Kalman Filter–Based CMORPH
Transcript of Kalman Filter–Based CMORPH
Kalman Filter–Based CMORPH
ROBERT J. JOYCE
NOAA/Climate Prediction Center, Camp Springs, Maryland, and Wyle, Inc., McLean, Virginia
PINGPING XIE
NOAA/Climate Prediction Center, Camp Springs, Maryland
(Manuscript received 22 February 2011, in final form 19 May 2011)
ABSTRACT
A Kalman filter (KF)-based Climate Prediction Center (CPC) morphing technique (CMORPH) algorithm
is developed to integrate the passive microwave (PMW) precipitation estimates from low-Earth-orbit (LEO)
satellites and infrared (IR) observations from geostationary (GEO) platforms. With the new algorithm, the
precipitation analysis at a grid box of 8 3 8 km2 is defined in three steps. First, PMW estimates of in-
stantaneous rain rates closest to the target analysis time in both the forward and backward directions are
propagated from their observation times to the analysis time using the cloud system advection vectors
(CSAVs) computed from the GEO–IR images. The ‘‘prediction’’ of the precipitation analysis is then defined
by averaging the forward- and backward-propagated PMW estimates with weights inversely proportional to
their error variance. The IR-based precipitation estimates are incorporated if the gap between the two PMW
observations is longer than 90 min. Validation tests showed substantial improvements of the KF-based
CMORPH against the original version in both the pattern correlation and fidelity of probability density
function (PDF) of the precipitation intensity. In general, performance of the original CMORPH degrades
sharply with poor pattern correlation and substantially elevated (damped) frequency for light (heavy) pre-
cipitation events when PMW precipitation estimates are available from fewer LEO satellites. The KF-based
CMORPH is capable of producing high-resolution precipitation analysis with much more stable performance
with various levels of availability for the PMW observations.
1. Introduction
Despite its crucial importance in many meteoro-
logical, hydrological, and water resources management
applications, accurate measurements of regional and
global precipitation remain a challenging task. In par-
ticular, precipitation variations of fine spatial and tem-
poral scales are not well observed over most of the
globe, although they represent a substantial portion of
the overall variability and play an important role in
hydrological cycle and land–atmosphere interactions
(Dai et al. 1999, 2007; Tian et al. 2007; Hong et al. 2007).
While precipitation products based on observations
from individual platforms and instruments (gauge, ra-
dar, satellites, etc.) have been widely utilized in a variety
of operational and research applications, integrating
information from multiple satellite sensors as well as
ground observations (gauges and radars) further im-
proves the quality and resolution of precipitation anal-
ysis (Adler et al. 1994; Huffman et al. 1997; Xie and Arkin
1996, 1997; Sapiano et al. 2008). Substantial progress has
been made in the past decade to generate precipitation
estimates at high spatial and temporal resolutions through
combined use of infrared (IR) and passive microwave
(PMW) observations from multiple satellites. Two cate-
gories of techniques have been developed to combine the
precipitation and/or cloud information from individual
satellite PMW and IR observations. The first category in-
cludes the Precipitation Estimation from Remote Sensing
Information using Artificial Neural Network (PERSIANN;
Hsu et al. 1997), the Naval Research Laboratory (NRL)
blended satellite precipitation estimates (Turk et al. 2003),
and the Tropical Rainfall Measurement Mission (TRMM)
Multisatellite Precipitation Analysis (TMPA; Huffman
et al. 2007, 2009). A temporally changing and regionally
Corresponding author address: Dr. Pingping Xie, NOAA/Climate
Prediction Center, 5200 Auth Road, Suite 605, Camp Springs, MD
20746.
E-mail: [email protected]
DECEMBER 2011 J O Y C E A N D X I E 1547
DOI: 10.1175/JHM-D-11-022.1
� 2011 American Meteorological SocietyUnauthenticated | Downloaded 04/22/22 11:18 PM UTC
dependent empirical relationship between precipitation
intensity and cloud-top temperature is defined using col-
located IR and PMW data, assuming that PMW estimates
represent the ‘‘truth’’ of instantaneous precipitation rates at
the ground. This relationship is then applied to estimate
precipitation over the globe from the high-resolution IR
data observed from geostationary (GEO) platforms. In
some of the algorithms, the IR-based precipitation esti-
mates are further merged with PMW estimates from low-
Earth-orbit (LEO) satellites, wherever available, to form
the final satellite-based precipitation products. Key to this
category of techniques is the IR–precipitation relationship,
established through a sophisticated artificial neural net-
work system in the PERSIANN (Hsu et al. 1997) and
through matching the probability density function (PDF) of
IR against that of the PMW precipitation estimates in the
NRL (Turk et al. 2003) and TMPA (Huffman et al. 2003,
2007). While utility of the GEO–IR data ensures the pro-
duction of high-resolution precipitation estimates over a
quasi-global domain, lack of direct physical linkage be-
tween the precipitation and cloud-top temperature results
in substantial error in the IR-based precipitation estimates
and thereby the final PMW–IR merged analyses. Currently,
the official TRMM (Simpson et al. 1988; Kummerow et al.
2000) takes the 3-hourly precipitation analysis generated by
the TMPA algorithm (3B42) as its official level 3 product.
For convenience purposes, here we call the techniques of
this category ‘‘Euler approach’’ to reflect their common
feature that only the PMW and IR observations available
locally inside the target grid box are used directly in the
definition of analysis.
In the second category of techniques, called ‘‘Lagrangian
approach,’’ estimates of instantaneous rain rates from the
PMW observations from LEO are propagated and in-
terpolated in the combined time–space domain through
the use of the precipitating cloud system advection vec-
tors (CSAVs) computed from consecutive IR images
from the GEO platforms (Joyce et al. 2004; Ushio et al.
2009). PMW estimates are propagated along the advec-
tion vectors from the time of observation to that of the
target analysis. Since the IR observations are not used
directly to estimate precipitation, the error inherent in the
IR estimates (especially when PMW observations are
available around the target analysis time) is excluded as
a potential contaminate of the merged analysis. Recently,
Behrangi et al. (2010) proposed a conceptual model to
construct high-resolution precipitation analysis by prop-
agating PMW estimates along motion vectors defined by
cloud tracking with consideration of intensity changes
during the propagation.
The Lagrangian approach of propagating and morph-
ing high-resolution satellite precipitation estimates was
first adopted by Joyce et al. (2004) in the development of
their Climate Prediction Center (CPC) morphing tech-
nique (CMORPH). The success of CMORPH inspired
other agencies such as Japan Aerospace Exploration
Agency (JAXA) with their Global Satellite Mapping of
Precipitation (GSMaP) algorithm to follow CMORPH’s
methodology of using PMW rainfall in a Lagrangian
framework with IR-derived vectors as the method of
propagation (Ushio et al. 2009). Recent evaluation results
showed superior performance of CMORPH and GSMaP
compared to other high-resolution satellite precipitation
products derived using an Euler approach (Xie et al. 2007;
Ebert et al. 2007; Shen et al. 2009). The high temporal and
spatial resolution global precipitation estimates that are
created by these algorithms are applied in a wide variety
of research and operational applications.
