right) precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h).
Daily precipitation error statistics for the Sheep Range (Figures 13a, 13b, 14a, and
14b) showed JA to have higher correlations and lower MAE at all stations during both
seasons. At several locations during the cold season (SH1, YG, SH2, and HP), Daymet
R2 was low (< 0.1) and MAE was high (> 300%), while JA R
2 generally remained above
0.4. This may be partly due to the additional information regarding hourly precipitation
that NLDAS-2 obtains from radar, satellite, and NARR; whereas Daymet relies only on
station data and underlying regression relationships. The poor correlations in the cold
season are partly due to tipping bucket measurements, while additional uncertainty comes
41
from a lack of GDP station data input in this region. It should also be noted that in this
arid climate, precipitation occurs on only a small fraction of days (i.e. an average of 13%
of days in the cold season), so correlations will decrease rapidly for each day that GDPs
don’t match observed precipitation. The combination of no GDP input from surface
observations in the Sheep range, and primarily tipping bucket rain gauges, leads to great
uncertainty in both GDP estimates and NevCAN observations of precipitation in the
Sheep Range.
Daymet and JA Tmax correlations (Figures 13c and 14c) were quite similar and
indicate good agreement to observations (R2 always > 0.82), and the noticeably higher
MAE at HP (Figures 13d and 14d) could be attributed to the large differences between
grid cell and station elevation. For Tmin, error statistics were generally still good, but
lower than Tmax, which is again due to GDP errors with inversions. Daily Tdew error
statistics (Figures 13g, 13h, 14g, and 14h) were consistent with the Snake Range, with JA
always indicating less error than Daymet due to previously described deficiencies.
42
Figure 13: Sheep Range cold season R2 (left) and MAE (right) computed at the daily time
step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and JA.
For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and
Tdew is expressed in °C.
43
Figure 14: Sheep Range warm season R2 (left) and MAE (right) computed at the daily
time step for precipitation (a, b), Tmax (c, d), Tmin (e, f), and Tdew (g, h) using Daymet and
JA. For precipitation, MAE is expressed as a percentage, while MAE for Tmax, Tmin, and
Tdew is expressed in °C.
Summary and Conclusions
In this study, we utilized the Nevada Climate-ecohydrological Assessment
Network (NevCAN) data to quantify elevational gradients of precipitation, maximum and
minimum temperature (Tmax and Tmin), and dew point temperature (Tdew), along a west-to-
east transect in the Snake Range and a south-to-north transect in the Sheep Range.
NevCAN, along with additional observations, were used to evaluate four gridded data
products (GDPs) of varying spatial resolution (4-km to 800-m).
44
We have highlighted the challenges of providing reliable “ground truth” for
evaluating GDP precipitation estimates in remote areas. By identifying large differences
in water year (2012) precipitation totals between SN5 and the Wheeler Peak SNOTEL
(WPS) station (161 mm) and through the comparison of tipping bucket and weighing
gauge measurements presented in Section 3.1, we have highlighted several difficulties
associated with comparing measurements of precipitation to GDP estimates of
precipitation. The high GDP totals that were found with respect to NevCAN totals may
largely be due to WPS being the only GDP input in the Snake Range. At the Sheep
Range, perceived GDP “overestimation” is partly due to the use of tipping bucket rain
gauges as the source of baseline NevCAN measurements used for comparison. A second
contribution to the large differences between GDPs and Sheep Range observed
precipitation is due to lack of any stations in the Sheep Range being used as GDP input.
Potential users of gridded precipitation data should be aware that large uncertainty exists
where station density is low, and especially when considering small, remote mountain
ranges with no observations used as GDP input (such as the Sheep Range). It is highly
recommend that any observing network with automated precipitation measurements be
equipped with weighing-type gauges and wind shields, as this work and previous studies
(e.g., Humphrey et al., 1997, Rasmussen et al., 2011) have noted large errors associated
with tipping bucket measurements.
A key finding of this study was that temperature inversions at the alluvial fan
locations were identified at both NevCAN transects, highlighting the importance of
mountain transect observation networks. These findings are consistent with previous
research that has identified cold air drainage as being the cause of this inverted Tmin-
45
elevation relationship (Gustavsson et al., 1998; Daly et al., 2009; Holden et al., 2011).
The only GDP not able to replicate this temperature feature was Daymet; however, the
magnitude of the Tmin inversions observed at the Snake Range (mean monthly lapse rates
of > +15°C/km in some cases) was not estimated well by any GDP. Both Daymet and
PRISM use local linear regressions of climate and elevation. However, the slope of the
PRISM regression line can vary sharply with elevation, based on local inversion height
and topographic position information. In contrast, the Daymet regression function is
monotonic through the entire elevation range (Daly, 2006). Lack of realistic Tmin in GDPs
was one motivation for development of the new data set, Topography Weather (TopoWx;
Olyer et al. 2014), that uses satellite based land surface temperature as a predictor for
Tmax and Tmin.
Given that maximum temperature showed a strong relationship with elevation,
biases between GDPs and observations were strongly related to differences between GDP
and station elevation. In general, PRISM 800-m and Daymet contained smaller biases
than PRISM 4-km and JA for both Tmax and Tmin, indicating that spatial resolution of less
than 4 km can provide valuable details regarding temperature features.
We have highlighted a limitation of using an overly simplistic estimation for Tdew
(Tmin = daily average Tdew) in semi-arid to arid environments, which results in an
unrealistic increase of Tdew with elevation in the Snake Range and generally a large bias
compared to observations of Tdew. The combination of Daymet’s assumption of Tmin =
daily average Tdew and the inability to reproduce temperature inversions make the
application of Daymet to estimate humidity levels in semi-arid and arid areas largely
uncertain. Reasonable Tdew estimates were provided by PRISM and JA. However, the
46
calculation of Tdew from JA and Daymet daily data was a contributing source of error
when comparing these estimates to NevCAN Tdew, which was computed with 10-minute
data.
This research highlights the importance of conducting local analyses of
observations and potential measurement errors to gain an understanding of potential GDP
biases prior to use in hydro-climatic applications. Although local results may vary, this
work complements other hydro-climatic studies throughout the Great Basin region where
geographical attributes are similar to the NevCAN transects. Procedures and results from
this study are useful for improving our understanding of GDP evaluation and analyses
related to hydro-climatic assessments in semi-arid and arid climates.
Acknowledgements
The authors would like to thank three anonymous reviewers for helping to improve this
manuscript through constructive feedback. We would also like to acknowledge Greg
McCurdy, Dr. Lynn Fenstermaker, Brad Lyles, and Dr. Jay Arnone for their valuable
contributions regarding the NevCAN instrumentation and site characteristics. This work
was partially funded by National Science Foundation under grant number EPS-0814372
and by the Desert Research Institute Divisions of Atmospheric Science and Hydrologic
Science faculty and graduate research. The work was also funded by Landsat Science
Team funding under USGS grant number G12PC00068 and the U.S. Bureau of
Reclamation Nevada Water Resources Evaluation Program in collaboration with the
Nevada State Engineer’s Office, funded by a grant under Public Law 109-103, Section
208(a), Cooperative Agreement 06FC204044.
47
References
Abatzoglou, J. T., 2011: Development of gridded surface meteorological data for
ecological
applications and modeling. Int. J. Climatol. 33, 121–131, DOI: 10.1002/joc.3413.
Ackerly, D. D., S. R. Loarie, W. K. Cornwell, S. B. Weiss, H. Hamilton, R. Branciforte,
and N. J. B. Kraft, 2010: The geography of climate change: implications for
conservation biogeography. Diversity Distrib., 16, 476-487.
Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration,
guidelines for
computing crop water requirement. FAO Irrig. and Drain. Paper 56, Food and
Agricultural
Organization of the United Nations. Rome, Italy. 300.
Alter, J. C., 1937: Shielded storage precipitation gages. Mon. Wea. Rev., 65, No. 7, 262-
265.
Bandyopadhyay, J. et al., 1997: Highland waters – a resource of global significance. In:
Messerli, B. and Ives, J. D. (eds.) Mountains of the world – a global priority.
NewYork and London: Parthenon, pp. 131-155.
Barnett, T. P., J. C. Adam, and D. P. Lettenmaier, 2005: Potential impacts of a warming
climate on water availability in snow-dominated regions. Nature, 438, 303-309.
Bradley, B. A., 2009: Regional analysis of the impacts of climate change on cheatgrass
invasion shows potential risk and opportunity. Glob. Change Biol., 15, 196-208, doi:
10.1111/j.1365-2486.2008.01709.x.
48
Burns, A. G. and W. Drici, 2011: Hydrology and water resources of Spring, Cave, Dry
Lake, and
Delamar Valleys, Nevada and vicinity. Presentation to the Office of the Nevada State
Engineer: Southern Nevada Water Authority, Las Vegas, Nevada.
Crago, R. D., R. J. Qualls, and M. Feller, 2010: A calibrated advection-aridity
evaporation model requiring no humidity data, Water Resour. Res., 46, W09519,
doi:10.1029/2009WR008497.
Cosgrove, B. A., and Coauthors, 2003: Real-time and retrospective forcing in the North
American Land Data Assimilation System (NLDAS) project. J. Geophys. Res., 108,
doi:10.1029/2002JD003118.
Daly, C., R. P. Neilson, and D. L. Phillips, 1994: A statistical-topographic model for
mapping
climatological precipitation over mountainous terrain. J. Appl. Meteor., 33, 140-158.
——, 2006: Guidelines for assessing the suitability of spatial climate data sets. Int. J.
Climatol., 26, 707-721.
——, M. Halbleib, J. I. Smith, W. P. Gibson, M. K. Doggett, G. H. Taylor, J. Curtis, and
P. P. Pasteris, 2008: Physiographically sensitive mapping of climatological
temperature and precipitation across the conterminous United States. Int. J. Climatol.,
doi: 10.1002/joc.1688.
——, C., D. R. Conklin, and M. H. Unsworth, 2009. Local atmospheric decoupling in
complex topography alters climate change impacts. Int. J. Climatol., DOI:
10.1002/joc.2007.
Eakin, T. E., 1966: A regional interbasin ground-water system in the White River area,
49
southeastern Nevada, Water Resour. Res, 2, no. 2, 251-271.
Epstein, B. J., G. M., Pohll, J. L. Huntington, R. H. Carrol, 2010: Development and
uncertainty analysis of an empirical recharge prediction model for Nevada’s Desert
Basins. Journal of the Nevada Water Resources Association. 5:1, 1-22. Available at:
http://www.dri.edu/images/stories/divisions/dhs/dhsfaculty/Justin-
Huntington/Epstein_et_al_2010.pdf.
Fassnacht, S. R., 2004: Estimating Alter-shielded gauge snowfall undercatch, snowpack
sublimation, and blowing snow transport at six sites in the coterminous USA. Hydrol.
Process., 18, 3481-3492.
Feld, S. I., N. C. Cristea, and J. D. Lundquist, 2013: Representing atmospheric moisture
content along mountain slopes: examination using distributed sensors in the Sierra
Nevada, California. Water Resour. Res., 49, 4424-4441, doi:10.1002/wrcr.20318.
Flint, A. L., Flint, L. E., Hevesi, J. A., and Blainey, J. M., 2004: Fundamental concepts of
recharge in the Desert Southwest: A regional modeling perspective, in Hogan, J.F.,
Phillips, FM., and Scanlon, B.R., eds., Ground-Water Recharge in a Desert
Environment: The Southwestern United States: American Geophysical Union, Water
Science and Application Series 9, 159-184.
Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solid precipitation
measurement intercomparison. WMO Instruments and Observing Methods Report
No. 67, WMO/TD No. 872.
Gustavsson, T., M. Karlsson, J. Bogren, and S. Lindqvist, 1998: Development of
temperature patterns during clear nights. J. Appl. Meteor., 37, 559-571.
50
Holden, Z. A., J. T. Abatzoglou, L. S. Baggett, and C. Luce, 2011: Empirical
downscaling of daily minimum air temperature at very fine resolutions in complex
terrain. Agr. Forest Meteorol., doi: 10.1016/j.agrformet.2011.03.01.
Houghton, J. G., 1979: A model for orographic precipitation in the north-central Great
Basin. Mon. Wea. Rev., 107, 1462-1475.
Humphrey, M. D., J. D. Istok, J. Y. Lee, A. Hevesi, and A. L. Flint, 1997: A new method
for automated dynamic calibration of tipping-bucket rain gauges. J. Atmos. Ocean.
Tech., 14, 1513-1519.
Huntington, J. L., and D. J. McEvoy, 2011: Climatological estimates of open water
evaporation from selected Truckee and Carson River basin water bodies, California
and Nevada. Desert Research Institute Publication No. 41254. Available at:
http://www.dri.edu/images/stories/divisions/dhs/dhsfaculty/Justin-
Huntington/Huntington_and_McEvoy_2011.pdf
——, J. L. Szilagyi, S. W. Tyler, and G. M. Pohll, 2011: Evaluating the complementary
relationship for estimating evapotranspiration from arid shrublands, Water Resour.
Res., doi:10.1029/2010WR009874.
——, and R. G. Niswonger, 2012: Role of surface-water and groundwater interactions on
projected summertime streamflow in snow dominated regions: an integrated
modeling approach. Water Resour. Res., 48, W11524, doi:10.1029/2012WR012319.
Jeton, A. E., S. A. Watkins, T. J. Lopes, and J. L. Huntington, 2006: Evaluation of
precipitation
estimates from PRISM for the 1961–90 and 1971–2000 data sets, Nevada. Scientific
51
Investigations Report 2005-5291. U.S. Geological Survey, Carson City, NV.
Available at: http://pubs.usgs.gov/sir/2005/5291/PDF/SIR2005_5291.pdf.
Leger, E. A., 2013: Annual plants change in size over a century of observations. Glob.
Change Biol. 19, 2229-2239. doi: 10.1111/gcb.12208.
Loarie, S. R., P. B. Duffy, H. Hamilton, G. P. Asner, C. B. Field, and D. D. Ackerly,
2009: The velocity of climate change. Nature, 462, 1052-1055.
