Post on 01-Jan-2016
Coupling of Atmospheric and Hydrologic Models:
A Hydrologic Modeler’s Perspective
George H. Leavesley1, Lauren E. Hay1, Martyn P. Clark2, William J. Gutowski, Jr.3, and
Robert. L. Wilby4
1U.S. Geological Survey, Denver, CO 2University of Colorado, Boulder, CO 3Iowa State University, Ames, IA 4King’s College London, London, UK
Topics
• Water resources issues
• Hydrologic modeling approaches
• Spatial and temporal distribution issues
• Hydrologic forecasting methodologies
• Downscaling approaches and applications
Water Resources Simulation and Forecast Needs
• Long-term Policy and Planning (10’s of years)
• Annual to Inter-annual Operational Planning (6 - 24 months)
• Short-term Operational Planning (1 - 30 days)• Flash flood forecasting (hours)• Land-use change and climate variability
PRMS
PRMS Snowpack Energy-Balance Components
LUMPED MODELSLUMPED MODELS - No account of spatial variability of processes, input, boundary conditions, and system geometry
DISTRIBUTED MODELSDISTRIBUTED MODELS - Explicit account of spatial variability of processes, input, boundary conditions, and watershed characteristics
QUASI-DISTRIBUTED MODELSQUASI-DISTRIBUTED MODELS - Attempt to account for spatial variability, but use some degree of lumping in one or more of the modeled characteristics.
SPATIAL CONSIDERATIONS
MAXIMUM TEMPERATURE-ELEVATION RELATIONS
PRECIPITATION-ELEVATION RELATIONS
Precipitation and Temperature Distribution
Methodologies• Elevation adjustments
• Thiessan polygons
• Inverse distance weighting
• Geostatistical techniques
• XYZ method
• …
Monthly Multiple Linear Regression (MLR) equations
developed for PRCP, TMAX, and TMIN using the predictor
variables of station location (X, Y) and elevation (Z).
XYZ Methodology
XYZ DISTRIBUTION
San Juan Basin Observation Stations 37
XYZ Spatial Redistribution
of Precipitation & Temperature
1. Develop Multiple Linear Regression (MLR) equations (in XYZ) for PRCP, TMAX, and TMIN by month using all appropriate regional observation stations.
2. Daily mean PRCP, TMAX, and TMIN computed for a subset of stations (3) determined by the Exhaustive Search analysis to be best stations
3. Daily station means from (2) used with monthly MLR xyz relations to estimate daily PRCP, TMAX, and TMIN on each HRU according to the XYZ of each HRU
Precipitation and temperature stations
XYZ Spatial Redistribution
Z
PR
CP
2. PRCPmru = slope*Zmru + intercept
where PRCPmru is PRCP for your modeling response unit
Zmru is mean elevation of your modeling response unit
x
One predictor (Z) example for distributing daily PRCP from a set of stations:
1. For each day solve for y-intercept
intercept = PRCPsta - slope*Zsta
where PRCPsta is mean station PRCP and
Zsta is mean station elevation
slope is monthly value from MLRs Plot mean station elevation (Z)
vs. mean station PRCP
Slope from monthly MLR used to find the
y-intercept
XYZ Methodology
XYZ DISTRIBUTIONEXHAUSTIVE SEARCH ANALYSIS
• Select best station subset from all stations
• Estimate gauge undercatch error for snow events (Bias in observed data)
• Select precipitation frequency station set (Bias in observed data)
Forecast Methodologies
- Historic data as analog for the future
Ensemble Streamflow Prediction (ESP)
-Synthetic time-series
Weather Generator
- Atmospheric model output
Dynamical Downscaling
Statistical Downscaling
Animas River @ Animas River @ DurangoDurangoMeasureMeasure
ddSimulatedSimulated
Animas Animas Basin Basin Snow-Snow-
covered covered Area Area Year 2000Year 2000SimulateSimulate
dd
MeasurMeasured ed
(MODI(MODIS S
SatellitSatellite)e)
Error Range <= Error Range <= 0.10.1
ESP Animas River @ Durango
0
2000
4000
6000
8000
10000
120004/
3/20
05
4/17
/200
5
5/1/
2005
5/15
/200
5
5/29
/200
5
6/12
/200
5
6/26
/200
5
7/10
/200
5
7/24
/200
5
8/7/
2005
8/21
/200
5
9/4/
2005
9/18
/200
5
Str
eam
flo
w (
cfsd
)
1981 (.68)
1982 (.45)
1983 (.23)
1986 (.50)
1987 (.95)
1988 (.73)1989 (.99)
1990 (.09)
1991 (.91)
1992 (.86)
1993 (.64)
1994 (.59)
1995 (.14)
1996 (.05)
1997 (.41)