While CMORPH consistently presents excellent
performance in estimating the spatial distribution and
temporal variations of precipitation over most of the
global regions and for all seasons, shortcomings exists in
the current version CMORPH and its high-resolution
precipitation products. Further improvements are needed
for the CMORPH technique to better meet the require-
ments of both science and societal communities.
First, the current CMORPH technique does not take
full advantage of precipitation information from PMW
observations and other sources. In the process of prop-
agating the PMW precipitation estimates, only data from
one scan closest to the target analysis time from each of
the forward and backward directions are included; IR-
based precipitation estimates are not used at all to adjust
the rainfall estimate. Despite their relatively poor accu-
racy, IR-based estimates may provide useful information
when no PMW estimates are available around the target
analysis time. The weights used to define the final analysis
from the propagated PMW estimates are approximated
as inversely proportional to the length of the propagation
time without consideration of the instrument dependency,
temporal propagation direction, surface type, season, lat-
itude, or the nonlinear nature of the estimation error. In
addition, the intensity of the precipitation is assumed un-
changed over the duration of both the forward and back-
ward propagation of the PMW estimates.
The objective of this work is to develop a prototype
model of the Kalman filter (KF)-based CMORPH that
is capable of producing high-resolution global pre-
cipitation analysis with improved accuracy through the
incorporation of additional IR-based information and
through the integration of all PMW- and IR-based in-
formation available using more precise weights. Section
2 of this paper describes the current CMORPH algo-
rithm and individual datasets used as inputs to the merg-
ing process, sections 3 and 4 present the development of
the prototype model of KF-based CMORPH over the
1548 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
conterminous United States (CONUS) and its implemen-
tation over the globe, and a summary and discussion is
provided in section 5.
2. Input data to the CMORPH processing system
a. A brief description of the current CMORPHalgorithm
CMORPH derives high-resolution (8 3 8 km2) half-
hourly precipitation analyses over a quasi-global domain
(608S–608N) through the integration of PMW estimates
from all available LEO satellites in a Lagrangian frame-
work. First, advection vectors of precipitating systems over
the globe are determined by computing spatially lagged
correlation between consecutive IR images observed by
GEO satellites. The CSAVs are then further refined as a
proxy to rainfall propagation through adjustments de-
termined from comparison studies against radar rainfall
motion. Precipitating cloud systems detected from PMW
satellite observations are then propagated along the ad-
vection vectors from the orbit observation time to the
target analysis time under the assumption that precipi-
tation intensity remains the same over the period. This
propagation is performed in both forward and backward
directions in time. The final precipitation analysis is de-
fined as the weighted mean of the propagated PMW es-
timates with the weights inversely proportional to the
length of propagation.
b. PMW precipitation estimates
Previously the only source of the information used
by CMORPH to define global high-resolution precip-
itation analyses was the level 2 instantaneous rain-rate
products retrieved from PMW observations of LEO
satellites (Ferraro 1997; Ferraro et al. 2000; Kummerow
et al. 2001). At the time of this writing, PMW pre-
cipitation estimates from up to nine LEO satellites are
available and utilized as inputs to the CMORPH pro-
cessing system. The most recent Goddard profiling algo-
rithm (GPROF; Kummerow et al. 1996; Olson et al. 1999)
is used to derive rainfall from PMW imagers. The tech-
nique used to retrieve precipitation from the PMW Ad-
vanced Microwave Sounding Unit (AMSU)/Microwave
Humidity Sounder (MHS) is the latest National Environ-
mental Satellite, Data, and Information Service (NESDIS)
algorithm (Vila et al. 2007), which reduces the error as-
sociated with the over- or underestimation of rainfall over
nadir–limb beam positions of the cross-track scanners
observed in the previous version of the algorithm. Fairly
well spaced in orbit time, observations from these plat-
forms combined provide critical information in defining
precipitation analysis around the diurnal cycle.
A series of innovative preprocessing procedures are
developed and implemented to prepare the PMW pre-
cipitation estimates to be used in the morphing process.
First, level 2 PMW estimates of instantaneous rain rates
generated at individual retrievals [or field of view (FOV)]
are mapped onto a global grid of 0.07278 latitude–longitude
(8 3 8 km2, at equator). Multiple 8-km grid boxes cov-
ered by a single satellite retrieval are assigned with the
same rain-rate value to ensure the spatial completeness and
representativeness of the resulting 8-km gridded fields of
PMW estimates.
To remove the systematic differences between PMW
estimates derived from observations of different instru-
ments using different algorithms, PMW estimates from
the TRMM Microwave Imager (TMI) are selected as
the reference standard and those from all other plat-
forms are calibrated against the TMI estimates through
matching the rain-rate PDF. The TMI estimates are cho-
sen as the normalization standard because of the finer
spatial resolution and emission detection (over ocean) of
the imaging sensors along with the dynamic ability of the
TRMM satellite to underfly all polar orbiting satellites,
allowing precise temporal and spatial matching of the re-
spective estimates. Data pairs of collocated TMI and tar-
get PMW estimates observed within 30 min or closer are
collected from the mapped 8 3 8 km2 grid fields for a
10-day period up to the target date. Accumulated PDF
tables are then constructed and utilized to adjust the target
PMW estimates. The PDF tables are created for the land
and oceanic regions separately and for each 108 latitude
band, using collocated data pairs over a wide domain of 308
in latitudes centering at the target band. Since the TMI
PMW estimates are available only from 408S to 408N, cali-
bration for PMW estimates from other satellites beyond the
408 parallels is performed against the PMW estimates from
the Advanced Microwave Scanning Radiometer (AMSR).
A preliminary examination showed very close agreements
in rain-rate PDF between the AMSR and TMI over tropics
and subtropics. PMW estimates from the future PMM core
satellite are expected to provide an improved calibration
standard especially over extratropics. Figure 1 illustrates
an example of the PDFs for the PMW instantaneous rain
rates from the TMI (green), original (blue), and cali-
brated (red) AMSU estimates for June–August 2005. The
PDF of the calibrated AMSU estimates matches much
better with that of the TMI than the original AMSU.
While the procedures described above work effec-
tively for the calibration of PMW estimates from most
LEO satellites, an additional calibration must be applied
for oceanic precipitation estimates derived from the PMW
observations of AMSU/MHS aboard National Oceanic
and Atmospheric Administration (NOAA) polar orbiting
platforms and the Meteorological Operational Satellite
DECEMBER 2011 J O Y C E A N D X I E 1549
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
(MetOp). Because of the limitations of the PMW sensors,
the AMSU/MHS is unable to detect some of the light
precipitation over water, resulting in a PDF of sub-
stantially reduced raining frequency (Vila et al. 2007).