Lundmark, K. W., G. M. Pohll, and R. W. H. Carroll, 2007: A steady-state water budget
accounting model for the carbonate aquifer system in White Pine County, Nevada and
adjacent areas in Nevada and Utah: University of Nevada, Desert Research Institute
Publication 41235, 56 p. Available at:
http://www.dri.edu/images/stories/research/projects/BARCAS/41235_2007_barcas_d
sc_report.pdf.
Lundquist, J. D., J. R. Minder, P. J. Neiman, and E. Sukovich, 2010: Relationship
between barrier jet heights, orographic precipitation gradients, and streamflow in the
Northern Sierra Nevada. J. Hydrometeor., 11, 1141-1156.
McEvoy, D. J., J. L. Huntington, J. T. Abatzoglou, and L. M. Edwards, 2012: An
evaluation of multiscalar drought indices in Nevada and eastern California. Earth
Interact., 16, No. 18, DOI: 10.1175/2012EI000447.1
Mensing, S., S. Strachan, J. Arnone, L. Fenstermaker, F. Biondi, D. Devitt, B. Johnson,
B. Bird, and E. Fritzinger, 2013: A network for observing Great Basin climate
change. Eos, 94, No. 11, 105-106.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer.
Meteor. Soc., 87, 343–360.
52
Mitchell, K. E., D. Lohmann, P. R. Houser, E. F. Wood, J. C. Schaake, A. Robock, B. A.
Cosgrove, J. Sheffield, Q. Duan, L. Luo, R. W. Higgins, R. T. Pinker, J. D. Tarpley,
D. P. Lettenmaier, C. H. Marshall, J. K. Entin, M. Pan, W. Shi, V. Koren, J. Meng, B.
H. Ramsay, A. A. Baily, 2004: The multi-institution North American Land Data
Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a
continental distributed hydrological modeling system. J. Geophys. Res., 109, D07S90,
doi:10.1029/2003JD003823.
Murray, F. W., 1967: On the computation of saturation vapor pressure. J. Appl.
Meteorol., 6, 203-204.
Nevada State Engineer’s Office (NSEO), 2012. Ruling (#6164) In the Matter of
Applications 54003 through 54021, Inclusive, Filed to Appropriate the Underground
Waters of the Spring Valley Hydrographic Basin (184), Lincoln and White Pine
Counties, Nevada,
218 pp. Available at:
http://www.snwa.com/assets/pdf/ws_gdp_spring_ruling_2012.pdf.
Nevada Bureau of Land Management, 2012: Clark, Lincoln, and White Pine counties
groundwater development project draft environmental impact statement. Vol. 1 A.
Nevada Division of Water Resources, Division of Energy Management, and State
Climate
Office, 2012: State of Nevada Drought Plan. Reno, NV.
Oyler, J.W., A. Ballantyne, K. Jencso, M. Sweet, and S.W. Running, 2014: Creating a
topoclimatic daily air temperature dataset for the conterminous United States using
homogenized station data and remotely sensed land skin temperature. Int. J. Climatol.
53
Porinchu, D. F., S. Reinemann, B. G. Mark, J. E. Box, and N. Rolland, 2010: Application
of a midge-based inference model for air temperature reveals evidence of late-20th
century warming in sub-alpine lakes in the central Great Basin, United States.
Quatern. Int., 215, 15-26. doi:10.1016/j.quaint.2009.07.021.
Rangwala, I. and J. R. Miller, 2012: Climate change in mountains: a review of elevation-
dependent warming and its possible causes. Climatic Change, 114, 527-547.
Rasmussen, R., B. Baker, J. Kochendorfer, T. Meyers, S. Landolt, A. P. Fischer, J. Black,
J. M.
Theriault, P. Kucera, D. Gochis, C. Smith, R. Nitu, M. Hall, K. Ikeda, and E.
Gutmann, 2011: How well are we measuring snow? The NOAA/FAA/NCAR winter
precipitation test bed. Bull. Amer. Meteor. Soc., 93, 811-829, DOI:10.1175/BAMS-D-
11-00052.1.
Rauscher, S. A., J. S. Pal, N. S. Diffenbaugh, and M. M. Benedetti, 2008: Future changes
in snowmelt-driven runoff timing over the western US. Geophys. Res. Lett, 35,
doi:10.1029/2008GL034424.
Sieck, L. C., S. J. Burges, and M. Steiner, 2007: Challenges in obtaining reliable
measurements of point rainfall. Water Resour. Res., 43, W01420,
doi:10.1029/2005WR004519.
Smith, R. B., 1979: The influence of mountains on the atmosphere. Adv. Geophys., 21,
87-230.
Thornton, P. E., S. W. Running, and M. A. White, 1997. Generating surfaces of daily
meteorology variables over large regions of complex terrain. J. Hydrol., 190, 214-
251.
54
——, and S. W. Running, 1999: An improved algorithm for estimating incident daily
solar radiation from measurements of temperature, humidity, and precipitation. Agr.
Forest Meteorol., 93, 211-228.
——, H. Hasenauer, and M. A. White, 2000: Simultaneous estimation of daily solar
radiation and humidity from observed temperature and precipitation: an application
over complex terrain in Austria. Agr. Forest Meteorol., 104, 255-271.
——, M. M. Thornton, B. W. Mayer, N. Wilhelmi, Y. Wei, R. B. Cook, 2012: Daymet:
daily surface weather on a 1 km grid for North America,1980 - 2008. Acquired online
(http://daymet.ornl.gov/) on 01/01/2012 from Oak Ridge National Laboratory
Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A.
doi:10.3334/ORNLDAAC/Daymet_V2.
Welch, A. H., D. J. Bright, and L. A. Knochenmus, eds., 2007: Water resources of the
Basin and Range carbonate-rock aquifer system, White Pine County, Nevada, and
adjacent areas in Nevada and Utah: U.S. Geological Survey Scientific Investigations
Report 2007–5261, 96 p.
Willmott, C. J. and S. M. Robeson, 1995: Climatologically aided interpolation (CAI) of
terrestrial air temperature. Int. J. Climatol., 15, 221-229.
Yang, D., B. E. Goodison, and J. R. Metcalfe, 1998: Accuracy of NWS 8” standard
nonrecording precipitation gauge: results and application of WMO intercomparison.
J. Atmos. Ocean. Tech., 15, 54-68.
Zhu, J. and Young, M. H., 2009: Sensitivity and Uncertainty of Ground-Water Discharge
Estimates for Semiarid Shrublands. J. Am. Water Resour. As., 45, 641–653.
doi: 10.1111/j.1752-1688.2009.00312.x
55
The Evaporative Demand Drought Index: CONUS-wide Assessment Against
Common Drought Indicators
Daniel McEvoy
1, 2, Justin Huntington
1, Mike Hobbins
2, Andrew Wood
3, Charles Morton
1
1Desert Research Institute, Reno, NV
2University of Nevada, Reno
3National Integrated Drought Information System, Boulder, CO
4National Center for Atmospheric Research, Boulder, CO
56
Abstract
Precipitation, soil moisture, and air temperature are the most commonly used
climate variables to monitor drought, however other climatic factors such as solar
radiation, wind speed, and specific humidity can be important drivers in the depletion of
soil moisture and evolution and persistence of drought. This work provides an assessment
of the Evaporative Demand Drought Index (EDDI) at multiple time scales for several
hydroclimates as a companion study to Hobbins et al. (2015) by examining EDDI and
individual evaporative demand components as they relate to the dynamic evolution of
flash drought over the central US, characterization of hydrologic drought over the
western US, and comparison to commonly used drought metrics of the US Drought
Monitor, Standardized Precipitation Index (SPI), Standardized Soil Moisture Index (SSI),
and the Evaporative Stress Index (ESI). Results show that EDDI has the strongest
relationships to SPI and SSI over Texas, Oklahoma, and much of the desert Southwest,
while comparisons to summer ESI revealed a hotspot over much of the central US. At
short time scales, spatial distributions and time series results illustrate that EDDI is useful
for flash drought identification, and can serve as a leading indicator by as much as two
months in advance of the USDM, SPI, and SSI. Our results illustrate the benefits of
physically based evaporative demand estimates, and demonstrate EDDI’s utility and
effectiveness in an easy-to-implement operational early warning and long-term
hydrologic drought monitoring tool for agricultural and drought monitoring, and potential
application to seasonal forecasting and fire-weather monitoring.
57
Introduction
Drought is a complex and naturally occurring process with adverse effects on
society, primarily through degradation and loss of agricultural crops and depletion of
water resources (i.e., streamflow and reservoir storage). Recent examples are instructive:
in California, the extended drought that began in late 2011 is still ongoing, and the 2011-
2014 three-year average precipitation (Prcp) record indicates that this period is the second
driest in recorded history (Seager et al., 2015); in 2011, Texas experienced extreme Prcp
deficits; while in 2011 and 2012 record-breaking temperatures (Tair) and high wind speed
(Uz) played a significant role in drought intensification over much of the central US (Karl
et al. 2012, Cattiaux and Yiou 2013). Total economic losses are estimated to be $2.7
billion, $7.7 billion, and more than $35 billion for the California, Texas, and central US
droughts, respectively. While conditions in Texas deteriorated over many months in
2011, the depletion of moisture over the central US in 2011 occurred at a much faster
rate. This fast onset of drought has recently been termed “flash drought” (Svoboda et al.
2002). The physical mechanisms driving flash droughts have been largely neglected from
traditional drought metrics. Hence there is a growing need for continued development of
physically based drought metrics that capture important land surface-atmospheric
feedbacks, and provide sufficient early warning.
It has been common practice in recent decades to monitor and analyze drought using
metrics driven by Prcp and Tair only. The two most commonly used drought indices are
the Palmer Drought Severity Index [PDSI; Palmer (1965)], which relies on monthly Tair
and Prcp, and the Standardized Precipitation Index [SPI; McKee (1993)], which relies on
58
Prcp only. While the PDSI and SPI have proven useful for providing valuable
information regarding hydrologic and meteorological drought, these metrics have
limitations at short time scales and fail to account for the effects of other important
drought meteorological and radiative forcings such as specific humidity (q), Uz, and
downwelling shortwave radiation (Rd). The most heavily used dataset for decision
making with regards to drought is the US Drought Monitor [USDM; Svoboda et al.
(2002)], which relies on a blend of metrics (including PDSI and SPI) and climate data
(e.g., soil moisture (SM), streamflow, and snow water equivalent) to produce weekly
maps of drought severity. The USDM could be improved through the inclusion of
important hydrometeorological forcings key to identifying flash and long-term drought
through the use of physically based evaporative demand (E0) estimates.
Other operational products could similarly be improved with the inclusion of
physically based E0 estimates. For example, the U.S. operational PDSI, produced by the
National Oceanic and Atmospheric Administration (Heddinghaus and Sabol 1991),
continues to use Tair-based E0 estimates (i.e. Thornthwaite 1948) within the PDSI
formulation despite the fact that there have been a number of studies that recommend the
use of physically based formulations of E0 (Milly and Dunne 2011; Hobbins et al. 2008,
2012; Hobbins 2015). Both Dai (2011) and van der Schrier et al. (2011) found PDSI to be
largely insensitive to E0 parameterization during the 20th
and early 21st century. On the
other hand, Sheffield et al. (2012) found major differences between the PDSI driven with
Tair- and physically-based E0 estimates, especially from the mid-1990s through 2008,
with Tair-based E0 estimates showing a significant drying trend in PDSI, and physically
based E0 estimates indicating no significant trend in global drought severity. The role of
59
physically based E0 estimates in drought monitoring and prediction remains an active—
and to some degree, controversial—area of research, and is a focus of this paper.
Recent studies have shown that actual evapotranspiration (ET), which is obtained
through the use of thermal and optical satellite remote sensing or land surface models,
used in combination with physically based E0 can be used as a drought indicator by
inherently accounting for feedbacks between the land surface-atmosphere interface
through the use of ratios of ET to E0 (Yao et al. 2010; Anderson et al. 2007a, 2007b,
2011; Mu et al. 2013; Otkin et al. 2013a, 2013b). However, the use of thermal and optical
remote sensing data for operational drought monitoring has limitations, such as cloud
cover, spurious ET estimates in semi-arid and arid regions, satellite inter-arrival times
that have to be interpolated, and uncertain simulated surface energy balance in
mountainous regions, especially where seasonal snowpack exists.
In an effort to complement and overcome some of the limitations of the
aforementioned metrics, the companion paper (Hobbins et al. – this issue) developed the
Evaporative Demand Drought Index (EDDI), which relies solely on physically based E0
estimates derived from a near-real-time (2-5 day latency), easily accessible land surface
forcing dataset: the North American Land Data Assimilation System Phase-2 [NLDAS-2;
Mitchell et al. (2004)]. Hobbins et al. (this issue) describe two primary physical
feedbacks between ET and E0 that form the rationale for EDDI: a complementary
relationship under water-limited conditions (extended drought) where ET and E0 vary in
opposing directions (Bouchet 1963), and parallel variations under energy-limited
conditions at the onset of flash drought. Under both scenarios, EDDI was found to
60
respond to drying and wetting anomalies of major components of the hydrologic cycle at
various time scales (Hobbins et al. - this issue).
This paper builds upon the work of Hobbins et al. (this issue) through a robust
CONUS-wide assessment of EDDI against several commonly used drought indices, and
outlines a second standardization option that acts to reduce errors in comparing multiple
drought indices through space and time. Data sources, E0 formulation, and statistical
procedures to calculate EDDI are presented first, followed by comparisons of EDDI to
other commonly used drought metrics, a flash drought case study over the central US,
and finally, extended drought case studies over the western US.
Data and Methods
Evaporative demand
Daily bias-corrected and spatially disaggregated (from 12 km to 4 km) NLDAS-2
gridded meteorological data [METDATA; Abatzoglou (2011)] are used to compute E0 on
a daily basis for 1979 to 2013. Maximum and minimum temperature at 2-m (Tmax and
Tmin), q at 2-m, Rd, and 10-m wind speed (U10) were obtained from the University of
Idaho (http://metdata.northwestknowledge.net/). A variety of methods has been
developed to compute E0 including Tair-based methods (e.g., Thornthwaite 1948,
Hargreaves and Samani 1985), radiation-based methods (Priestley and Taylor 1972), and
radiation - aerodynamic combination methods that incorporate Tmax, Tmin, Rd, U10, and q,
such as the Penman-Monteith (PM) approach (Monteith 1965). A priori, it is generally
assumed that if the necessary data resources are available, a full-form physically based
61
method, such as PM, should be used over methods based only on Tair or radiation.