1998 (.55)
1999 (.36)2000 (.82)
2001 (.32)
2002 (.77)
2003 (.18)
2004 (.27)
Probability of Exceedance
ESP – Animas River @ DurangoESP – Animas River @ Durango(Frequency Analysis on Peak Flows)
ESP – Animas River @ DurangoESP – Animas River @ DurangoESP Animas River @ Durango
0
2000
4000
6000
8000
10000
120004/
3/20
05
4/17
/200
5
5/1/
2005
5/15
/200
5
5/29
/200
5
6/12
/200
5
6/26
/200
5
7/10
/200
5
7/24
/200
5
8/7/
2005
8/21
/200
5
9/4/
2005
9/18
/200
5
Str
eam
flo
w (
cfsd
)
1981
1982
1983
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
Forecast 4/2/05 Observed 4/3 – 6/30/05
Representative Elevation of
Atmospheric ModelOutput based on Regional Station
Observations
Elevation-based Bias Correction
Performance Measures
Coefficient of Efficiency E
Nash and Sutcliffe, 1970, J. of Hydrology
Widely used in hydrology Range – infinity to +1.0 Overly sensitive to extreme values
Nash-Sutcliff Coefficient of Efficiency Scores Simulated vs Observed Daily
StreamflowAnimas River, Colorado USA
Statistical Statistical vs vs
Dynamical Dynamical DDownscaling
NNational CCenters for EEnvironmental
PPrediction/National Center for Atmospheric Research
Reanalysis NCEPNCEP
Global-scale model
NCEP
210 km grid spacingRetroactive 51 year recordEvery 5 days there is an 8-day forecast
Compare SDS and DDS output by using it to drive the distributed hydrologic model
PRMS in 4 basins
(DAY 0)
East Fork of the Carson
Cle Elum
Animas
Alapaha
Snowmelt Dominated922 km2
Snowmelt Dominated526 km2
Snowmelt Dominated1792 km2
Rainfall Dominated3626 km2
Study Basins
Statistical Downscaling
Carson
Cle Elum
Animas
Alapaha
NCEP Grid nodes
Basins
500km Buffer radius
Dynamical Downscaling
52 km grid node spacing
10 year run
Regional Climate Model – RegCM2
nested within NCEP
Dynamical Downscaling
East Fork of
the Carson
Cle Elum
Animas
Alapaha
Animas River Basin
RegCM2 grid nodes
Buffer
52 km
Dynamical Downscaling
Use grid-nodes that fall within 52km buffered area
Climate Stations
1. Station Data
- BEST-STAStations used to calibrate
the hydrologic model
Input Data Sets used in Hydrologic ModelInput Data Sets used in Hydrologic Model
Climate Stations
1. Station Data
- BEST-STA
Input Data Sets used in Hydrologic ModelInput Data Sets used in Hydrologic Model
- ALL-STAAll stations within the RegCM2 buffered area (excluding BEST-STA)
RegCM2 Grid Nodes
1. Station Data
- BEST-STA
Input Data Sets used in Hydrologic ModelInput Data Sets used in Hydrologic Model
2. DDS
- ALL-STA
1. Station Data
- BEST-STA
Input Data Sets used in Hydrologic ModelInput Data Sets used in Hydrologic Model
2. DDS
3. SDS
- ALL-STA
NCEP Grid Nodes
NCEP Grid Nodes
1. Station Data
- BEST-STA
Input Data Sets used in Hydrologic ModelInput Data Sets used in Hydrologic Model
2. DDS
3. SDS
4. NCEP
- ALL-STA
Nash-Sutcliffe Goodness of Fit StatisticComputed between measured and simulated runoff
Best-Sta
Nash-Sutcliffe Goodness of Fit StatisticComputed between measured and simulated runoff
Best-Sta
INPUT TIME SERIES:
Test1
Best-Sta PRCP
Bias-DDS TMAXBias-DDS TMIN
Test2 Bias-DDS PRCPBest-Sta TMAXBias-DDS TMIN
Test3Bias-DDS PRCP
Bias-DDS TMAX
Best-Sta TMIN
Nash-Sutcliffe Goodness of Fit StatisticComputed between measured and simulated runoff
Best-Sta
R-Square Values between Daily “Best” timeseries and:
All-Sta, Bias-All, DDS, Bias-DDS, NCEP, Bias-NCEP, and SDS
Minimum TemperatureMaximum TemperaturePrecipitationAlapahaAnimasCarsonCle Elum
R-S
quar
e
R-S
quar
e
R-S
quar
e
All-
Sta
DD
S
Bia
s-A
llB
ias-
DD
SN
CE
PB
ias-
NC
EP
SD
S
All-
Sta
DD
S
Bia
s-A
llB
ias-
DD
SN
CE
PB
ias-
NC
EP
SD
S
All-
Sta
DD
S
Bia
s-A
llB
ias-
DD
SN
CE
PB
ias-
NC
EP
SD
S
Rainfall-dominated basin – highly controlled by daily variations in precipitation
Snowmelt-dominated basins – highly controlled by daily variations in temperature and radiation
Minimum TemperatureMaximum TemperaturePrecipitationAlapahaAnimasCarsonCle Elum
R-S
quar
e
R-S
quar
e
R-S
quar
e
All-
Sta
DD
S
Bia
s-A
llB
ias-
DD
SN
CE
PB
ias-
NC
EP
SD
S
All-
Sta
DD
S
Bia
s-A
llB
ias-
DD
SN
CE
PB
ias-
NC
EP
SD
S