Performing no correction for cases of zero rainfall, the
PDF matching procedure described in the last para-
graph would generate corrected AMSU precipitation
estimates with negative bias in overall magnitude com-
pared to the TMI estimates. To solve this problem,
a recursive filter is developed and applied to revise the
correction table slightly until the bias between the re-
sulting adjusted AMSU and TMI estimates is negligible.
c. Cloud system advection vectors
Key to the CMORPH technique is the propagation of
PMW observations of instantaneous rain rates in the
combined time–space domain along the CSAVs. Follow-
ing Smith and Phillips (1972) and Purdom and Dills (1994),
the CSAVs are derived by computing the displacements
of cloud systems detected from full-resolution (4 3
4 km2) GEO–IR images in 30-min intervals. Spatially
lagged correlation is calculated for the brightness tem-
perature (Tb) arrays on two consecutive GEO–IR im-
ages and the displacement with the highest correlation
is used to define the CSAV. Assuming the CSAVs do not
present substantial variations on small spatial scales,
definition of the CSAVs is performed on a 2.58 latitude–
longitude grid over the globe from 608S to 608N using the
GEO–IR data over a 58 latitude 3 58 longitude domain
centering at the target grid point.
Early versions of CMORPH used CSAVs directly to
propagate PMW-derived precipitation. However, it was
soon determined that the west–east and south–north
advection rates were too fast in the North Hemisphere
midlatitudes (Joyce et al. 2004). To correct this, a speed
adjustment procedure was developed. First, rainfall
system advection vectors were computed by spatially
lagging hourly U.S. Next Generation Weather Radar
(NEXRAD) stage II (Klazura and Imy 1993) radar
rainfall (mapped to the same 8 3 8 km2 grid) in the exact
same dimensions and manner CSAVs are computed from
IR. The frequency distribution of CSAV and radar rain-
fall advection rates indicated that north–south rates are
quite similar but that west–east CSAV speeds were about
3–4 times as fast compared to the radar-derived vectors,
and south–north rates were twice as fast (Joyce et al.
2004, their Fig. 7). These systematic differences are con-
sistent with several case studies that show the tendency of
IR features to quickly stream to the northeast on the east
side of long-wave troughs, with the actual rainfall also
moving in this direction but at a slower rate. The CSAVs
computed from the IR images over midlatitudes are ad-
justed accordingly based on the comparison results. For
consistency with the Northern Hemisphere, the meridio-
nal adjustment is applied to vectors of the opposite sign
in the Southern Hemisphere. The incorporation of this
adjustment procedure into the CMORPH processing has
resulted in improved propagation of precipitation features.
d. IR-based precipitation estimates
In the first-generation CMORPH algorithm (Joyce
et al. 2004), the GEO–IR data are utilized only to com-
pute the CSAVs for propagating the PMW instantaneous
precipitation estimates toward the targeted analysis time.
In developing the KF-based CMORPH, precipitation
estimates derived from GEO–IR observations will also be
incorporated to improve the quantitative accuracy of the
integrated precipitation analysis when PMW observations
are not available over an extended period of time. To this
end, a new technique is developed to estimate precipi-
tation from the full-resolution global GEO–IR data of
Janowiak et al. (2001) by matching the PDF of Tbs from
GEO–IR observations to that of the PMW instantaneous
rain rates.
First, combined PMW precipitation estimates
(MWCOMB) are defined in a 30-min interval on the 8 3
8 km2 global grid system by averaging the calibrated
PMW rain rates from individual LEO satellites. Mean-
while, global arrays of IRTb are created on the same
time–space resolution by utilizing the full-resolution
(4 km–30 min) GEO–IR data of Janowiak et al. (2001).
Collocated pairs of the IRTb and MWCOMB rainfall
data are then collected and used to create PDFs. An
FIG. 1. The cumulated PDF of instantaneous rain rates for the
original (blue) and calibrated (red) PMW estimates from the
AMSU aboard the Polar-orbiting Operational Environmental
Satellites (POES), and the PMW estimates from the TRMM TMI
(green) over Northern Hemisphere land for 1 Jun–31 Aug 2005.
The cumulated PDF is plotted as the contribution to the mean
rainfall (mm day21, y axis) from all cases with instantaneous rain
rates equal to or larger than a selected intensity (mm hr21, x axis).
1550 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
estimation of 30-min precipitation over an 8 3 8 km2
grid box is assigned by matching the cumulated PDF of
the IRTb with that of the MWCOMB rain rates. To
ensure statistical stability and temporal–spatial conti-
nuity for the estimation of global precipitation, the PDF
tables are constructed for each 30-min time step and for
each 58 3 58 latitude–longitude grid box using collocated
data pairs collected over a 158 3 158 latitude–longitude
region centered at the target 58 latitude–longitude grid
box and covering a 9.5-hr time window centered at the
target estimation time. The short time window is used to
capture the current location in the diurnal cycle. Esti-
mation procedures are performed separately for land
and ocean to account for the differences in the IRTb–
precipitation relationship. Called IR rainfall frequency
(IRFREQ), precipitation estimates generated by this
technique are capable of capturing the spatial distribution
and temporal variations of precipitation with reasonable
quality over tropical and subtropical regions during warm
seasons (Fig. 2). A comprehensive evaluation of several
IR-based precipitation algorithms revealed superior per-
formance of the IRFREQ precipitation estimates com-
pared to other IR-based products (R. Kuligowski 2009,
personal communication).
3. Development of the prototype KF-basedCMORPH over CONUS
a. Conceptual model of the KF-CMORPH
Utilized widely in the atmospheric and oceanic data
assimilation community, the Kalman filter is an efficient
and effective recursive data processing algorithm to es-
timate the state of a linear system from a series of obser-
vations with different error characteristics. In general,
a Kalman filtering process consists of two sequential steps:
a ‘‘forecast step’’ that creates the ‘‘forecast’’ of the final
analysis as well as the error variance for the forecast, fol-
lowed by an ‘‘analysis step’’ that modifies the forecast with
the observations (Kalnay 2003).
The forecast of a state variable at step i (Fi) is com-
puted from the analysis at the previous step (i 2 1) Ai21
through a forecast model M:
Fi 5 Mi21(Ai21), (1)
while the error variance for the forecast (sfi )2 can be
defined as
(sfi )2
5 P2 1 Q2, (2)
where P2 is the error variance attributable to the prop-
agation of the error in the initial condition over the
forecast process and can be computed using the forecast
model. The Q2, meanwhile, is the error variance created
in the forecast process with the perfect initial condition.
The final analysis Ai is defined by updating the fore-
cast (Fi) with the observations (Oi):
Ai 5 Fi 1 Ki(Oi 2 Fi), (3)
where Ki is the Kalman gain computed as a function of
the forecast error variance (sfi )2 and the observation
error variance (s0i )2:
Ki 5 (sfi )2=[(s0
i )21 (s
fi )2]. (4)
In our application to the modification of CMORPH,
a simplified implementation of the KF approach has
been taken as a first step to demonstrate the effective-
ness of the statistical framework in integrating pre-
cipitation from multisensors. PMW estimates from
multiple LEO platforms are first propagated backward
and forward from their observation times to the analysis
time. Weighted mean of the propagated PMW estimates
are then defined as the forecast at the analysis time, with
the weights inversely proportional to the error variance
for the propagated PMW estimates. The model adopted
here therefore only forecasts the precipitation by propa-
gating PMW estimates, assuming no changes in the pattern
and intensity of the precipitating systems. The forecast is
then refined by incorporating information from IR-based
precipitation estimates at the analysis time to form the fi-
nal precipitation analysis.