Hobbins et al. (2012) and Hobbins (2015) demonstrated that the primary drivers of E0
variability differ across the US, and with aggregation period (e.g., monthly vs. annual)
and season. For example, during summer months U10 is the primary driver of E0
variability over much of the Great Basin, while Rd is the primary driver of variability
over much of the southeast US. In this study, we use reference ET (ET0) from the PM-
based American Society of Civil Engineers Standardized Reference ET equation (ASCE-
EWRI, 2005) for E0.
Evaporative Demand Drought Index
A probability-based standardized climate variable can be obtained using parametric or
non-parametric methods. Parametric methods use a single probability distribution to fit a
time series (e.g., Gamma distribution for SPI), where probabilities are transformed to
standardized values through an inverse normal approximation. However, a single
probability distribution may not always be appropriate at large spatial scales, and several
studies have documented these limitations with SPI (Guttman 1999; Quiring 2009) and
Standardized Streamflow Index (Vicente-Serrano et al. 2012). The Evaporative Demand
Drought Index (EDDI) presented in Hobbins et al. (this issue) is calculated from a simple
Z-score based on the mean and standard deviation of a given accumulated ET0 time
series. Here, we deviate from Hobbins et al. (this issue) by using a probability-based
approach for EDDI to allow for more consistent comparisons between EDDI against
other standardized indices.
To overcome the limitations of a parametric approach, ET0 probabilities (P(x)) are
obtained through the empirical Tukey plotting position (Wilkes 2011):
62
𝑃(𝑥𝑖) =𝑖 − 0.33
𝑛 + 0.33 ,
where i is the rank in the historical time series (from 1 to 35, with 1 being the max ET0
value and 35 being the min) of the observed value, and n is the number of observations.
EDDI values are obtained from empirically derived probabilities through an inverse
normal approximation (Abramowitz and Stegun 1965) at time scales of 1, 3, 6, 9, and 12
months. Comparisons between EDDI values derived from the simple z-score outlined in
Hobbins et al. (this issue) and the formulation presented here showed negligible
differences in identifying wet and dry periods, but the plotting position approach was
ultimately chosen in this paper to maintain consistency when comparing multiple indices
outlined below. This method follows Hao and AghaKouchak (2014), where the plotting
position approach was used to compute SPI, Standardized Soil Moisture Index (SSI) and
Multivariate Standardized Drought Index (MSDI). Farahmand and AghaKouchak (2015)
recommend this plotting position approach to maintain consistency when comparing
several standardized drought indices.
NLDAS-based drought metrics
To assess the ability of EDDI to identify historical drought periods, EDDI is
compared to SPI and SSI using monthly Prcp and simulated SM from NLDAS-2 (Xia et
al. 2012a, 2012b). NLDAS-2 Prcp is primarily derived from Climate Prediction Center
gridded daily gauge data {with a topographical adjustment from the Parameter-elevation
Regressions on Independent Slopes Model [PRISM; Daly et al. (1994)]}. NLDAS-2 SM
is derived from the Variable Infiltration Capacity land surface model [VIC; Liang et al.
(1994)], and represents the average SM from the top 100 cm of the soil column. Monthly
63
NLDAS-2 data were obtained for the period of 1979 to 2013 with a native grid spacing of
0.125°. To compare EDDI to NLDAS-2 drought indices, all NLDAS-2 data were
resampled to the 4-km (~1/16°) UI METDATA grid using a bilinear interpolation.
Monthly Prcp and SM were accumulated at five time scales (1, 3, 6, 9, and 12 months),
and standardized following the EDDI methodology of plotting positions and inverse
normal approximation. Pearson linear correlation coefficients between EDDI and
standardized NLDAS-2 variables were computed for each month (n = 35 years) at the
five time scales.
Evaporative Stress Index
The ESI (Anderson et al. 2007b, 2011) represents standardized anomalies in the ET
fraction of reference ET (i.e., ET/ET0), with ET obtained through satellite-assisted
modeling of the land surface energy balance. ET and other land-surface energy balance
components are retrieved using satellite optical and thermal imagery, to force the
Atmosphere-Land Exchange Inverse surface energy balance model [ALEXI; Anderson et
al. (1997, 2007a)]. Atmospheric variables needed to drive ALEXI come from the North
American Regional Reanalysis [NARR; Mesinger et al. (2006)].
Weekly ESI data were provided (courtesy of Martha Anderson, USDA, and Chris
Hain, University of Maryland) over the US for 2000 to 2013 at a 4-km spatial resolution
and were aggregated to time scales of 1, 2, and 3 months. To obtain a constant
comparison between EDDI and ESI, EDDI was recalculated using the same period of
record as the ESI, and the same aggregation time scales. ESI data were resampled using a
bilinear interpolation to match the EDDI grid. No downscaling was necessary as both
64
grids were of identical spatial resolution. Pearson linear correlation coefficients between
EDDI and ESI were computed for each week over the 14-year period and at all five time
scales.
United States Drought Monitor
The USDM (Svoboda et al. 2002) was used as another metric to validate EDDI, with
the primary goal of identifying differences between the two metrics during the evolution
of drought through time and space. The USDM is derived from a blend of drought
metrics adjusted using local expert knowledge to develop weekly drought severity maps
over CONUS (Svoboda et al. 2002; Anderson et al. 2013). The USDM classification
system of drought ranges from D0 (abnormally dry) to D4 (exceptional drought). For
results where the USDM is compared, all drought metrics were converted to USDM
classes (Table 1). The comparisons of EDDI to the USDM are necessarily qualitative
because the USDM is a blend of information at several different time scales, whereas
EDDI represents a single time scale.
USDM data (2000 to 2013) were downloaded as ESRI shapefiles provided by the
National Drought Mitigation Center, and rasterized to match the 4-km EDDI grid, to
create a USDM class map of integer values of drought intensity ranging from 0 to 4 (i.e.,
D0 = 0, D1 = 1, D2 = 2, D3 = 3, and D4 = 4);
Table1. Drought classes for comparing USDM to SPI, SSI, ESI, and EDDI.
Category Description SPI, SSI, and ESI
percentiles
EDDI percentiles
D0 Abnormally Dry 21-30 70-79
D1 Moderate Drought 11-20 80-89
D2 Severe Drought 6-10 90-94
D3 Extreme Drought 3-5 95-97
D4 Exceptional Drought 0-2 98-100
65
Results and Discussion
NLDAS-2 drought index correlations with EDDI
Correlations between EDDI and NLDAS-2 drought indices (EDDI-SPI and EDDI-
SSI) for 1, 6, and 12 month time scales are shown in Figure 1. Positive EDDI values
indicate drought, and negative SPI and SSI values indicate drought, therefore strong
negative correlations represent similar drought signals between EDDI and both SPI and
SSI over the 35-year period of record. At the 1-to 12-month time scales correlations
between EDDI and SPI and SSI are strongest (more negative) over much of the
southwestern and southcentral US (with the exception of 1-month SSI), and highest in
Texas (r <-0.7). The northeast is region of general weak correlations for both EDDI-SPI
and EDDI-SSI, with the Midwestern states of OH, IN, and MI being a weak spot for
EDDI-SPI only. Spatial correlations at 6 and 12 month time scales are quite similar
(Figure 1c-1f), and generally much stronger than at the 1-month time scale (Figure 1a and
1b). Over the northeastern US, EDDI-SPI correlations remain fairly weak at longer
timescales, while EDDI-SSI correlations improve over OH, WV, NY, and PA (Figure 1c-
1f).
Weak correlations to 1-month SSI over the west may be explained by above average
Tair and Rd (driving EDDI upwards) that can lead to increased snow melt and SM, and a
short term wetting signal from SSI, particularly during the winter months. Positive
correlations of EDDI-SPI and EDDI-SSI over the northeastern US are caused by energy-
limited conditions as opposed to water-limited conditions. In such energy-limited regions,
66
the rate of change in ET is generally proportional and in the same direction as ET0 (Han
et al. 2014; Hobbins et al. - this issue).
Figure 1: Correlation coefficient between EDDI and SPI at (a) 1-month, (c) 6-month, (e)
12-month, and SSI (b) 1-month, (d) 6-month, and (f) 12-month time scales.
Figure 2 highlights four regions of interest selected for individual monthly correlation
analysis. The Central Valley of California (CA) and Iowa (IA) are two major agricultural
regions where drought impacts can have adverse effects on crop production. East-central
67
Texas (TX) is part of a region that has been identified as a global “hot spot” for strong
land surface-atmospheric coupling (Koster et al. 2004, 2006); therefore strong correlation
of SM and Prcp to EDDI is expected. Pennsylvania (PA) is an area identified by Koster et
al. (2009) where SM is generally high and exerts little control on ET due to prevailing
energy limiting conditions, even during times of severe meteorological drought. This
observation is consistent with low correlations found in Figure 1 in parts of the northeast
US. The following section further highlights how ET0 anomalies (i.e., EDDI) in PA relate
to SM- and Prcp-driven droughts.
Figure 2: Shading indicates METDATA terrain height (m) and red boxes indicate area-
averaging domains for Figures 3 and 4. IA, TX, and PA boxes are 50 x 100 4-km
METDATA pixels (200 km x 400 km), and CA box is 25 x 25 pixels (100 km x 100 km).
Individual monthly correlations between EDDI and NLDAS-2 derived indices at
various time scales are shown in Figure 3 for these regions of interest. For each of the
68
selected regions shown in Figure 2, EDDI correlations to SSI and SPI were area-averaged
over all pixels. For the TX region (Figure 3a and 3e), seasonality and time scale had little
impact on the strength of correlations, and generally showed strong inverse relationships
(r < -0.6 for SPI and r < -0.7 for SSI) during most months and time scales, reinforcing the
conclusions of Koster et al. (2004, 2006).
For the CA region, large seasonal and time scale dependent variations were found,
especially at the 1-month time scale for both SPI and SSI (Figure 3b and 3f). Correlations
ranged from +0.20 to -0.82, with the highest correlations occurring at the 6- to 12-month
time scales during the growing season. An exceptionally weak correlation (-0.13) was
found with SPI during July at the 1-month time scale. July is the driest month of the year
for the Central Valley of CA, and most Julys see zero Prcp accumulation. This limits the
negative range of the 1-month SPI (McEvoy et al. 2012) causing poor correlations with
EDDI. Furthermore, when it does rain during dry summer months it occurs from isolated
convective activity over a single day: even if most of the month was warm, cloud-free,
and dry (leading to a drought signal from EDDI), the SPI will show a wet anomaly. A
more consistent stepped correlation pattern was revealed at longer time scales, where r
values < -0.7 were found during the spring (April, May, and June) for 3-month, spring
and summer (July, August, and September) for 6-month, and summer and fall (October,
November, and December) for 9- and 12-month periods.
Iowa was similar to Texas in that little variability was found in correlations (r-values
only ranged from -0.5 to -0.7), with the exception of the 1-month time scale. Lower
correlations at 1-month time scale during the fall and winter should be expected with SSI,
since the top 100 cm of ground is typically frozen during these months, and land surface-
69
atmospheric coupling is weak. There is a rapid increase in correlation at the 1-month time
scale during the late spring and summer.
Correlations for PA region were the weakest of the four analyzed, with notably higher
correlation to SSI (Figure 3h) when compared to SPI (Figure 3d). For SPI (Figure 3d), r-
values never exceed -0.56, while for SSI (Figure 3h) r-values ranged from -0.60 to -0.69
during the summer and early fall at 1-, 3- and 6-month time scales. Weak correlations
were found to be both slightly positive and negative (-0.30 < r < +0.20) for SPI and SSI
at the 1-month time scale during fall and winter, and for winter and spring months at
other time scales. Results shown in Figure 3 illustrate that EDDI may be particularly
useful for flash drought and seasonal drought monitoring, especially during the growing
season.
70
Figure 3: Monthly correlations between EDDI and SPI (top row) and SSI (bottom row)
at all time scales for (a, e) TX, (b, f) CA, (c, g) IA, and (d, h) PA. Y-axis indicates
ending month of each time scale, and x-axis shows time scale (months). Shading
indicates correlation coefficients.
Soil moisture is typically a slowly varying component of the hydro-climatic
system compared to variations in ET0; therefore EDDI could serve as a leading indicator
for identifying soil moisture deficits. Correlations between EDDI and SSI at coincident
time scale and ending month (as presented in Figure 3) may not be the most robust due to
this time lag between SM and ET0. To demonstrate the potential value of EDDI as a
leading drought indicator during the growing season a lagged correlation analysis was
71
performed between 3-month SSI ending in August and EDDI at every time scale and
ending month.
Figure 4 shows that in all four regions EDDI leads SSI, where 3-month SSI ending in
August (blue dots in Figure 4 show fixed time scale and ending month for SSI) is better
correlated to 3-month EDDI ending in June (CA; Figure 4a) or July (TX, IA, and PA;
Figure 4b, 4c, and 4d respectively). An interesting feature of Figure 4 is shown for IA,
where 12-month EDDI ending in August was found to have highest correlation to 3-
month SSI ending in August, highlighting the extremely low summer SM moisture
variability in this region. This is further reinforced later in Figure 6, where monthly SSI
variability was found to be low relative to EDDI and SPI during the 2012 drought. These
results highlight that EDDI is a leading indicator when compared to SSI, and therefore
could be used to complement and perhaps improve the USDM since SM percentiles are
primary inputs for USDM objective blends.
72
Figure 4: Lagged correlation between 3-month SSI ending in August and EDDI for (a)
CA, (b) TX, (c) IA, and (d) PA. Y-axis indicates EDDI ending months and x-axis
indicate EDDI time scale. Green dots are placed in the ending month containing the
strongest correlation for each time scale, and blue dots are used as a reference to show
SSI time scale and ending month.