All-
Sta
DD
S
Bia
s-A
llB
ias-
DD
SN
CE
PB
ias-
NC
EP
SD
S
Compare SDS and ESP Compare SDS and ESP Forecasts using PRMSForecasts using PRMS
Perfect model scenario
-Ensemble Spread• Range in forecasts
-Ranked Probability Score• measure of probabilistic forecast skill• forecasts are increasingly penalized as more probability is assigned to event categories further removed from the actual outcome
Ranked Probability Skill Score (RPSS) for forecast Ranked Probability Skill Score (RPSS) for forecast days 0-8 and month using measured runoff and days 0-8 and month using measured runoff and simulated runoff produced using: (1)simulated runoff produced using: (1) SDSSDS output output
and (2)and (2) ESPESP techniquetechnique
For
ecas
t D
ay
Month MonthJ F M A M J J A S O N D J F M A M J J A S O N D
8
6
4
2
0
8
6
4
2
0
0.1 0.3 0.5 0.7 0.9
RPSSRPSS
ESPSDS
Perfect Forecast: RPSS=1
Forecast Spread for forecast days 0-8 and month Forecast Spread for forecast days 0-8 and month using measured runoff and simulated runoff using measured runoff and simulated runoff produced using: (1)produced using: (1) SDSSDS output and (2)output and (2) ESPESP
techniquetechnique
For
ecas
t D
ay
Month MonthJ F M A M J J A S O N D J F M A M J J A S O N D
8
6
4
2
0
8
6
4
2
0
500 1500 2500 3500 4500Forecast SpreadForecast Spread
ESPSDS
Comparison of hydrologic model inputs -- Precipitation
For
ecas
t D
ay
Month MonthJ F M A M J J A S O N D J F M A M J J A S O N D
8
6
4
2
0
8
6
4
2
0
0.1 0.2 0.3 0.4 0.5Pearson CorrelationPearson Correlation
ESPSDS
R-square values calculated between daily basin-mean measured and (1) SDS and (2) ESP precipitation values
Daily basin precipitation mean by month and
forecast day for ESP (red line) and SDS (boxplot)
Comparison of hydrologic model inputs – Maximum Temperature
For
ecas
t D
ay
Month MonthJ F M A M J J A S O N D J F M A M J J A S O N D
8
6
4
2
0
8
6
4
2
0
0.1 0.3 0.5 0.7 0.9Pearson CorrelationPearson Correlation
ESPSDS
R-square values calculated between daily basin-mean measured and (1) SDS and (2) ESP maximum temperature values
Daily basin maximum temperature mean by
month and forecast day for ESP (red line) and SDS
(boxplot)
Nested Domains for MM5
Evaluating MM5 Output Using PRMS
HRU Configurations
Daily Precipitation Mean by
Month
Percent Rain Days by Month
XYZ vs MM5
Daily Basin Maximum and Minimum Temperature Mean by Month
XYZ vs MM5
HRU HRU ConfigurationsConfigurations
Results
Wet period
Dry period
5 Years of data
2-yr calibration
3-year evaluation
Improving Flash Flood Prediction Through a Synthesis of NASA
Products, NWP Models and Flash Flood Decision Support Systems
NASA NOAA NCAR USGS
Proposed Flash Flood Prediction Program
Components• 1km Noah LSM in the LIS framework (ensemble mode) • WRF (ensemble runs) • DSS – USGS Modular Modeling System (MMS)
– Sacremento
– CASC2D
– PRMS
• Forecasts– 15 minutes out to 24 hours
– 1 km resolution
USGS Modular Modeling System (MMS)Toolbox for Modeling,
Analysis, and DSS Development and
Application
Summary
• Statistical and dynamical downscaling can provide reasonable input to drive hydrologic models for a variety of applications.
• Statistical downscaling with XYZ distribution– handles spatial and elevation effects– most effective for frontal type storms – limited value for convective storm systems– based on historic climatology which may limit use
for future climate scenarios
Summary• Dynamical downscaled output handles spatial
and elevational effects, and frontal and convective storm types. However bias correction required which results in similar limit on use with future climate scenarios.
• Statistical downscaling shows improvements over ESP based simulations for short-term forecasts.
Summary• Demonstrated higher degree skill of one
method over the over varies with the climatic and physiographic region of the world and the performance measures.
• Need for hydrologic and atmospheric modelers to work collaboratively to improve downscaling methods. Each community can provide valuable feedback to the other.