The forecast error, as defined in Eq. (2), should be
composed of two portions: i) propagation of the error in
the input PMW estimates composed mainly of the level
2 PMW retrieval error, and ii) KF-CMORPH model
error arising in the process of defining the forecast from
the PMW estimates caused by, among many other fac-
tors, inaccurate estimation of the propagation vectors
and unrealistic assumption of unchanged precipitating
systems (pattern and intensity) over the period of propa-
gation. In the development of this conceptual model,
however, the error for the forecast is simplified as a func-
tion of the propagation time and defined separately for
PMW estimates of different sensor types (section 3b). It is
critically important that future work will quantify the error
in the PMW estimates, its propagation in the processing,
and the error caused by various imperfect assumptions and
estimations in the integration process to further improve
the quantitative accuracy of the combined satellite pre-
cipitation estimates.
Our final goal is to construct a KF-based system
to produce high-resolution pole-to-pole global pre-
cipitation analysis through integration of information
from satellite IR, PMW observations, numerical model
DECEMBER 2011 J O Y C E A N D X I E 1551
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
simulations, and other available sources. As a first step,
we have developed a prototype model of the KF-based
CMORPH to combine precipitation estimates derived
from PMW and IR observations over a global domain
from 608S to 608N. The IR-based estimates are included
only when no PMW data are available within a window
of 90 min centered at the target analysis time.
b. Development of the prototype model over CONUS
Key to the development of the KF-CMORPH is the
definition of error structure for the propagated PMW- and
IR-based estimates as a function of instrument type,
propagation time and temporal direction, location, and
season. A preliminary test is first performed to define the
error through comparison of the IRFREQ and the PMW
estimates propagated through various lengths of time
against the stage-II radar observations (Klazura and Imy
1993) over CONUS for summer 2007. Estimation error
is expressed here as the correlation between the satellite
estimates and the stage-II radar data. Correlation for the
PMW estimates degrades sharply as they are propagated
from their observation time (Fig. 3). The magnitude and
FIG. 2. Evolution of hourly precipitation in association with the development of a severe
weather system over CONUS from (top to bottom) 2200 UTC on 4 Aug through 0100 UTC on
5 Aug 2009, as depicted by the current version of CMORPH (Joyce et al. 2004) based on (left)
PMW observations and by (right) the IRFREQ.
1552 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
the degradation rate of the correlation, however, differ
for different instruments, indicating the importance of
defining the error separately for estimates from different
platforms. During this summertime study over land, the
IR-based precipitation estimates outperform the PMW
estimates when the propagation time is longer than 90 min
or so, suggesting potential improvements in the final anal-
ysis through the inclusion of IR estimates to fill in PMW
observation gaps.
Using these error statistics, a prototype model KF-
CMORPH is developed to construct high-resolution
precipitation estimates over CONUS. The IR-based pre-
cipitation estimates are included as part of the inputs to
the KF-based CMORPH when no PMW observations are
available within a window of 90 min.
c. Evaluation of the prototype KF-CMORPHmodel over CONUS
To assess the performance of the KF-CMORPH anal-
yses relative to the original CMORPH analyses, we
compared both analyses with estimates of precipitation
from radar over CONUS during July–August 2009. The
radar data used for this comparison is the surface rain-rate
estimation at 1-km/5-min resolution over CONUS gen-
erated by the National Mosaic and Quantitative Pre-
cipitation Estimation (QPE) system (NMQ/Q2; Zhang
et al. 2009) developed at the NOAA National Severe
Storms Laboratory (NSSL) and the University of Okla-
homa. The NMQ/Q2 system combines information from
all ground-based radars constituting the NEXRAD
network and calibrates the radar rain rates with gauge
observations. The Q2 radar rainfall data at its original
resolution are integrated to define mean rain rates at
a space–time scale of 0.258 latitude–longitude and 30 min
to compare against the CMORPH satellite estimates over
CONUS.
Figure 4a illustrates time series of pattern correlation
between the radar observations and three sets of satel-
lite precipitation estimates: the IR-based IRFREQ, the
original CMORPH, and the KF-based CMORPH. Cor-
relation used here is Pearson’s correlation coefficient
between two variables, defined as the covariance of the
two variables divided by the product of their standard
deviations:
rX,Y 5cov(X, Y)
sXsY
5E[(X2 mX)(Y2 mY)]
sXsY
. (5)
The KF-CMORPH performs consistently better than
the original CMORPH throughout the two-month pe-
riod. The improvements in the overall pattern correla-
tion, however, are marginal. The correlation computed
over the combined space–time domain is 0.717 and 0.725
for the original and KF-CMORPH, respectively (Table 1,
first line from bottom). Since the majority of PMW ob-
servations used as inputs to the CMORPH are from
LEO platforms that fly over different geographical lo-
cations at fixed local times, performance of the satellite-
based precipitation estimates are evaluated according to
different local hours (Fig. 5a). Pattern correlations for
both the original and KF-CMORPH estimates present
wave-shaped variations, showing higher correlation over
the half-hourly slots with frequent PMW flights (0214,
0618, and 0921 LST; see Table 2), a reflection that PMW
retrievals with shorter propagation time contain less error,
as shown in Fig. 3. The KF-CMORPH presents substantial
improvements upon the original CMORPH over the half-
hourly slots existing in PMW observation gaps, reflecting
the relative additional skill gained in the KF-CMORPH
through the incorporation of IR-based estimates.
Both the number and the configuration of the LEO
satellites are sensitive to the performance of satellite
precipitation estimates derived through the integration
of individual PMW estimates through a Lagrangian
approach. As shown in Table 1 and Fig. 4, while corre-
lation for both the original and KF-CMORPH improves
with increased number of LEO–PMW satellites, corre-
lation differences for precipitation estimates based on
four- and seven-satellite configurations are very small. A
careful examination (results not shown) revealed that
this ‘‘level off’’ in performance is caused primarily by the
orbit patterns for the seven-satellite configuration,
which contains a ‘‘hole’’ between 19:30 and 01:30 local
time in sampling the diurnal cycle (Table 1). By opti-
mizing diurnal sampling, the skill of the four-satellite
FIG. 3. Correlation for the IR-based and PMW-propagated 0.258
latitude–longitude precipitation estimates as a function of in-
strument type and propagation time computed by comparisons
with stage-II radar data over CONUS for June–September (JJAS)
2007. Positive and negative propagation times indicate forward and
backward propagation, respectively.
DECEMBER 2011 J O Y C E A N D X I E 1553
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
configuration offsets the addition of PMW satellites in
a poorly designed seven-satellite configuration. Turk et al.
(2010) performed a similar experiment and found that
performance for precipitation estimates based on an Euler
approach will degrade when all cross-track sounders or the
morning local time crossing satellites were removed. While
it is not the central topic of this work to examine the sen-
sitivity of integrated precipitation estimates to the avail-
ability, quality, and configuration of input PMW and IR
satellite observations, future work is planned to address
this critically important issue to get a better understanding
of how the integration algorithm should be designed for
the estimation of precipitation for periods observed with
fewer satellites.