ESI correlations with EDDI
Seasonal temporal correlations between EDDI and ESI for CONUS are shown in
Figure 5. Only spring (April, May, and June) and summer (July, August, and September)
periods are evaluated due to limited availability of continuous monthly ESI data during
fall and winter. ESI data were frequently missing in snow-covered mountainous regions
of the west during spring and summer periods, and ESI pixels were masked (indicated by
white shading in Figure 5, as in the mountain ranges of western US) when less than 75%
73
of the monthly time series was available over the period of 2000 to 2013. Pixels with
spurious ESI data (ESI <-5 and >5) were also masked. One benefit of EDDI over ESI and
other remote sensing based metrics is that EDDI can be used during all seasons. This may
be particularly useful for high-elevation hydrometeorological monitoring in seasonally
snow-covered areas.
Figure 5 illustrates fairly large differences between spring and summer periods, with
negligible differences between different time scales of 4-, 8-, and 12-weeks. During the
spring period negative correlations are strongest (r values < -0.7) over much of TX, the
desert SW, and central valley of CA, while weaker relationships were found over the NE,
and parts of the Pacific NW (Figure 5a, 5c, and 5e). The low positive correlations in the
NE are due to energy-limited evaporative conditions described in section 3.1. Summer
correlations (Figure 5b, 5d, and 5f) are strongest and spatial patterns most consistent over
the central US, and lower correlations are evident over parts of NV, CA and into the
Pacific Northwest when compared to the spring period. Inspection of the summer time
series from the regions of low correlation in the west and Pacific Northwest showed that
during certain summers ESI and EDDI were strongly negatively correlated, but positively
correlated in others (not shown). ET rates in semi-arid regions are typically low during
summer periods; therefore small variations in ET can potentially lead to large changes in
ESI, making for poor correlations with EDDI. For example, most of NV experienced
below normal Prcp and high temperatures for July of 2005, and EDDI and SPI indicated
drought conditions, whereas ESI indicated wet conditions (not shown). In general, EDDI
is strongly correlated to ESI (r values < -0.7) during spring and summer months over
much of the southwest, southcentral, and northcentral US.
74
Figure 5: Seasonal correlation coefficient (left column spring and right column summer)
between ESI and EDDI at (a, b) 4-week, (c, d) 8-week, and (e, f) 12-week time scales.
Areas shaded in white indicate an insufficient amount of ESI data.
Flash drought over the central US
Flash drought can develop even during periods of excess Prcp, and evaporative
drivers can potentially uniquely identify the onset and evolution of flash drought. For
example, in some situations (i.e., the 2011 central CONUS case), a T-based E0 would fail
to identify rapid drying due to below normal Tair coincident with high U2 and low q. The
following highlights the Midwest droughts of 2011 and 2012 as a case study to
75
demonstrate how EDDI can serve as an effective early warning of flash droughts, as well
as extended droughts.
Area-averaged time series of 1-month EDDI are compared to 1-month SPI and SSI
during 2011 and 2012 in Figure 6a for the IA domain. Figure 6b illustrates the sensitivity
of EDDI to individual ET0 forcings averaged over the IA domain. Note that in Figures 6a
and 6b the vertical axis of EDDI is reversed to better visualize drought onset and duration
when compared to SPI and SSI. Figure 6a illustrates that in April, 2011, all indices are
near neutral (i.e., close to zero), and over the next two months EDDI changes to a
moderate drought class (<-0.78 or USDM D1 class), while both SPI and SSI increase to
slightly wet conditions. SPI and SSI values do not decrease towards moderate drought
conditions until July of 2011. SPI falls below moderate drought in September, and SSI
follows one month later in October. Both EDDI and SSI maintain extended drought
conditions throughout all of 2012, with the exception of February when EDDI is slightly
above moderate drought (-0.78), but still below zero. During this extended drought of
2012, SPI is highly variable and indicates wet conditions for many months.
To highlight the ET0 drivers that caused EDDI to signal first a flash drought and then
an extended drought, a simple sensitivity analysis of EDDI was performed (Figure 6b and
6c). For this analysis, ET0 was calculated while constraining the variable of interest to
daily climatology values in order to isolate the impact of each forcing on the EDDI
drought signal. Results are presented as estimates of EDDI with a notation of the variable
constrained to its daily climatology (i.e., EDDI-T, EDDI-q, EDDI-Rd, and EDDI-U2). For
example, EDDI-T was calculated using the daily climatology of Tmax and Tmin, and with
METDATA-observed forcings values of all other variables. During the period of 20 May
76
to 25 May, EDDI-q and EDDI-U2 had the greatest separation from standard EDDI values
in the negative direction (note y-axis is reversed), which indicates that the drying power
of the air term in the ET0 equation, (U2 multiplied by vapor pressure deficit), initiated the
flash drought signal—approximately 20 May through 5 June—in EDDI via increased U2
and below normal q (Figure 6c). In this case, using daily climatology q and U2 values
mitigated the drought signal relative to the standard EDDI. By June, 2011, EDDI
decreased below the moderate drought threshold (-0.78), with the primary difference
from May being that U2 and Tair were then acting in combination to exacerbate the
drought signal—as opposed to Tair moderating it in May. Despite below-normal Tair
conditions in September, 2011, the standard EDDI drought signal was maintained due to
extremely low q values evidenced by a large difference between EDDI and EDDI-q
(absolute difference of 1.17). From November, 2011, through the following May, Tair
dominated the EDDI signal, as seen by the large differences between EDDI and EDDI-T.
This increase in Tair and ET0 likely contributed to the persistent SSI drought signal
throughout 2012, despite above-normal Prcp for February, April, October, and December
(Figure 6a).
Results illustrated in Figure 6 and in the companion paper of Hobbins et al.
(2015) highlight two major focal points of this research: (1) EDDI is a leading indicator
of flash and extended drought conditions, and (2) a physically based E0 is required to
capture this signal. This reinforces the work of Hobbins et al. (2012) and Hobbins (2015)
who concluded that Tair is not always the dominant driver of ET0, and T-based
parameterizations could lead to false drying (or wetting) signals when used for drought
monitoring applications. Our findings illustrated in Figure 6 also contradict the notion
77
that 2012 should be considered a flash drought case over the central US (e.g. Mo and
Lettenmaier 2015): our results clearly indicate a well-established and persistent drought
signal by both EDDI and SSI, with SPI being the only indicator to signal a rapid
transition from wet to dry over the period of April through July. Figure 6 illustrates that
the flash drought signal appeared in EDDI starting in May, 2011, and in SPI and SSI
starting in August, 2011.
Figure 6: EDDI under sustained and flash drought conditions. (a) Monthly time series of
1-month EDDI, SSI, and SPI area averaged over the IA domain. (b) Monthly time series
of 1-month EDDI and EDDI constrained by climatology Tair (EDDI-T), q (EDDI-q), Rd
(EDDI-Rd), and U2 (EDDI- U2). Black box highlights time period shown in (c). (c) Daily
time series of 1-month EDDI, EDDI-T, EDDI-q, EDDI-Rd and EDDI-U2 for May and
June 2011 shown to highlight details of flash drought initiation. Note that the vertical axis
of EDDI is reversed to clearly visualize drought onset and duration when compared to
78
SPI and SSI. Light green reference line indicates start of moderate drought classification
(-0.78).
To spatially assess EDDI during the extended 2012 drought a comparison was made
between the USDM, SPI, SSI, and ESI. Recall from Section 2.5 that the USDM is at a
blended time-scale, against which a fixed time-scale EDDI is being compared: thus, the
EDDI and the USDM distributions should not be expected to look similar. The objective
of the EDDI and USDM comparisons is to show that EDDI can presage rapid onset
droughts before the impacts show up in the USDM, thus highlighting the substantial
added value gained by using EDDI in conjunction with other drought-monitoring metrics
for decision-making applications.
Figure 7 shows the evolution of the 1-month EDDI, ESI, SSI, and SPI, and
USDM through time and space over the spring and summer of 2012. The USDM
generally indicated no drought or D1-D2 over much of the central US of 1 May. This is
likely a result of the near-normal to slightly above normal Prcp during April, as
illustrated in the April SPI spatial distribution. In contrast, EDDI indicates at least
moderate drought conditions over most of the same region, and looks similar to the
USDM spatial distribution two months later (i.e., of 3 July, 2012). EDDI responded to
anomalously high Tair, U2, and Rd across the region during the second half of April. ESI
showed widespread neutral conditions for April with a rapid intensification in May. SSI
and SPI show a slower progression and more local intensification (non-uniform spatial
distribution) when compared to EDDI and ESI. The 2012 drought evolution illustrated by
the USDM over the central US expands in both spatial extent and severity throughout the
summer, however the progression from D0 to D3 and D4 takes approximately three
79
months. Figure 7 illustrates that 1-month EDDI presaged the onset of USDM extreme to
exceptional drought by as much as two months. ESI also led the onset of extreme to
exceptional drought, but was limited in extent when respectively compared to April
through July EDDI, and July USDM drought spatial distributions.
80
Figure 7: Evolution of the 1-month EDDI (top row), USDM (second row), 1-month ESI
(third row), 1-month SSI (fourth row), and 1-month SPI (fifth row) through spring and
summer of 2012. USDM data are from 1 May, 2012 (April column), 5 June, 2012 (May
column), 3 July, 2012 (June column), and 31 July, 2012 (July column). EDDI, ESI, SSI,
and SPI are at 1-month time scales at the end of each month: April, May, June, and July.
All drought metrics have been converted to USDM categories according to Table 1.
Extended drought in arid to semi-arid regions
In this section we examine whether EDDI can be used to characterize historical
extended droughts over the western US. Droughts in arid to semi-arid regions of the US
are generally slower to develop than in the central US, primarily due to the manner in
which water resources are both naturally and anthropogenically stored. Natural water
storage occurs as winter snowpack at high elevations that typically reach maximum depth
in March or April. During spring and summer snowmelt, runoff is stored in reservoirs.
Hydrologic and agricultural drought severity in the west are strongly linked to reservoir
storage and streamflow (McEvoy et al. 2012, Abatzoglou et al. 2014).
Two extended drought case studies using the USDM, EDDI, SPI, and SSI are shown
in Figure 8. The first case focuses on the drought of the 2007 water year (October 2006
through September 2007) (Figure 8, left column). The USDM from 02 October, 2007,
indicates 78% (percent area) of the western US in at least a D0 drought class. Figure 8c
illustrates the 12-month EDDI ending in September, 2007, and has the strongest spatial
coherence and severity when compared to the USDM, while SSI and SPI (Figure 8e and
8g, respectively) underrepresent the spatial extent shown by USDM and EDDI,
particularly over NV, ID, and western MT. The second case focuses on the extreme
southwestern drought of 2002 (Figure 8, right column), with the USDM mapped at 25
June, 2002, and the 6-month EDDI, SPI, and SPI mapped for January through June, 2002.
81
All metrics show a similar spatial structure of drought extent, although EDDI and SPI
indicate little to no drought in MT. Temperatures were lower than normal over much of
MT, WY, and the northern portion of UT and CO, and slightly above normal for the Four
Corners region (not shown). This indicates that Tair was likely driving EDDI negative in
MT, however Tair, q and U2 must have all played a role in driving EDDI in the positive
direction over UT and CO.
82
Figure 8: USDM from 02 October, 2007 (a) and 25 June, 2002 (b), 12-month (October-
September) EDDI (c), SSI (e), and SPI (g) ending September, 2007, and 6-month
(January-June) EDDI (d), SSI (f), and SPI (h) ending June, 2002.
83
The potential usefulness of EDDI to aid in the interpretation of hydroclimatic states at
multiple time scales and over long time periods was assessed for an area of interest.
Figure 9 illustrates time series of EDDI averaged over the northern Sierra Nevada for
1979-2013. The northern Sierra Nevada provides much of the water resources to western
NV and CA, therefore the use of multiple complementary drought metrics for evaluating
short and extended drought in this region is invaluable. EDDI at the 2-wk and 1-month
time scales (Figure 9a and 9b, respectively) closely correspond to documented heat
waves and extreme fire weather in the region (Burt 2007; Trouet et al. 2009), however
the high frequency of the time series make it difficult to characterize hydrologic drought.
At longer time scales EDDI (Figure 9c, 9d, and 9e, respectively) clearly identify all of the
major documented hydrologic droughts over the period from 1979 to 2013 (Seager 2007;
Weiss et al. 2009; McEvoy et al. 2012). The longest duration drought to occur during the
period of record analyzed was during the early 2000s, when the 12-month EDDI
remained positive for five continuous years (late 1999 to 2005). Fast recovery of
hydrologic droughts are also well captured by EDDI at nearly all time scales when
compared to known “drought-buster” precipitation events (Ralph and Dettinger 2010;
Dettinger 2013), and wet periods associated with El Niño (1982-83 and 1997-98), and La
Niña (2010-11).
84
Figure 9: Area-averaged time series of EDDI over the northern Sierra Nevada from 1979
to 2013 aggregated at 2-week (a), 1-month (b), 3-month (c), 6-month (d), and 12-month
time scales.
Summary and Conclusions
This work highlights an application and assessment of EDDI at multiple time
scales and for several hydroclimates as a companion study to Hobbins et al. (this issue).
The methods and results of Hobbins et al. (this issue) are reinforced and a robust
85
CONUS-wide evaluation is performed, by examining EDDI and individual evaporative
demand components as they relate to the dynamic evolution of flash drought over the
central US, characterization of hydrologic drought over the western US, and comparison
to commonly used drought metrics (USDM, SPI, SSI, and ESI). Results highlight the
advantages and limitations of EDDI as a monitor of drought at multiple time scales, and
provide leading indications of flash and extended hydrologic drought. Correlations of
EDDI to NLDAS-2 forced drought metrics of SSI and SPI indicate that over much of the
CONUS, EDDI spatial distributions are generally similar to SPI and SSI. Over parts of
the western US where weak correlations were found, EDDI often contained drought
information not found in SPI or SSI. For example, Prcp is bounded by zero at short time
scales (1 to 2 months) over many western states, which can lead to a skewed SPI,
whereas EDDI will maintain a consistent distribution during months with no Prcp. At
short time scales, spatial distributions and time series results illustrate that EDDI can be
useful for flash drought identification, and can serve as a leading indicator by as much as
two months in advance of the USDM, SPI, and SSI (i.e. Figures 4, 6, 7; and Figures
shown in Hobbins et al. - this issue).
Comparisons of EDDI to remotely sensed ESI products also show strong correlations,
with the exceptions of the northeast US during spring, and over parts of the western US
during summer. Weak correlations with ESI over the northeastern US are largely due to
energy-limited land-surface energy-balance conditions over the region, where ET and
ET0 are often positively correlated. Weak correlations with ESI over the western US
during summer months are likely due to the low and effectively zero-bounded actual ET
rates that occur in arid environments. Low soil moisture and low ET rates make it
86
difficult to accurately estimate ET with thermal and optical remote sensing. These
uncertainties combined with the high variability of estimated ET relative to average
conditions often led to spurious ESI values and low correlations with EDDI.
Comparisons of EDDI with ESI generally demonstrate that EDDI can be effectively used
in conjunction with ESI and other remote sensing products to provide year-round data,
with no limitations during cloudy days or over snow covered areas.
For drought monitoring in arid and semi-arid regions of western US, EDDI
aggregation to longer time scales (3 to 12 months) is best suited to capture the
complementary relationship found between ET and ET0 (Bouchet 1963; Hobbins et al.
2004), and therefore identify extended hydrologic droughts typical of this region. Results
illustrate that in most cases, when Prcp deficits at the 3- to 12-month time scales were
fairly large, EDDI was strongly positive. However, the complementary relationship was
found to not hold true in regions and time periods where weak land surface-atmospheric
coupling and energy limited conditions exist (Figures 3 and 5).
Despite some noted limitations, EDDI is shown to provide useful information on the
less-understood and documented dynamical processes associated with drought evolution
and persistence. Results highlighted in this work illustrate the benefits of assimilating
physically based E0 estimates and EDDI into operational monitoring products such as the
USDM. The additional information and early warning provided by EDDI could greatly
contribute to a stronger understanding of drought evolution and dynamics, land surface-
atmosphere interactions, and perhaps more importantly, reduce and/or mitigate future
adverse societal effects that have been associated with past droughts. EDDI could also
prove very useful and effective for easy-to-implement operational early warning for
87
agricultural and fire-weather monitoring (Ham et al. 2014) and seasonal forecasting of
drought.
Acknowledgments
This research was supported by the Desert Research Institute (DRI) Maki Endowment
for enhancing water resource monitoring in Southern, Nevada, U.S. Bureau of
Reclamation Climate Analysis Tools WaterSMART program, the National Integrated
Drought Information System (NIDIS) program, and U.S. Geological Survey and DRI
Great Basin Cooperative Ecosystem Study Unit collaborative project on drought
monitoring and fallow field tracking through cloud computing of Landsat, MODIS, and
gridded climate data archives.
References
Abatzoglou, J.T., 2011: Development of gridded surface meteorological data for
ecological
applications and modelling. Int. J. Climatol. 33, 121–131, doi:10.1002/joc.3413.
Abatzoglou, J.T., R. Barbero, J. Wolf, and Z. Holden, 2014: Tracking interannual
streamflow variability with drought indices in the U.S. Pacific Northwest. J.
Hydrometeor. 15, 1900–1912, doi:10.1175/JHM-D-13-0167.1.
Abramowitz, M., and I.A. Stegun, 1965: Handbook of mathematical functions, with
formulas, graphs, and mathematical tables. Dover Publications, 1046 pp.
88
Anderson, M.C., J.M. Norman, G.R. Diak, W.P. Kustas, and J.R. Mecikalski, 1997: A
two-source time-integrated model for estimating surface fluxes using thermal infrared
remote sensing. Remote Sens. Environ., 60, 195–216.
Anderson, M.C., J. M. Norman, J.R. Mecikalski, J.A. Otkin, and W.P. Kustas, 2007: A
climatological study of evapotranspiration and moisture stress across the continental
United States based on thermal remote sensing: 1. Model formulation, J. Geophys.
Res., 112, D10117, doi:10.1029/2006JD007506.
Anderson, M.C., J.M. Norman, J.R. Mecikalski, J.A. Otkin, and W.P. Kustas, 2007: A
climatological study of evapotranspiration and moisture stress across the continental
United States based on thermal remote sensing: 2. Surface moisture climatology, J.
Geophys. Res., 112, D11112, doi:10.1029/2006JD007507.
Anderson, M.C., C. Hain, B. Wardlow, A. Pimstein, J.R. Mecikalski, and W.P. Kustas,
2011: Evaluation of drought indices based on thermal remote sensing of
evapotranspiration over the continental United States. J. Climate, 24(8), 2025–2044,
doi:10.1175/2010JCLI3812.1.
Anderson, M.C., C. Hain, J. Otkin, X. Zhan, K. Mo, M. Svoboda, B. Wardlow, and A.
Pimstein, 2013: An intercomparison of drought indicators based on thermal remote
sensing and NLDAS-2 simulations with U.S. Drought Monitor classifications. J.
Hydrometeor., 14, 1035–1056, doi:10.1175/JHM-D-12-0140.1.
ASCE-EWRI, 2005: The ASCE Standardized Reference Evapotranspiration Equation.
Report 0-7844-0805-X, 59 pp. [Available online at
http://www.kimberly.uidaho.edu/water/asceewri/ascestzdetmain2005.pdf.]
89
Bouchet, R.J., 1963: Évapotranspiration réelle et potentielle, signification climatique.
Proc. International Association Scientific Hydrology Symp., Publ. No. 62, Berkeley,
CA, International Association Scientific Hydrology, 134–142.
Burt, C.C., 2007: Extreme weather: a guide and record book. W.W. Norton and Co., 320
pp.
Cattiaux, J., and P. Yiou, 2013: U.S. heat waves of spring and summer 2012 from the
flow-analogue perspective [in “Explaining Extreme Events of 2012 from a Climate
Perspective”]. Bull. Amer. Meteor. Soc., 94(9), S10–S13.
Dai, A., 2011: Characteristics and trends in various forms of the Palmer Drought Severity
Index during 1900-2008, J. Geophys. Res., 116, D12115,
doi:10.1029/2010JD015541.
Daly, C., R.P. Neilson, and D.L. Phillips, 1994: A statistical-topographic model for
mapping climatological precipitation over mountainous terrain. J. Appl. Meteorol.,
33(2), 140–158.
Dettinger, M.D., 2013. Atmospheric rivers as drought busters on the U.S. West Coast. J.
Hydrometeor., 14(6), 1721–1732, doi: 10.1175/JHM-D-13-02.1.
Farahmand, A., and A. AghaKouchak, 2015: A generalized framework for deriving
nonparametric standardized indicators. Adv. Water Resour., 76, 140–145,
doi:10.1016/j.advwatres.2014.11.012
Guttman, N.B., 1999: Accepting the standardized precipitation index: a calculation
algorithm. J. Amer. Water Resour. Assoc. 35(2), 311–322.
Ham, C., M.T. Hobbins, K.L. Abt, and J.P. Prestemon, 2014: Using the Evaporative
Demand Drought Index and the Palmer Drought Severity Index to forecast the
90
number of large wildland fires on federal lands, Large Wildland Fires Conference,
Missoula, MT, Association for Fire Ecology and the International Association of
Wildland Fire. [Available online at http://largefireconference.org/wp-
content/uploads/2013/06/Oral-Presentation-Abstracts-V4.pdf.]
Han, S., F. Tian, and H. Hu, 2014: Positive or negative correlation between actual and
potential evaporation? Evaluating using a nonlinear complementary relationship
model. Water Resources Research, 50(2), 1322–1336.
Hao, Z., and A. AghaKouchak, 2014: A nonparametric multivariate multi-index drought
monitoring framework, J. Hydrometeor., 15, 89–101, doi:10.1175/JHM-D-12-0160.1.
Hargreaves, G.H., and Z.A. Samani, 1985: Reference crop evapotranspiration from
temperature, Appl. Eng. Agric., 1, 96–99.
Heddinghaus, T.R., and P. Sabol, 1991: A review of the Palmer Drought Severity Index
and where do we go from here? Preprints, Seventh Conf. on Applied Climatology,
Boston, MA, Amer. Meteor. Soc., 242–246. [Available online at
http://www.ncdc.noaa.gov/temp-and-precip/drought/docs/heddinghaus-sabol-pmdi-
article.pdf.]
Hobbins, M.T., J.A. Ramírez, and T.C. Brown, 2004: Trends in pan evaporation and
actual evaporation across the conterminous U.S.: Paradoxical or complementary?
Geophys. Res. Lett., 31(13), L13503, doi: 10.1029/2004GL0198426.
Hobbins, M.T., A. Dai, M.L. Roderick, and G.D. Farquhar, 2008: Revisiting the
parameterization of potential evaporation as a driver of long-term water balance
trends. Geophys. Res. Lett., 35, L12403, doi: 10.1029/2008GL033840.
91
Hobbins, M.T., A.W. Wood, D. Streubel, and K. Werner, 2012: What drives the
variability of evaporative demand across the conterminous United States? J.
Hydrometeor., 13, 1195–1214, doi: 10.1175/JHM-D-11-0101.1.
Hobbins, M.T., A. Wood, D.J. McEvoy, J.L. Huntington, C. Morton, and J. Verdin, 2015:
The Evaporative Demand Drought Index: Part I - linking drought evolution to
variations in evaporative demand. J. Hydrometeor. (this issue).
Hobbins, M.T., 2015: The variability of ASCE Standardized Reference
Evapotranspiration: a rigorous, CONUS-wide decomposition and attribution. Trans.
ASABE (accepted).
Karl T.R., and Coauthors, 2012: U.S. temperature and drought: recent anomalies and
trends. Eos Trans. AGU, 93(47), 473–474.
Koster, R.D., and Coauthors, 2004: Regions of strong coupling between soil moisture and
precipitation. Science, 305(5687), 1138–1140, doi: 10.1126/science.1100217.
Koster, R.D., and Coauthors, 2006: GLACE: The Global Land–Atmosphere Coupling
Experiment. Part I: Overview. J. Hydrometeor., 7, 590–610, doi:10.1175/JHM510.1.
Koster, R.D., S.D. Schubert, and M.J. Suarez, 2009: Analyzing the concurrence of
meteorological droughts and warm periods, with implications for the determination of
evaporative regime. J. Climate, 22, 3331–3341, doi: 10.1175/2008JCLI2718.1.
Liang, X., D.P. Lettenmaier, E.F. Wood, and S.J. Burges, 1994: A simple hydrologically
based model of land surface water and energy fluxes for general circulation models.
J. Geophys. Res., 99(D7), 14,415–14,428.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer.
Meteor. Soc., 87(3), 343–360, 10.1175/BAMS-87-3-343
92
McEvoy, D.J., J.L. Huntington, J.T. Abatzoglou, and L.M. Edwards, 2012: An evaluation
of multiscalar drought indices in Nevada and eastern California. Earth Interact., 16,
1–18, doi:10.1175/2012EI000447.1.
McKee, T.B., N.J. Doesken, and J. Kleist, 1993: The relationship of drought frequency
and duration to time scales. Preprints, Eighth Conf. on Applied Climatology,
Anaheim, CA, Amer. Meteor. Soc., 179–184. [Available online at
http://ccc.atmos.colostate.edu/relationshipofdroughtfrequency.pdf.]
Milly, P.C.D., and K.A. Dunne, 2011: On the hydrologic adjustment of climate-model
projections: the potential pitfall of potential evapotranspiration. Earth Interact., 15,
doi:10.1175/2010EI363.1.
Mitchell, K.E., and Coauthors, 2004: The multi-institution North American Land Data
Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a
continental distributed hydrological modeling system, J. Geophys. Res., 109, D07S90,
doi:10.1029/2003JD003823.
Mo, K.C., and D.P. Lettenmaier, 2015: Heat wave flash droughts in decline. Geophys.
Res. Lett., 42, 2823–2829, doi:10.1002/2015GL064018.
Monteith, J.L., 1965: Evaporation and environment. Symp. Soc. Exp. Biol., XIX, 205–
234.
Mu, Q., M. Zhao, J.S. Kimball, N.G. McDowell, and S.W. Running, 2013: A remotely
sensed global terrestrial drought severity index. Bull. Amer. Meteor. Soc., 94(1), 83–
98, doi:10.1175/BAMS-D-11-00213.1.
Otkin, J.A., M.C. Anderson, C. Hain, I.E. Mladenova, J.B. Basara, M. Svoboda, 2013a:
Examining rapid onset drought development using the thermal infrared–based
93
Evaporative Stress Index. J. Hydrometeor., 14, 1057–1074. doi:10.1175/JHM-D-12-
0144.1.
Otkin, J.A., M.C. Anderson, C. Hain, and M. Svoboda, 2013b: Examining the
relationship between drought development and rapid changes in the Evaporative
Stress Index. J. Hydrometeor., 15, 938–956, doi:10.1175/JHM-D-13-0110.1.
Palmer, W.C., 1965: Meteorological drought. Research Paper 45, 58 pp. [Available
online at http://www.ncdc.noaa.gov/temp-and-precip/drought/docs/palmer.pdf.]
Penman, H.L., 1948: Natural evaporation from open water, bare soil and grass. Proc. R.
Soc. London, Ser. A, 193(1032), 120–146, doi:10.1098/rspa.1948.0037.
Priestley, C.H.B., and R.J. Taylor, 1972: On the assessment of surface heat flux and
evaporation using large-scale parameters. Mon. Weather Rev., 100(2), 81–92.
Quiring, S.M., 2009: Developing objective operational definitions for monitoring
drought. J. Appl. Meteor. Climatol. 48, 1217–1229, doi: 10.1175/2009JAMC2088.1.
Ralph, F.M., and M.D. Dettinger, 2012: Historical and national perspectives on extreme
West Coast precipitation associated with atmospheric rivers during December
2010. Bull. Amer. Meteor. Soc., 93, 783–790. doi:10.1175/BAMS-D-11-00188.1.
Seager, R., 2007: The turn of the century North American drought: global context,
dynamics, and past analogs. J. Climate, 20, 5527–5552,
doi:10.1175/2007JCLI1529.1.
Seager, R., M. Hoerling, S. Schubert, H. Wang, B. Lyon, A. Kumar, J. Nakamura, and N.
Henderson, 2014: Causes and predictability of the 2011-14 California drought.