A set of synthetic experiments are designed and im-
plemented to further understand the performance of the
KF-CMORPH in generating integrated precipitation es-
timates from various LEO–PMW configurations. To this
end, high-resolution precipitation estimates over CONUS
are generated for the two-month period in 2009 using the
original and the KF-based CMORPH algorithms with
input PMW observations from only one, two, four, and
seven satellites selected from the all nine available plat-
forms (Table 2). The selection of the satellite configura-
tion is to mimic the PMW instrument availability over the
entire time span from 1987 to the present.
With fewer PMW observations available as inputs,
pattern correlation for the original CMORPH degrades
sharply from 0.717 for the nine-satellite configuration to
0.469 for the one-satellite simulation (Table 1). The
correlation for the original CMORPH is especially low
for the half-hourly local time slots precisely between two
consecutive LEO orbit times (Figs. 5b–e). The pattern
correlation for the original CMORPH degrades linearly
with the propagation time and falls down to ;0.1 for
estimates defined by propagating PMW by more than
5.0 h (Fig. 6, top).
The KF-CMORPH shows much less degradation in
correlation with PMW observations from a reduced
TABLE 1. Correlation of daily Q2 radar rainfall with the original
and Kalman filter CMORPH with input PMW estimates from
various LEO satellite configurations. Comparisons were made for
daily 0.258 latitude–longitude rainfall over the CONUS for a two-
month period from 1 Jul to 31 Aug 2009.
PMW satellite
configuration CMORPH
Kalman filter
CMORPH
One satellite 0.469 0.653
Two satellites 0.562 0.673
Four satellites 0.663 0.704
Seven satellites 0.659 0.702
Nine satellites 0.717 0.725
FIG. 4. Time series of correlation between the Q2 radar-estimated precipitation and
IRFREQ (red), the original CMORPH (black), and the prototype KF-CMORPH (green).
Comparisons are for daily 0.258 latitude–longitude precipitation over the CONUS for July–
August 2009. Results for CMORPH with (a) nine-, (b) seven-, (c) four-, (d) two-, and (e) one-
satellite configurations.
1554 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
number of platforms (Table 1, Figs. 4b–e), thanks to the
contribution from the IR-based precipitation estimates.
The daily 0.258 pattern correlation, computed using all
data for the two-month period and over all grid boxes
over CONUS, is 0.725 and 0.653, respectively, for the
KF-CMORPH with the nine- and one-satellite configu-
rations (Table 1). Thirty-minute 0.258 pattern correlation
for KF-CMORPH decreases initially as the propagation
time extends and then becomes stabilized at ;0.45 for
propagation time of 1.5 h and longer (Fig. 6, bottom),
a substantial improvement upon that for the original
CMORPH.
The PDF of the rainfall intensity generated by the
original and KF-based CMORPH is virtually identical
for estimates defined with propagation time within
30 min and it is very close to that of the IR-based esti-
mates (Fig. 7a). All of the three satellite-based products
(the two versions of CMORPH and the IRFREQ), how-
ever, exhibit overestimation for events of precipitation
(.25 mm h21) compared to the radar observations from
Q2 (Fig. 7a) because of the systematic error in the level
2 PMW retrievals. The frequency for heavy precipitation
estimates generated by the original CMORPH drops sig-
nificantly over grid boxes where analysis is defined by
TABLE 2. Names and equatorial crossing times (ECT, local time) used in the synthetic experiments for CMORPH with different LEO
satellite configurations.
No. Satellite ECT One satellite Two satellites Four satellites Seven satellites Nine satellites
1 TRMM Precessing X X X
2 Aqua 1340 X X X X
3 NOAA-18 1345 X X
4 NOAA-19 1355 X X
5 NOAA-15 1645 X X
6 NOAA-16 1745 X X X X
7 DMSP-13* 1820 X X
8 NOAA-17 2125 X
9 MetOp 2135 X X X
* Defense Meteorological Satellite Program (DMSP).
FIG. 5. Correlation between the Q2 radar-estimated precipitation and IRFREQ (red), the
original CMORPH (black), and the prototype KF-CMORPH (green), plotted as a function of
local time (x axis). Comparisons are for 0.258 latitude–longitude 30-min precipitation rates over
the CONUS for July–August 2009. Results for CMORPH with (a) nine-, (b) seven-, (c) four-,
(d) two-, and (e) one-satellite configurations.
DECEMBER 2011 J O Y C E A N D X I E 1555
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
propagation of PMW over an extended period of time
(Fig. 7, black lines). The PDF for the KF-CMORPH,
meanwhile, is relatively stable for precipitation estimates
over grid boxes with different propagation times (Fig. 7,
green lines), thanks to the incorporation of IR-based es-
timates that retain the original PMW rainfall rate distri-
bution (Fig. 7, red lines).
Compared to the original CMORPH, the KF-CMORPH
described in this paper exhibits substantial improve-
ments and much more stable performance in integrating
high-resolution precipitation estimates with limited sam-
pling from the PMW observations. This shows clearly the
potential strength of the KF-CMORPH in constructing
high-resolution precipitation estimates for the entire
TRMM Global Precipitation Mission (GPM) era with
relatively stable performance.
4. Global implementation of KF-CMORPH
a. Development of a prototype system forgenerating global precipitation analyses
This conceptual KF-CMORPH developed using the
CONUS data is implemented for constructing the
precipitation estimates over the global domain from 608S
to 608N. Error statistics for PMW- and IR-based
precipitation estimates are defined for individual in-
struments as a function of region and season through
comparisons with the concurrent PMW estimates from
the TRMM TMI. Error functions for the TMI are taken
to be the same as those for the AMSR for Earth Ob-
serving System (EOS) (AMSR-E), based on an early
comparison against the stage-II radar observations over
CONUS. Over land, the error functions are computed
for each 108 latitude band using data collected over
a 308-wide latitude band centered on the target band. No
zonal differences in the error are considered because of the
limited sampling of the data. Over ocean, the error func-
tions are defined for each 208 3 208 latitude–longitude
box using data over a 408 3 408 latitude–longitude region
centered on the target box. Over both land and ocean, the
error functions are calculated for each month using data
over a five-month period centered on the target month to
account for the seasonal variations. The comparisons
against stage II were done once, while those against TMI
are updated monthly.
As shown in Fig. 8, evolution of estimation error, shown
as correlation with the TMI estimates for the Special Sen-
sor Microwave Imager (SSM/I) and AMSR-E estimates
over the propagation period, presents strong regional
variations because of differences in the time scales of
the target precipitation systems. In particular, error for
PMW estimates exhibit distinct contrasts over land and
ocean, implying the importance of defining the error
separately for land and ocean.