Assessment Report, 40 pp. [Available online at
http://cpo.noaa.gov/MAPP/californiadroughtreport.]
94
Sheffield, J., E.F. Wood, and M.L. Roderick, 2012: Little change in global drought over
the past 60 years. Nature, 491, 435–438, 15 November, doi:10.1038/nature11575.
Svoboda, M., and Coauthors, 2002: The Drought Monitor. Bull. Amer. Meteor. Soc., 83,
1181–1190.
Thornthwaite, C.W., 1948: An approach toward a rational classification of climate.
Geogr. Rev., 38, 55–94.
Trouet, V., A.H. Taylor, A.M. Carleton, and C.N. Skinner, 2009: Interannual variations
in fire weather, fire extent, and synoptic-scale circulation patterns in northern
California and Oregon. Theor. Appl. Climatol., 95(3-4), 349–360,
doi:10.1007/s00704-008-0012-x.
van der Schrier, G., P.D. Jones, and K.R. Briffa, 2011: The sensitivity of the PDSI to the
Thornthwaite and Penman-Monteith parameterizations for potential
evapotranspiration. J. Geophys. Res., 116, doi:10.1029/2010JD015001.
Vicente-Serrano, S.M., J.I. López-Moreno, S. Beguería, J. Lorenzo-Lacruz, C. Azorin-
Molina, and E. Morán-Tejeda, 2012: Accurate computation of a streamflow drought
index. J. Hydrol. Eng., 17(2), 318–332, doi:10.1061/(ASCE)HE.1943-5584.0000433.
Weiss, J.L., C.L. Castro, and J.T. Overpeck, 2009: Distinguishing pronounced droughts
in the southwestern United States: Seasonality and effects of warmer temperatures. J.
Climate, 22, 5918–5932, doi: 10.1175/2009JCLI2905.1.
Wilks, D.S., 2011: Empirical distributions and exploratory data analysis. Statistical
methods in the atmospheric sciences. Academic Press, 23–70.
Xia, Y. and Coauthors, 2012a: Continental-scale water and energy flux analysis and
validation for the North American Land Data Assimilation System project phase 2
95
(NLDAS-2): 1. Intercomparison and application of model products. J. Geophys. Res.,
117(D3), doi:10.1029/2011JD016048.
Xia, Y. and Coauthors, 2012b: Continental-scale water and energy flux analysis and
validation for the North American Land Data Assimilation System project phase 2
(NLDAS-2): 2. Validation of model-simulated streamflow, J. Geophys. Res.,
117(D3), doi:10.1029/2011JD016051.
Yao, Y., S. Liang, Q. Qin, and K. Wang, 2010: Monitoring drought over the
conterminous United States using MODIS and NCEP Reanalysis-2 data. J. Appl.
Meteor. Climatol., 49, 1665–1680. doi:10.1175/2010JAMC2328.1.
96
Exploring the Use of Physically Based Evaporative Demand Anomalies to Improve
Seasonal Drought Forecasts
Daniel J. McEvoy1, 2
, John F. Mejia1, Justin L. Huntington
1, and Michael Hobbins
3, 4
1Desert Research Institute, Reno, Nevada
2Atmospheric Science Graduate Program, University of Nevada, Reno, Nevada
3University of Colorado-Cooperative Institute for Research in Environmental Sciences,
Boulder, CO 4NOAA-Earth Systems Research Laboratory, Physical Sciences Division, Boulder, CO
97
Abstract
Providing reliable seasonal drought forecasts continues to pose a major challenge
for scientists, end-users, and the water resources and agricultural communities. Recent
research has shown that satellite based evapotranspiration (ET) and land surface model
based evaporative demand (E0) anomalies accurately represent drought at different time
scales. Yet, to this end minimal research has been conducted on seasonal prediction skill
of E0 and application to drought forecasting. In this study, the first CONUS wide
evaluation of E0 forecast anomalies is performed using the National Center for
Environmental Prediction Climate Forecast System version 2 (CFSv2) reforecast data
covering the period of 1982-2009, and the American Society for Civil Engineers
Standardized Reference ET is used for E0. Skill evaluation is carried out using the
University of Idaho gridded meteorological data (METDATA), and E0 skill from CFSv2
is compared against precipitation (Prcp) skill. CONUS was divided into nine area
averaging climate regions, and moderate skill was found out to leads of five months in
the West, Southwest and South regions during the spring, and leads of one to three
months during the summer and fall in the Northeast, West North Central, East North
Central, and Central regions. Skill for E0 was consistently better than for Prcp with
improvements in anomaly correlation of 0.2 to 0.5.While probabilistic skill evaluation of
drought events revealed overall poor mean skill, the notable warm-season droughts of
1988 and 1999 in the West North Central, Central, and Northeast regions, and the
winter/spring drought of 1992 in the Northwest were all forecast with skill using E0
anomalies. Increased skill was found in the Northwest, West, Southwest, and Southeast
regions when CFSv2 forecast were initialized during moderate and strong El Niño-
98
Southern Oscillation events, and degraded skill was found in the East North Central,
Central, and Northeast regions.
Introduction
A growing literature indicates that current dynamical seasonal precipitation (Prcp)
forecasts contain limited skill past one-month lead time (e.g., Lavers et al. 2009, Yuan et
al. 2011, Yuan et al. 2013, Saha et al. 2014, Wood et al. 2015). Therefore, incorporating
new drought related variables with reasonable skill could add value and confidence to
operational seasonal forecasts. Our understanding of drought dynamics and variability
has evolved substantially over the last decade by using evapotranspiration (ET) and
physically based evaporative demand (E0) as a link between the land surface-atmosphere
interface. Recent studies have shown ET (primarily obtained through the use of remote
sensing) and E0 can be used to indicate drought by providing details on feedbacks at this
interface (Yao et al. 2010, Anderson et al. 2011, Mu et al. 2013, Otkin et al. 2013, Shukla
et al. 2015). However, remote sensing is limited to near real-time application and
dynamical forecasting is not possible.
On the other hand, variables needed to compute E0 (air temperature (Tair), wind
speed (WS), dowelling shortwave radiation at the surface (Rd), and specific humidity
(SH)) are all available from the Climate Forecast System version 2 (CFSv2; Saha et al.
2014). Studies on E0 forecasts from CFSv2 are limited to Tian et al. (2014), who
evaluated bias-corrected maximum and minimum Tair (Tmax and Tmin, respectively), WS,
Rd, and offline computations of E0 over the southeast CONUS. Dew point temperature
was approximated using Tmin, and therefore CFSv2 SH was not evaluated. Tian et al.
(2014) found CFSv2 E0 forecasts to have moderate skill during the cold season with the
99
greatest skill when forecasts were initialized during El Niño-Southern Oscillation
(ENSO) events (El Niño and La Niña conditions existed), and no skill during the warm
season due to the inability of CFSv2 to fully resolve summer convection. An extensive
analysis over CONUS of CFSv2 E0 and the application to seasonal drought forecasting
has yet to be conducted.
As an example of Prcp and E0 anomalies in drought, Figure 1 shows the E0
(Figure 1a) and Prcp (Figure 1b) anomaly percentiles for the AMJ 2002 period from the
University of Idaho’s gridded meteorological data (METDATA; Abatzoglou 2011),
during one of the most severe droughts in the recorded history of the Southwest (e.g.,
Weiss et al. 2009 and references therein). Both E0 and Prcp anomalies identify similar
spatial patterns of wet (MT and the Great Lakes) and dry regions (Southwest, NC, and
VA), which clearly shows that E0 can successfully identify drought periods. In Figure 2
the CONUS average percent area in drought based on percentiles of 3-month
accumulated E0 (Figure 2a) and Prcp (Figure 2b) are shown, with drought being defined
as E0 values above the 80th
percentile and Prcp values below the 20th
percentile. While
differences in intensity and timing certainly exist, both metrics consistently identify the
major drought periods of 1987-1989, 1999-2003, and 2006-2007. The wet periods of
1982-1984 (excluding summer of 1983) and 1991-1998 are also consistent. Much of
CONUS experienced well above normal temperatures during 2006 and 2007, which
likely drove the percent area of drought based on E0 much higher relative to Prcp.
In this study E0 anomalies from CFSv2 (CFSRF; Saha et al. 2014) reforecasts are
evaluated over CONUS against METDATA in order to determine if E0 anomalies contain
improved skill over Prcp, and can therefore be used to add confidence to seasonal
100
drought predictions. This study aims not only to evaluate CFSRF, but also to explore the
drivers of drought dynamics and variability in an effort to understand why certain
droughts are more predictable than others.
Figure 1: Accumulated E0 (a) and Prcp (b) anomaly percentiles from METDATA for
AMJ 2002. Note that upper E0 and lower Prcp percentiles indicate drought (brown
shading). NCDC climate regions (described in Section 2) used as area averaging domains
for Section 3 results are shown in the bottom panel (c). Regions are named as follows:
101
Northwest (NW), West (We), Southwest (SW), West North Central (WNC), South (So),
East North Central (ENC), Central (Ce), Southeast (SE), and Northeast (NE).
Figure 2: CONUS average percent area in drought based on 3-month accumulated E0 (a)
and Prcp (b) percentiles.
Data and Methodology
Nine-month continuous reforecasts (CFSRF) from CFSv2 were obtained from
NCEP, covering the retrospective period of 1982-2009. A detailed description of CFSRF
can be found in Saha et al. (2014), and several other papers have laid out the CFSRF
format (e.g., Yuan et al. 2011, Dirmeyer 2013). In this study, true monthly ensembles
102
were created from CFSRF leads of 1-9 months, resulting in a range of 20 to 28 ensemble
members per month (Table 1). (This is in contrast to the ensembles provided by NCEP,
which overlap initialization months. For example, the NCEP January ensemble contains
24 members, with initialization dates of January 11th
, 16th
, 21nd
, 26th
, and 31st, and
February 5th
.) The January ensemble used here contains 28 members initialized on
January 1st, 6
th, 11
th, 16
th, 21
nd, 26
th, and 31
st. This method is used to identify the impact
of initialization month on forecast skill.
Table 1: CFSv2 monthly ensembles are listed and each initial day consists of four
members initialized at 00Z, 06Z, 12Z, and 18Z.
Initial Month Number of Members Initial Days
January 28 1, 6, 11, 16, 21, 26, 31
February 20 5, 10, 15, 20, 25
March 24 2, 7, 12, 17, 22, 27
April 24 1, 6, 11, 16, 21, 26
May 28 1, 6, 11, 16, 21, 26, 31
June 24 5, 10, 15, 20, 25, 30
July 24 5, 10, 15, 20, 25, 30
August 24 4, 9, 14, 19, 24, 29
September 24 3, 8, 13, 18, 23, 28
October 24 3, 8, 13, 18, 23, 28
November 24 2, 7, 12, 17, 22, 27
December 24 2, 7, 12, 17, 22, 27
To evaluate CFSRF, daily METDATA covering the period of 1982-2009 were
obtained (http://metdata.northwestknowledge.net/) and averaged to monthly values.
METDATA is a bias-corrected and spatially disaggregated (from 12 km to 4 km) product
that combines the Parameter Regression on Independent Slopes Model (PRISM; Daly et
al. 1994) with the North American Land Data Assimilation System version 2 (NLDAS-2;
Mitchell et al. 2004). To match the CFSRF spatial resolution, METDATA were re-
103
gridded from 4 km to 1° using a bilinear interpolation, and negligible differences were
found between spatially averaged native and regridded data (Figure 3). Variables
obtained include Prcp, Tmax (at 2 m), Tmin (at 2 m), SH (at 2 m), WS (at 10 m), and Rd.
Figure 3. Comparison between 1982-2009 CONUS average annual E0 from the
METDATA native grid of 4-km (x-axis) and the regridded 1° spatial resolution (y-axis).
Following Allen et al. (1998, 2005), the American Society of Civil Engineers
Standardized Reference ET (ET0) was computed from METDATA and CFSRF, and is
used as the E0 in this study. A priori, it is generally assumed that if the necessary data
resources are available, a physically based E0 method should be used over a Tair- or
104
radiation- based method. Hobbins et al. (2012) and Hobbins (2015) demonstrated that
the dominant drivers of E0 variability change across CONUS depending on factors such
as aggregation period (monthly vs. annual) and season, and is not always Tair. For
example, during the summer WS is the dominant driver over much of the Great Basin,
while Rd dominates over much of the southeast CONUS. For ET0 computations at the
daily time step the soil heat flux (G) is assumed to zero. However, at monthly time steps
G must be calculated to account for warming and cooling of the soil in spring and
autumn, respectively. Following Allen et al. (1998; equations 43 and 44) mean monthly
Tair was used to calculate G. ET0 was first calculated for each ensemble member, and the
ensemble mean ET0 was then calculated.
A skill analysis was carried out over CONUS, and area averaging domains were
constructed using the nine National Climatic Data Center (NCDC) climate regions
(Figure 1c; Karl and Koss 1984). Monthly METDATA anomalies were computed relative
to the 1982-2009 METDATA mean. Following the recommendation of Kumar et al.
(2014), monthly CFSRF anomalies were calculated relative to the CFSRF climatology
(1982-2009) from the corresponding initialization month and lead time. Season 1
forecasts were generated as the accumulated anomaly over the first three months (i.e.,
season 1 JFM forecasts were initialized in December, and represents the accumulated
anomaly over JFM).
To assess CFSRF general skill, anomaly correlation (AC) between CFSRF
ensemble means and METDATA was computed for monthly one- to nine-month leads,
and season-1 forecasts, and is defined as follows:
105
𝐴𝐶 =(𝑓 − 𝑓𝑐)(𝑜 − 𝑜𝑐)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅
√(𝑓 − 𝑓𝑐)2̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ (𝑜 − 𝑜𝑐)2̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅
where f is the forecast, fc is the forecast climatology, o is the observed value, and oc is
the observed climatology (Miyakoda et al. 1972; equation 11). Sample size for each AC
was 28 (number of reforecast years).
Skill during individual drought events was assessed based on the probability
forecasts, using the categorical Heidke skill score (HSS; O’Lenic et al. 2008) computed
over individual climate regions, and is defined as
𝐻𝑆𝑆 =(ℎ − 𝑒) ∗ 100
(𝑡 − 𝑒)
where h is the number of grid points with hits, or correct tercile forecast, t is the total
number of grid points, and e is the number of grid points expected to be correct by chance
(i.e., t/3).
All forecasts were used for the HSS, including grid points with “equal chances”,
following Peng et al. (2013).
Results
Deterministic skill
Figure 4 shows the spatially averaged monthly AC over each NCDC climate
region for ET0. Large regional and seasonal variability in skill was found, which indicates
that using only the CONUS average AC would not provide much valuable insight.
Maximum average AC for each region ranged from 0.41 (Southeast) to 0.66 (West). The
West, Southwest, and South regions all show a similar pattern of greatest skill during
January through June, with moderate skill (AC = 0.3 to 0.6) out to 5-months lead time,
106
and a sharp decrease in skill during the months of July through December. The CPC
currently uses an AC value of 0.3 as the threshold for a skill mask (AC < 0.3 considered
to have no skill) in the real time CFSv2 forecasts. A similar pattern was revealed in the
West and Southwest Prcp skill (Figure 5), where enhanced predictability primarily is
associated with the initial conditions of the ENSO region sea surface temperatures (SSTs)
(e.g., Wood et al. 2005, Yuan et al. 2013). This lends credence to the idea that ET0
predictability in these regions may also be related to SST initial conditions from the
ENSO region in the late-summer and fall months, and is further investigated in
subsequent sections.
107
Figure 4: Average ET0 anomaly correlation between METDATA and CFSRF over each
region (refer to Figure 1c for full region names and locations). Labels on the x-axis
indicate lead time (months) and labels on the y-axis indicate target month.
Figure 5: Average precipitation anomaly correlation between METDATA and CFSRF
over each region (refer to Figure 1c in main manuscript for full region names and
locations). Labels on the x-axis indicate lead time (months) and labels on the y-axis
indicate target month.
Moderate skill at the longest lead times occurred in June for the West, and May
for the Southwest and South, which indicates December and January are important
initialization months for reliable forecasts in these regions. Consistent with Tian et al.
(2014), the Southeast ET0 skill was greatest during the cold season, with a major decrease
108
in skill during the warm season. An intriguing feature of Figure 4 is that moderate skill
was maintained during the important growing season months of July through September
at leads of 1-3 months over the East North Central, Central, and Northeast regions, where
Prcp skill is basically nonexistent (Figure 5c, 5d, 5e).
Patterns of Tmax and Tmin skill are a bit noisy with little regional or seasonal
consistency, and moderate skill was found to extend to slightly longer lead times when
compared to ET0 (Figures 6 and 7). Tair skill will be important for determining ET0 skill
in much of the interior central US, where Tair is found to be the dominant driver of ET0
variability year round (Hobbins 2015).
109
Figure 6: As in Figure 4, but for maximum temperature.
Figure 7: As in Figure 4, but for minimum temperature.
110
SH was found to have moderate skill out to lead times of five months, and showed
similar skill patterns to ET0 in several regions (Figure 8). Skillful SH predictions will be
important for parts of the Northeast (Figure 8e) and Southeast (Figure 8i) regions in the
months of NDJF, when SH drives much of the ET0 variability (Hobbins 2015).
Figure 8: As in Figure 4, but for specific humidity.
111
Generally lower skill was found in Rd predictions, but pockets of moderate skill
were still found in several regions out to lead times of two to four months (Figure 9). The
lack of skill in spring and summer is likely hindering the ET0 skill during these seasons in
the Southeast (Figure 9i), when Rd was found to drive much of the ET0 variability in this
region (Hobbins 2015).
Figure 9: As in Figure 4, but for downwelling shortwave radiation at the surface.
112
Wind speed forecasts were found to contain little usable skill (Figure 10), with the
exception of a few pockets with consistent AC of 0.3-0.4 in the Southwest (Figure 10g)
and South (Figure 10h) regions. Future improvements to WS predictions will be
important for improving ET0 skill in the Southwest and parts of the West during the
summer months, when WS controls much of the ET0 variability (Hobbins (2015).
113
Figure 10: As in Figure 4, but for wind speed.
There is an important consistency in E0 and Prcp with respect to temporal
variability: quantities of both variables can change dramatically at the daily and monthly
time scales in response to synoptic scale weather patterns and persistent atmospheric
blocking. This is in contrast to the soil moisture column, which is slower to respond to
atmospheric circulation patterns. Therefore, a comparison of regional E0 and Prcp skill
can provide pertinent information on where and when E0 forecasts could add value to
seasonal drought forecasts.
114
Season 1 AC for both E0 and Prcp averaged over CONUS and its constituent nine
climate regions is shown in Figure 11. CONUS wide, E0 skill is greater than Prcp during
all seasons, and remains above the 0.3 AC threshold of moderate skill for more than half
of the year. Most regions contain at least one or two seasons when E0 skill exceeds Prcp
skill by ~0.2 to 0.5. Of particular interest are regions where E0 skill is high compared to
Prcp skill during the growing season, such as the East North Central, Central, and
Northeast. In these regions, enhanced reliability in seasonal drought forecasts could
greatly benefit agricultural operations during the heart of the growing season, as well as
harvest operations during the late summer and early fall. A second area of interest is in
the Southwest region, where moderate E0 skill during the fall, winter, and spring could
improve water supply outlooks, which is particularly important in UT and CO.
115
Figure 11: Season 1 anomaly correlation area averaged over CONUS and individual
climate regions. The black reference line is anomaly correlation of 0.3, which indicates
the start of moderate skill.
Categorical skill of probability forecasts in drought events
Thus far, the skill analysis has considered all months and seasons using
deterministic forecasts (ensemble means). Probability forecasts (i.e., the likelihood of
occurrence of a specific event—such as above normal, near normal, or below normal
E0—calculated from all ensemble members) are considered next during drought events to
provide additional insight into the potential of using E0 anomaly forecasts to add
confidence in seasonal drought forecasts. Drought events were defined based on
116
METDATA when seasonal (three-month accumulation) anomalies indicate at least 50%
of the pixels in a region are above the 80th
percentile for E0 and below the 20th
percentile
for Prcp. This excludes events where E0 and Prcp show contrasting drought signals. The
HSS was then calculated based on the observed terciles (below normal, near normal, or
above normal) and highest chance probability forecast, where negative HSS indicates
worse skill than the reference forecast (climatology), HSS of zero indicates the same skill
as the reference forecast, and positive HSS indicates skill as percent improvement over
the reference forecast (Olenic et al. 2008; Peng et al. 2013).
The number of matching drought events ranged from 20 to 31 for each region,
and a scatter plot of E0 HSS (x-axis) and Prcp HSS (y-axis) is shown in Figure 12. Points
in the lower right quadrant (E0 HSS > 0 and Prcp HSS < 0) indicate skill from E0 only, in
the upper left quadrant (E0 HSS < 0 and Prcp HSS > 0) indicate skill from Prcp only, in
the upper right quadrant (E0 HSS > 0 and Prcp HSS > 0) indicate skill from both Prcp and
E0, and in the lower left quadrant (E0 HSS < 0 and Prcp HSS < 0) indicate no skill from
either metric. Overall, only two regions (Southwest and East North Central) were found
to have a positive (skillful) mean HSS using E0; in no regions was Prcp found to have
positive mean HSS. Despite mean HSS being poor, a number of notable events were
forecast well for several regions, with the lower right quadrant in Figure 12 containing
more than double the number of events compared to the upper left quadrant (15 and 38
total combined events in all regions for the upper left and lower right quadrants,
respectively). Certain events like the 1988 JJA drought were forecast well using both E0
and Prcp in the West North Central (E0 HSS = 100 and Prcp HSS = 78) and East North
Central (E0 HSS = 100 and Prcp HSS = 21), but positive skill was only found for E0 in
117
the Central region (E0 HSS = 51 and Prcp HSS = -5). The other prominent drought that
was forecast well using E0 but not Prcp was summer and fall of 1999 which had
devastating impacts on the Central and Northeast regions. In the Central region E0 HSS
was 3, 44, and 83, and Prcp HSS was -20, -25, and -3 for the consecutive 3-month
periods of JAS, ASO, and SON 1999. In the Northeast region E0 HSS was 64 and 95,
while Prcp HSS was -25 and 0 for the consecutive 3-month periods of MJJ and JJA 1999.
The 1992 winter and spring drought in the Northwest (JFM, FMA, and MAM) is another
example of consistent and positive HSS obtained from E0 forecasts.
Figure 12: The HSS for season-1 forecasts for cases when both E0 and Prcp indicate
drought (>80th
percentile for E0 and <20th
percentile for Prcp). Labels inside of each
118
panel indicate region and mean HSS. Red circles show notable drought events of JFM,
FMA, and MAM 1992 in the NW, JJA 1988 in the WNC, ENC, and Ce, JAS, ASO, and
SON 1999 in the Ce, and MJJ and JJA 1999 in the NE. These events are described in
further detail in the text.
ENSO as a source of predictability
Predictability of both Tair and Prcp over CONUS and E0 over a portion of the SE
region have been found to increase when CFSRF and NCEP global spectral model
(predecessor to CFS) forecasts are initialized during moderate and strong ENSO events
(Wood et al. 2005, Yuan et al. 2013, Tian et al. 2014). To investigate ENSO as a source
of E0 predictability over CONUS, all season 1 forecasts were compared against ENSO
conditional forecasts defined as when the Oceanic Nino Index (ONI) from the CPC
(http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml)
exceeds +/- 1° C. Note that the forecasts initialized during ENSO events were considered,
not the forecast target month. A total of 76 events were classified during the period of
record out of 336 possible seasons.
Figure 13 shows the difference in AC between ENSO conditional events and all
forecasts using E0 (Figure 13a) and Prcp (Figure 13b). Spatial patterns of forecast
improvement are quite similar for both E0 and Prcp over parts of the West, Southwest,
South and Southeast regions. Precipitation improvements are consistent with previous
research in the West (Wood et al. 2005, Yuan et al. 2013) and Southeast (Yuan et al.
2013) regions, and our results are also consistent with Yuan et al. (2013) for the
Southwest (this contradicts the results of Wood et al. (2005) for this region). Tian et al.
(2014) also found improvements in E0 forecasts in parts of the Southeast. Differences in
spatial patterns between E0 and Prcp arise in much of the northern half of CONUS, where
119
E0 forecasts are greatly improved in the Northwest and Prcp forecasts are improved in the
Northeast and Central regions. Overall, the magnitude of AC differences are similar, but
E0 skill is considerably higher than Prcp during ENSO conditional forecasts (Figure 13c
and 13d). This is particularly evident in the Northwest and Southwest, with similar skill
found in the West and Southeast. Mostly minor changes in skill were found in the South,
West North Central, and East North Central regions.
Figure 13: The difference in AC (ENSO conditional forecasts – All forecasts) and
regionally averaged AC for E0 (a and c) and Prcp (b and d) forecasts.
Discussion and Conclusions
120
We have provided the first CONUS wide study assessing seasonal predictions of
E0 anomalies using the CFSv2 reforecast (CFSRF) and METDATA observations. Nine
climate regions were used as area-averaging domains to assess regional skill variability.
Monthly ensemble mean forecasts were analyzed at one- to nine-month leads, and lead
one-month, season 1 ensemble mean forecasts were used to highlight seasonal skill
variability. Monthly E0 ACs revealed large regional and seasonal variability, with
moderate skill (AC of 0.3-0.6) found at one- to five-month leads but no consistent skill at
longer lead times.
Seasonal skill from E0 was found to be consistently greater than Prcp when
averaged over CONUS, and the same could be said for most regions where large AC
differences (0.2-0.5) were found depending on season. Assessment of probability
forecasts during drought using the HSS events revealed no skill on average, with the
exception of the Southwest and East North Central using E0. However, two of the most
severe warm-season droughts during the period of record (summer of 1988 and summer
and fall of 1999) and the winter and spring drought of 1992 in the Northwest were all
forecast with reasonable skill using E0 and mostly poor skill using Prcp. While it is still
unclear exactly what makes certain droughts predictable and others not, one prominent
feature of the 1988, 1992, and 1999 (Northeast only) events was high temperatures that
clearly exacerbated the E0 anomalies, and likely improved predictions over Prcp
forecasts. Another interesting finding is that a poor Prcp forecast would not necessarily
lead to a poor E0 forecast, which indicates a potential lack of land surface-atmospheric
coupling in CFSv2. These results indicate that including E0 anomalies in operational
seasonal drought forecasts could provide additional skill and boost overall confidence for
121
various applications such as water resource outlooks in regions that depend on seasonal
snow pack, and agricultural outlooks in the many important farming belts throughout
CONUS.
We have shown results that are consistent with previous research, and suggest that
some portion of E0 predictability comes from the initial state of tropical Pacific SSTs.
This is evident in the similar skill patterns found between E0 and Prcp in the West,
Southwest, and Southeast regions, where several studies have found enhanced Prcp and
Tair skill during strong ENSO events (e.g., Wood et al. 2005, Yuan et al. 2013), and in our
analysis of ENSO conditional vs. all events. Tian et al. (2014) also found enhanced E0
predictability when CFSv2 forecasts were initialized during ENSO events in the
Southeast cold season. Jia et al. (2015) found seasonal Prcp skill in a Geophysical Fluid
Dynamics Laboratory climate model to be mostly ENSO-related, but temperature skill to
be related to ENSO as well as changes in the external radiative forcings (i.e., a multi-
decadal warming signal in both summer and winter), which could be an additional factor
contributing to E0 skill from CFSv2. Initial state of the soil moisture column could also
be contributing to E0 predictability considering the ongoing feedbacks between the land
surface state and variables of Tmax, Tmin, and SH used to generate E0 estimates. Yuan et
al. (2013) found high skill in soil moisture forecasts at longer lead times and Yoon and
Leung (2015) found antecedent soil moisture to be as important as ENSO in seasonal
Prcp forecast skill over parts of CONUS.