Utilizing the error statistics defined for the propagated
PMW and the IR-based precipitation estimates, a pro-
totype KF-based CMORPH algorithm system has been
developed at NOAA/CPC to produce high-resolution
precipitation analysis parallel with the original version of
CMORPH. Figure 9 illustrates the evolution of a rapid
developing precipitating system over CONUS as depicted
by the original CMORPH (left), the IR-based IRFREQ
(second from left), the KF-CMORPH (third from left)
and the stage-II radar estimates (right). PMW observa-
tions were scanned over the target region at the first and
last half-hourly time slots. The original CMORPH, in-
terpolating precipitation estimates from the two PMW
orbits, therefore missed the peak of the precipitation event
around 21:30 UTC (fifth row from top) as observed by
the radar and the IRFREQ. Incorporating IR-based pre-
cipitation estimates in 30-min intervals enables the KF-
CMORPH to capture the rapid development of the system,
improving the overall performance of the new technology.
b. Quantitative examination of the globalKF-CMORPH
The same synthetic tests are performed for the origi-
nal and KF-based CMORPH using PMW estimates from
FIG. 6. Correlation between the Q2 radar precipitation and
precipitation estimates derived from (top) the original CMORPH
and (bottom) the KF-CMORPH using input PMW observations
from various configurations of LEO satellites. Correlation is
computed for 0.258 latitude–longitude 30-min mean rain rates over
the CONUS for July–August 2009.
1556 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
different LEO satellite configurations shown in Table 2.
‘‘Ground truth’’ for precipitation of high space–time res-
olution, such as the Q2 radar estimates, is not available
over most global regions for quantitative assessments of
the resulting CMORPH estimates. Therefore, we accu-
mulated the original and KF-based CMORPH to daily
precipitation averaged over a grid box of 0.258 latitude–
longitude and compared them with both the NOAA/CPC
unified global daily gauge analysis (Xie et al. 2010) over
land and the in situ measurements made by siphon gauges
installed on moored buoys of the Tropical Atmosphere
and Ocean (TAO) project (McPhaden et al. 1998) over
tropical oceans.
The CPC unified global daily gauge analysis is defined
by interpolating quality controlled gauge reports from
over 16 000 stations over the globe using the optimal
interpolation (OI) algorithm of Gandin (1965). Effects of
orographic influences to the precipitation are considered
in the definition of the gridded analysis (Xie et al. 2007).
The analysis is originally created on a 0.1258 latitude–
longitude grid and integrated to a 0.258 latitude–longitude
grid box mean for the verifications of CMORPH satellite
estimates in this study. The TAO buoy measurements of
daily rainfall used here are those observed at 40 moored
buoys over equatorial Pacific Ocean (http://www.pmel.
noaa.gov/tao/disdel/disdel.html). Since all of the TOA
buoys with in situ measurements are positioned at lo-
cations of even latitude–longitude (e.g., 88N, 1208E),
mean daily precipitation over four 0.258 latitude–longitude
grid boxes cornering at the buoy location is computed
for the CMORPH estimates and compared against the
buoy measurements. Although differences exist between
the area mean precipitation over a 0.58 latitude–longitude
grid box targeted by the satellite estimates and the ‘‘point’’
value measured at the buoy location, averaging over
a daily period reduces the discrepancies caused by the
differences in the spatial representativeness of the two
datasets. Same as in the CONUS experiments, exami-
nations are conducted for the two-month period from
1 July to 31 August 2009.
The KF-based CMORPH exhibits superior perfor-
mance over the original CMORPH in estimating pre-
cipitation over both land and ocean and for estimates
using PMW observations from all five LEO satellite
configurations (Table 3, Fig. 10). Pattern correlation
for the precipitation estimates over the global land
(equatorial Pacific) based on the original CMORPH
degrades sharply from 0.618 (0.596) when PMW ob-
servations from all nine satellites are included to 0.428
(0.437) when inputs are available from only one LEO
satellite. Performance of the KF-CMORPH, meanwhile,
is much more robust, with much smaller decreases in the
pattern correlation [0.556 (0.579) for only one LEO sat-
ellite] because of the inclusion of the IR precipitation
estimates. The two versions of the CMORPH present
biases of relatively close magnitude caused by the biases in
the PMW estimates used as primarily inputs to the in-
tegration algorithms.
To further examine the performance of the original
and KF-CMORPH, we analyzed the results from the
FIG. 7. Accumulated PDF for precipitation intensity derived from the Q2 radar (pink), the
IRFREQ (red), the original CMORPH (black), and the KF-CMORPH (green) using PMW
observations from four LEO satellites. The cumulated PDF is defined as the contribution to the
mean rainfall (mm day21, y axis) from all cases with instantaneous rain rates equal to or larger
than a selected intensity (mm h21, x axis). The PDF is derived from data over CONUS for July–
August 2009 from the synthetic satellite configuration experiments. Results are displayed
separately for PMW propagation time of a) 0, b) 30, and c) 240 min.
DECEMBER 2011 J O Y C E A N D X I E 1557
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
global synthetic experiment using four LEO satellites.
To this end, we compared the precipitation estimates
derived by the CMORPH algorithms from the four se-
lected LEO satellites with MWCOMB constructed from
the five withdrawn LEO satellites. Since the PMW es-
timates from the nine LEO satellites are all calibrated
against the same reference as described in section 2b,
comparisons against the MWCOMB based on the with-
drawn satellites’ estimates provide performance metrics
for the CMORPH integration algorithms, separating the
influence of the error (especially the bias) inherent in the
input PMW estimates to more clearly isolate the exami-
nation to the integration process performance.
As expected, the KF-CMORPH shows consistently
superior performance compared to the original CMORPH
(Figs. 11 and 12). Pattern correlation for precipitation
estimates constructed by the original CMORPH decreases
linearly with the propagation time for the PMW observa-
tions. Correlation for estimates of 30-min mean pre-
cipitation at a grid box of 0.258 latitude–longitude is as high
as ;0.8 when the PMW observations are available within
the 30-min window. It degrades down to less than 0.3 when
FIG. 8. Correlation between instantaneous rain rates derived from TRMM TMI and esti-
mates defined by propagating SSM/I and AMSR-E observations (left) forward and (right)
backward over various lengths of time, from (top) 0 to (bottom) 120. Correlation is computed
for the month of September 2009, using data over a five-month period from July to November
2009.
1558 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
FIG. 9. Evolution of a precipitating system over central United States depicted in 30-min intervals, from (top) 19:30 to (bottom)
23:30 UTC on 16 Jul 2009. (left to right) Precipitation distribution observed by the original CMORPH, IRFREQ, KF-CMORPH, and the
radar estimates.
DECEMBER 2011 J O Y C E A N D X I E 1559
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
defined by propagating an instantaneous PMW observa-
tion 5 h apart using the original CMORPH algorithm.
With the KF-CMORPH, the pattern correlation decreases
only slightly for the same propagation length, from ;0.8 to
;0.65, thanks to the integration of IR-based precipitation
estimates.
The KF-CMORPH also exhibits stronger capability in
generating precipitation estimates with much better fi-
delity in the PDF of precipitation intensity compared to
the original CMORPH (Fig. 12). PDF is identical for
precipitation estimates derived from the two versions of
CMORPH algorithms with propagation time less than
45 min, a reflection that no IR-based estimates are in-
cluded in the KF-CMORPH when PMW observations
are available nearby. The PDF for the CMORPH esti-
mates with propagation time of 0 min (PMW observa-
tions available within the 30-min target analysis window)
is very close to that for the withdrawn MWCOMB (Fig.