This work also opens the door for continued research on improved skill in E0
seasonal forecasts via downscaling approaches (statistical or dynamical). Downscaling
was intentionally left out of this study to avoid masking CFSv2 deficiencies in E0 drivers
122
at the native grid scale. With the regional results that we have provided, it can be seen
that downscaling of E0 and most of the drivers (Tair, SH, and Rd), could be useful for local
applications other than drought forecasting, like irrigation water demands. However,
downscaling WS would likely not provide additional benefit given that poor skill was
found everywhere during all seasons. It may be useful to replace forecast WS with
climatological values, which has shown to improve skill in long-range weather forecasts
of E0 (Tian and Martinez 2012).
A multi-model ensemble, such as the NMME (North American Multi-Model
Ensemble; Kirtman et al. 2014), would certainly provide a more robust analysis then than
just considering CFSv2. However, at the time of writing, CFSv2 was the only seasonal
forecast model to provide publically available reforecasts of Rd, SH, and WS needed to
compute E0, while NMME Phase-I reforecasts only provide the Tair component needed
for E0. The NMME Phase-II data distribution is currently underway, which includes
output of all the required variables for E0 computations from seven seasonal forecast
models. Once NMME Phase-II data distribution is complete, a follow-up study will
establish whether E0 forecast skill can be enhanced using a multi-model ensemble
approach.
Skillful seasonal predictions remain a major challenge in drought forecasting, and
the use of E0 anomalies from CFSv2 clearly show potential improvements depending on
region and season. However, further research and improvements are needed for practical
use in an operational setting, and E0 and its drivers should continue to be evaluated in
other dynamical models as well as statistical techniques. Identifying systematic
consistencies in the physical mechanisms driving drought events where E0 shows good
123
skill and Prcp shows poor skill will also be a crucial step towards implementation of E0 in
a drought forecasting framework.
Acknowledgements
This research was supported by the Desert Research Institute (DRI) IBM PureSystems,
the Bureau of Reclamation Climate Analysis Tools WaterSMART program, and the
National Integrated Drought Information System (NIDIS).
References
Abatzoglou, J. T., 2011: Development of gridded surface meteorological data for
ecological
applications and modeling. Int. J. Climatol. 33, 121–131, DOI: 10.1002/joc.3413.
Anderson, M. C., C. R. Hain, B. Wardlow, J. R. Mecikalski, and W. P. Kustas, 2011:
Evaluation of a drought index based on thermal remote sensing of evapotranspiration
over the continental United States. J. Climate, 24, 2025–2044.
Hobbins, M. T., A. Wood, D. Streubel, and K. Werner, 2012: What drives the variability
of evaporative demand across the conterminous United States? J. Hydrometeor, 13,
1195–1214, doi: 10.1175/JHM-D-11-0101.1.
Karl T. R., and W. J. Koss, 1984: Regional and national monthly, seasonal, and annual
temperature weighted by area, 1895-1983. Historical Climatology Series 4-3,
National Climatic Data Center, Asheville, NC, 38 pp.
——, Gleason B. E., Menne M. J., 2012: US temperature and drought: recent anomalies
and trends. E0s Trans Am Geophys Union, 93, 473–474.
Kirtman, B. P., D. Min, J. M. Infanti, J. L. Kinter, III, D. A. Paolino, Q. Zhang, H. van
den Dool, S. Saha, M. P. Mendez, E. Becker, P. Peng, P. Tripp, J. Huang, D. G.
124
DeWitt, M. K. Tippett, A. G. Barnston, S. Li, A. Rosati, S. D. Schubert, M.
Rienecker, M. Suarez, Z. E. Li, J. Marshak, Y. Lim, J. Tribbia, K. Pegion, W. J. M.,
B. Denis, and E. F. Wood, 2014: The North American Multimodel Ensemble: Phase-
1 Seasonal-to-Interannual Prediction; Phase-2 toward Developing Intraseasonal
Prediction. Bull. Amer. Meteor. Soc., 95, 585–601.
doi: http://dx.doi.org/10.1175/BAMS-D-12-00050.1
Kumar, S., P. A. Dirmeyer, and J. L. Kinter III, 2014: Usefulness of ensemble forecasts
from NCEP Climate Forecast System in sub-seasonal to intra-annual forecasting,
Geophys. Res. Lett., 41, 3586–3593, doi:10.1002/2014GL059586.
Miyakoda, K., G. D., Hembree, R. F. Strickler, and I. Shulman, 1972: Cumulative results
of extended forecast experiments I. Model performance for winter cases. Mon. Wea.
Rev.100. 836-855.
Mu, Q., M. Zhao, J. S. Kimball, N. G. McDowell, S. W. Running, 2013: A remotely
sensed global terrestrial drought severity index. Bull. Amer. Meteor. Soc., 94, 93-98.
Otkin, J. A., M. C. Anderson, C. Hain, I. E. Mladenova, J. B. Basara, M. Svoboda, 2013:
Examining Rapid Onset Drought Development Using the Thermal Infrared–Based
Evaporative Stress Index. J. Hydrometeor., 14, 1057–1074.
Saha S., and coauthors, 2014: The NCEP Climate Forecast System Version 2. J. Climate,
27, 2185-2208. doi: http://dx.doi.org/10.1175/JCLI-D-12-00823.1.
Shukla, S., M. Safeeq, A. AghaKouchak, K. Guan, and C. Funk, 2015: Temperature
impacts on the water year 2014 drought in California. Geophys. Res. Lett., 42,
doi:10.1002/2015GL063666.
125
Tian, D., and C. J. Martinez, 2012: Forecasting reference evapotranspiration using
retrospective forecast analogs in the southeastern United States. J. Hydrometeor., 13,
1874–1892,
doi:10.1175/JHM-D-12-037.1.
Tian, D., C. J. Martinez, and W. D. Graham, 2014: Seasonal prediction of regional
reference evapotranspiration based on Climate Forecast System Version 2. J.
Hydrometeor., 15, 1166-1188. doi: http://dx.doi.org/10.1175/JHM-D-13-087.1.
Yao, Y., S. Liang, Q. Qin, K. Wang, 2010: Monitoring Drought over the Conterminous
United States Using MODIS and NCEP Reanalysis-2 Data. J. Appl. Meteor.
Climatol., 49, 1665–1680. doi: http://dx.doi.org/10.1175/2010JAMC2328.1
Yoon, J., and L. R. Leung, 2015: Assessing the relative influence of surface soil moisture
and ENSO SST on precipitation predictability over the contiguous United States.
Geophys. Res. Lett., doi: 10.1002/2015GL064139 (in press).
Yuan, X., E. F. Wood, L. Luo, and M. Pan, 2011: A first look at Climate Forecast System
version 2 (CFSv2) for hydrological seasonal prediction. Geophys. Res. Lett., 38,
doi:10.1029/2011GL047792.
——, E. F. Wood, J. K. Roundy, and M. Pan, 2013: CFSv2-based seasonal hydroclimatic
forecasts over the conterminous United States. J. Hydrometeor, 26, 4828-4847.
Weiss, J. L., C. L. Castro, and J. T. Overpeck, 2009: Distinguished pronounced droughts
in the Southwestern United States: seasonality and effects of warmer temperature. J.
Climate, 22, 5918-5932.
Wood, A. W., A. Kumar, and D. P. Lettenmaier, 2005: A retrospective assessment of
National Centers for Environmental Prediction climate model–based ensemble
126
hydrologic forecasting in the western United States. J. Geophys. Res., 110, D04105,
doi:10.1029/2004JD004508.
Wood, E., S. Schubert, A. Wood, C. Peters-Lidard, K. Mo, A. Mariotti, and R. Pulwarty,
2015: Prospects for advancing drought understanding, monitoring and prediction. J.
Hydrometeor. doi:10.1175/JHM-D-14-0164.1, in press.
Xia, Y. and coauthors, 2012a: Continental-scale water and energy flux analysis and
validation for the North American Land Data Assimilation System project phase 2
(NLDAS-2): 1. Intercomparison and application of model products. J. Geophys Res.,
117 (D3), DOI: 10.1029/2011JD016048.
Xia, Y. and coauthors, 2012b: Continental-scale water and energy flux analysis and
validation for the North American Land Data Assimilation System project phase 2
(NLDAS-2): 2. Validation of model-simulated streamflow, J. Geophys Res., 117
(D3), DOI: 10.1029/2011JD016051.
127
Summary and Conclusions
Droughts are projected by global climate models to become more severe and
longer-lasting over portions of the US by the end of the 21st century, with drying
exacerbated by higher temperatures and increased evaporative demand (E0). Physically
based E0 has traditionally been neglected in drought monitoring, and there is a growing
need for the development of drought monitoring and forecasting methods and tools that
consider the effect of E0 and the drivers of E0 (temperature, humidity, wind speed, and
solar radiation) on drought development, intensification, and persistence.
Gridded data products (GDPs) are commonly used to estimate E0 without
thorough evaluation and understanding of biases and uncertainties in model
parameterizations. Several GDPs (PRISM, Daymet, and METDATA) were evaluated in
the Great Basin using a new observing network (independent of GDPs) in the Great
Basin—the Nevada Climate-Ecohydrological Assessment Network (NevCAN)—to
investigate the impact of terrain and GDP spatial resolution on estimates of precipitation,
temperature, and humidity.
Providing reliable “ground truth” for evaluating GDP precipitation estimates in
complex terrain was found to be challenging, with two co-located (~50 m apart) weather
stations (one dependent and one independent of GDPs) indicating about 30% difference
in water year precipitation totals. Nonetheless, GPDs were able to reproduce elevational
precipitation gradients, with a high bias (relative to NevCAN) in the cold season.
Temperatures were generally well replicated by GDPs with the exception of minimum
temperature (Tmin) at the alluvial fan locations. Nocturnal cold air drainage in complex
128
terrain results in large Tmin inversions at the alluvial fans (mean monthly lapse rates of >
+15°C km-1
were observed), which was captured by all GDPs (with the exception of
Daymet), but severely underestimated. This highlights the importance of using a non-
monotonic regression function, as in PRISM and METDATA, to estimate temperature.
Cold air pools and temperature inversions in complex terrain can lead to stagnant air and
long periods of poor air quality, and therefore spatially replicating this feature is of
utmost importance.
EDDI is the first drought indicator based solely on physically based E0 (the ASCE
Standardized Reference ET; ET0), and it was developed to fill a gap in monitoring of
drought dynamics driven by the aerodynamic (temperature, wind speed, and humidity)
and radiative (solar radiation and temperature) components of ET0. The Standardized
Precipitation Index (SPI), Standardized Soil Moisture Index (SSI), Evaporative Stress
Index (ESI), and United States Drought Monitor (USDM) were used to evaluate EDDI
across CONUS. The hypothesis was tested that EDDI can be a leading indicator during
rapid onset, or flash drought, due to both advective and radiative meteorological forcings
leading surface moisture depletion, and thus leading to a drought signal from EDDI prior
to other drought metrics. A sensitivity analysis on the drivers of EDDI during a flash
drought that transitioned to an extended drought revealed that it was high wind speed and
low humidity that initiated the flash drought, and extreme temperatures acted to
exacerbate the severity of EDDI and maintain an extended drought signal even during
periods of normal to slightly above normal precipitation. This reinforces the idea that a
temperature is not always the dominant driver of ET0 variability, especially at short time
129
scales, and temperature-based E0 cannot be used to capture realistic dynamics of flash
drought development.
Providing reliable seasonal drought forecasts remains a major challenge, and the
results from Chapter 2 motivated a study on the use of ET0 anomalies as a tool to
improve and add confidence to seasonal drought forecasts. Seasonal precipitation
forecasts hold little skill past the range of weather (~two weeks), with temperature
showing improved skill out to several months. Therefore, ET0 forecasts, with temperature
as an input, should be more skillful than precipitation. However, little is known about the
skill of seasonal humidity, solar radiation, and wind speed forecasts; the other drivers of
ET0 that can dominate the variability depending on season and region. In Chapter 3
reforecasts of ET0 and precipitation anomalies from the Climate Forecast System Version
2 (CFSv2) were evaluated against METDATA for the period of 1982-2009 over CONUS.
Forecasts of the individual drivers of ET0 were also evaluated. While overall probabilistic
skill of drought events was rather low, the severe warm-season droughts of 1988 and
1999 in the central and northeast US and the spring drought of 1992 in the Pacific
Northwest were all forecast with moderate skill using ET0.
The results presented in this dissertation can be useful to a number of groups both
in the scientific community and general public, and the conclusions and
recommendations are as follows:
Finer GDP spatial resolution does not always lead to less error when compared to
observations, and GDP performance varied greatly depending on region and
130
climate variable. It is strongly recommended that local analysis be conducted
prior to GDP use in hydro-climatic applications.
Using an over-simplified method to estimate dew point temperature (Tdew) in
Daymet—i.e., that daily average Tdew can be estimated as daily Tmin—is
unacceptable in semiarid to arid regions; this finding is particularly important
when considering GDPs for physically based E0 estimates.
METDATA is recommended for ET0 and EDDI calculations, with more realistic
humidity estimates compared to Daymet.
The hypothesis is confirmed that using a short time scale EDDI (two weeks to one
month) can be useful as an early warning indicator during flash drought
conditions, and EDDI was found to lead other indicators, such as SPI, SSI, and
the USDM, by up to three months.
EDDI could greatly benefit the agricultural, water resources, public health, and
recreation sectors through reactive emergency responses and implementation of
action plans in a timely manner.
EDDI can be useful to monitor hydrologic drought at longer time scales (6-12
months) in water-limited regions, which can be explained by the complementary
relationship between ET and ET0.
Seasonal forecasts of ET0 consistently contained greater skill than precipitation
relative to METDATA observations, with the greatest improvement found in parts
of the central and northeast US during the growing season, when precipitation
skill is nearly nonexistent.
131
Seasonal forecasts of ET0 anomalies may be particularly useful for increasing
confidence in season one outlooks during severe droughts driven by high
temperatures.
ET0 seasonal forecast skill is greatly enhanced in the northwest, west, southwest,
and southeast CONUS when forecasts are initialized during moderate and strong
El Niño-Southern Oscillation events.
Currently, EDDI is being updated in near real time (2-4 day latency), and continued
efforts should incorporate EDDI in operational drought-monitoring activities such as the
USDM. Seasonal forecasts of ET0 anomalies may be most beneficial to agricultural
sectors in the central and northeast US. Development of real-time seasonal ET0 forecasts
using CFSv2 is currently in progress, and will be made publically available in the near
future.