12a). The small differences in the PDF at low rain rates
are attributable to the fact that four of the five withdrawn
LEO satellites in this synthetic experiment carry sounder-
based PMW instruments that are relatively poor at de-
tecting weak precipitation, especially over midlatitude
oceans. Morphing the propagated PMW estimates re-
duces(increases) the PDF for heavy (light) precipitation,
even when the propagation is less than 30 min (Fig. 12b).
The PDF aliases degrade quickly with the propagation
time for precipitation estimates defined with the original
CMORPH. The KF-CMORPH, however, generates pre-
cipitation analysis with PDF close to that for the withdrawn
MWCOMB estimates for propagation time of various
lengths (Fig. 12c).
Degree of agreements in precipitation patterns ex-
amined in sections 3 and 4 is measured mainly using
pattern correlation as defined in Eq. (5). We did not
perform a significance test for each of the correlation
coefficients and the correlation differences shown in the
tables and figures. However, a brief estimation using the
methods described in the appendix of Xie and Arkin
(1995) revealed that most correlation coefficients shown
there and any correlation difference of 0.01 or larger is
statistically significant at a level of 95% or higher because
of the massive number of cases involved in the calcula-
tions (e.g., .600 000 cases in calculating each statistic
in Table 1). Taking together the statistical significance
of the correlation and the consistent trends of variation
patterns in the statistics, it is clear that all conclusions we
made are based on a solid physical foundation.
5. Summary
A new algorithm has been developed for CMORPH.
The Kalman filter technique is adopted to integrate the
TABLE 3. (top) Correlation and (bottom) bias (%) between the CMORPH daily 0.258 precipitation estimates and gauge observations for
July and August 2009.
a) Results over land
CMORPH version One satellite Two satellites Three satellites Seven satellites Nine satellites
Original CMORPH 0.428 0.496 0.580 0.587 0.618
9.7 82 8.7 8.6 7.3
KF-CMORPH 0.556 0.571 0.599 0.601 0.621
6.0 10.6 7.7 6.6 8.0
b) Results over equatorial Pacific
CMORPH version One satellite Two satellites Three satellites Seven satellites Nine satellites
Original CMORPH 0.437 0.498 0.573 0.581 0.596
31.3 32.6 32.9 34.0 31.0
KF-CMORPH 0.579 0.582 0.608 0.609 0.611
21.4 22.1 24.6 26.4 28.6
FIG. 10. Time series of pattern correlation between the
CMORPH daily precipitation estimates and the CPC unified
global daily 0.258 latitude–longitude precipitation analysis over the
global land. Only data over grid boxes with at least one reporting
gauge are included in the calculation. Results for (top) the original
and (bottom) Kalman filter–based CMORPH, with correlation for
CMORPH derived from different LEO–PMW satellite configu-
rations plotted in different colors.
1560 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
PMW precipitation estimates from LEO satellites and
IR observations from GEO platforms.
The KF-CMORPH derives the precipitation analysis
at a grid box of 8 3 8 km2 in three steps. First, PMW
estimates of instantaneous rain rates closest to the target
analysis time in both the forward and backward direc-
tions are propagated from their observation times to the
analysis time using the CSAVs computed from the GEO–
IR images. The forecast of the precipitation analysis is
then defined by averaging the forward- and backward-
propagated PMW estimates with weights inversely pro-
portional to their error variance. The IR-based precipitation
estimates are incorporated if the gap between the two
PMW observations is longer than 90 min.
The CSAVs used to propagate the PMW estimates
are calculated by computing the pattern correlation
between spatially lagged GEO–IRTb arrays from two
consecutive images. The spatial displacement with the
highest correlation is used to define the CSAVs. The IR-
based precipitation estimates used in this study are de-
fined by matching the PDF of GEO–IRTb with that of
the instantaneous PMW estimates.
Major differences between the KF-CMORPH and the
original CMORPH include i) the inclusion of the IR-based
precipitation estimates to fill in the gaps when PMW ob-
servations are not available nearby and ii) the improved
error definition for the PMW and IR precipitation esti-
mates as a function of instrument type, surface type, and
length of and temporal direction of propagation time, re-
gion, and season.
Validation tests showed substantial improvements of
the KF-CMORPH against the original version in both the
pattern correlation and fidelity of the precipitation intensity
PDF. In general, performance of the original CMORPH
FIG. 11. Correlation between the MWCOMB constructed from
withdrawn independent LEO satellites and precipitation estimates
derived from the IRFREQ (red), the original CMORPH (black),
and the KF-CMORPH (green) using PMW observations from four
LEO satellites. The MWCOMB is defined using the PMW esti-
mates from the five LEO satellites not included in the creation of
CMORPH analysis. Correlation is computed for 30-min mean rain
rates over 0.258 latitude–longitude grid boxes over the globe for
July–August 2009.
FIG. 12. Accumulated PDF of precipitation intensity derived from the MWCOMB based on
the withdrawn LEO satellites (purple), IRFREQ (red), the original CMORPH (black), and the
KF-CMORPH (green) using PMW observations from four LEO satellites. The cumulated PDF
is defined as the contribution to the mean rainfall (mm day21, y axis) from all cases with in-
stantaneous rain rates equal to or larger than a selected intensity (mm h21, x axis). The PDF is
derived from data from the synthetic experiments using four LEO satellites. Results are dis-
played separately for PMW propagation time of a) 0, b) 30, and c) 240 min.
DECEMBER 2011 J O Y C E A N D X I E 1561
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
degrades sharply with poor pattern correlation and sub-
stantially elevated (damped) frequency for light (strong)
precipitation when PMW precipitation estimates are
available from fewer LEO satellites. The KF-CMORPH is
capable of producing high-resolution precipitation analysis
with much more stable performance with various levels of
availability for the PMW observations.
Further improvements and enhancements are desir-
able for the KF-based CMORPH described in this pa-
per. In particular, error for the individual input PMW
and IR precipitation estimates and the propagation of
the error along the model integration process need to be
accurately quantified. This requires a comprehensive
examination of the error in the level 2 PMW precipita-
tion retrievals as well as that generated in the process
of propagating precipitating systems. The results of this
error analysis will not only lead to improved error def-
inition and thereby the final precipitation analysis, it will
also provide insights to how the integration algorithm
should be refined to reduce the generation and propa-
gation of errors. While this paper describes an algorithm
to construct high-resolution precipitation estimates through
the integration of information from individual sources,
success of such integration techniques is built upon the
production of the input level 2 PMW and IR precipi-
tation estimates with improved quality and refined error
quantification.
A prototype system has been developed at NOAA/
CPC to construct 30-min precipitation estimates on an
8 3 8 km2 grid over the globe from 608S to 608N by in-
tegrating PMW estimates from LEO satellites and IR
observations from GEO platforms through the KF-
CMORPH algorithm described in this paper. Further
work is under way to extend the analysis domain to
cover the entire globe from pole to pole and to remove
the bias in the integrated satellite estimates through
comparison against gauge observations.
Acknowledgments. The authors thank S.-H. Yoo and
Y. Yarosh for their technical support to part of the study
described in this paper. They are grateful to M. Sapiano
who kindly provided the SSM/I-based level 2 data used
in the synthetic experiments for July–August 2009.
Comments from P. A. Arkin, J. Janowiak, R. Ferraro,
G. Huffman, W. Shi, M. Chen, and anonymous reviewers
were invaluable for improvements to the manuscript. This
work is supported by NOAA/Climate Prediction Center
(CPC), NOAA/Climate Program Office (CPO), NOAA/
National Climatic Data Center (NCDC), and NOAA/
USWRP Hydrometeorology Testbed (HMT) as part of
NOAA’s contribution to the NASA Precipitation Mea-
surement Mission (PMM) and GEWEX Global Precipi-
tation Climatology Project (GPCP).
REFERENCES
Adler, R. F., G. J. Huffman, and P. R. Keehn, 1994: Global tropical
rain estimates from microwave-adjusted geosynchronous IR
data. Remote Sens. Rev., 11, 125–152.
Behrangi, A., B. Imam, K. Hsu, S. Sorooshian, T. J. Bellerby, and
G. J. Huffman, 2010: REFAME: Rain estimation using forward-
adjusted advection of microwave estimates. J. Hydrometeor.,
11, 1305–1321.
Dai, A., K. E. Trenberth, and T. R. Karl, 1999: Effects of clouds,
soil moisture, precipitation, and water vapor on diurnal tem-
perature range. J. Climate, 12, 2451–2473.
——, X. Lin, and K. Hsu, 2007: The frequency, intensity, and di-
urnal cycle of precipitation in surface and satellite obser-
vations over low- and mid-latitudes. Climate Dyn., 29, 727–
744.
Ebert, E. E., J. E. Janowiak, and C. Kidd, 2007: Comparison of
near-real-time precipitation estimates from satellite observa-
tions and numerical models. Bull. Amer. Meteor. Soc., 88, 47–64.
Ferraro, R. R., 1997: Special sensor microwave imager derived
global rainfall estimates for climatological applications.
J. Geophys. Res., 102, 16 715–16 735.
——, F. Weng, N. C. Grody, and L. Zhao, 2000: Precipitation
characteristics over land from the NOAA-15 AMSU sensor.
Geophys. Res. Lett., 27, 2669–2672.
Gandin, L. S., 1965: Objective Analysis of Meteorological Fields.
Israel Program for Scientific Translations, 242 pp.
Hong, Y., R. F. Adler, A. J. Negri, and G. J. Huffman, 2007: Flood
and landslide applications of near real-time satellite rainfall
products. Nat. Hazards, 43, 285–294, doi:10.1007/s11069-006-
9106-x.
Hsu, K.-L., X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Pre-
cipitation estimation from remotely sensed information using
artificial neural networks. J. Appl. Meteor., 36, 1176–1190.
Huffman, G. J., and Coauthors, 1997: The Global Precipitation
Climatology Project (GPCP) combined precipitation dataset.
Bull. Amer. Meteor. Soc., 78, 5–20.
——, R. F. Adler, E. F. Stoker, D. T. Bolvin, and E. J. Nelkin, 2003:
Analysis of TRMM 3-hourly multi-satellite precipitation es-
timates computed in both real and post-real time. Extended
Abstracts, 12th Conf. on Satellite Meteorology and Oceanog-
raphy, Seattle, WA, Amer. Meteor. Soc., P4.11. [Available
online at http://ams.confex.com/ams/annual2003/techprogram/
paper_54906.htm.]
——, and Coauthors, 2007: The TRMM multisatellite precipitation
analysis (TMPA): Quasi-global, multiyear, combined-sensor
precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55.
——, R. F. Adler, D. T. Bolvin, and E. J. Nelkin, 2009: The TRMM
Multi-satellite Precipitation Analysis (TMPA). Satellite Rain-
fall Applications for Surface Hydrology, M. Gebremichael and
F. Hossain, Eds., Springer, 3–22.
Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2001: A real-time
global half-hourly pixel-resolution infrared dataset and its
applications. Bull. Amer. Meteor. Soc., 82, 205–217.
Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004:
CMORPH: A method that produces global precipitation es-
timates from passive microwave and infrared data at high
spatial and temporal resolution. J. Hydrometeor., 5, 487–503.
Kalnay, E., 2003: Atmospheric Modeling, Data Assimilation and
Predictability. Cambridge University Press, 364 pp.
Klazura, G. E., and D. A. Imy, 1993: A description of the initial set
of analysis products available from the NEXRAD WSR-88D
system. Bull. Amer. Meteor. Soc., 74, 1293–1312.
1562 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 12
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC
Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified
scheme for obtaining precipitation and vertical hydrometeor
profiles from passive microwave sensors. IEEE Trans. Geosci.
Remote Sens., 34, 1213–1232.
——, and Coauthors, 2000: The status of the Tropical Rainfall
Measuring Mission (TRMM) after two years in orbit. J. Appl.
Meteor., 39, 1965–1982.
——, and Coauthors, 2001: The evolution of the Goddard Profiling
Algorithm (GPROF) for rainfall estimation from passive mi-
crowave sensors. J. Appl. Meteor., 40, 1801–1820.
McPhaden, M. J., and Coauthors, 1998: The Tropical Ocean-
Global Atmosphere observing system: A decade of progress.
J. Geophys. Res., 103, 14 169–14 240.
Olson, W. S., C. D. Kummerow, Y. Hong, and W.-K. Tao, 1999:
Atmospheric latent heating distributions in the tropics derived
from satellite passive microwave radiometer measurements.
J. Appl. Meteor., 38, 633–664.
Purdom, J. F. W., and P. N. Dills, 1994: Cloud motion and height
measurements from multiple satellites including cloud heights
and motions in polar regions. Preprints, Seventh Conf. on
Satellite Meteorology and Oceanography, Monterey, CA,
Amer. Meteor. Soc., 408–411.
Sapiano, M. R. P., T. M. Smith, and P. A. Arkin, 2008: A new
merged analysis of precipitation utilizing satellite and re-
analysis data. J. Geophys. Res., 113, D22103, doi:10.1029/
2008JD010310.
Shen, Y., A. Xiong, Y. Wang, and P. Xie, 2009: Performance of
high-resolution satellite precipitation products over China.
J. Geophys. Res., 115, D02114, doi:10.1029/2009JD012097.
Simpson, J., R. F. Adler, and G. R. North, 1988: A proposed
tropical rainfall measuring mission (TRMM) satellite. Bull.
Amer. Meteor. Soc., 69, 278–295.
Smith, E., and D. Phillips, 1972: Measurements from satellite
platforms. SSEC Annual Satellite Rep. NASS-11542, 53 pp.
Tian, Y., C. D. Peters-Lidard, B. J. Choudhury, and M. Garcia,
2007: Multitemporal analysis of TRMM-based satellite pre-
cipitation products for land data assimilation applications.
J. Hydrometeor., 8, 1165–1183.
Turk, F. J., E. E. Ebert, B.-J. Sohn, H.-J. Oh, V. Levizzani, E. A.
Smith, and R. Ferraro, 2003: Validation of an operational
global precipitation analysis at short time scales. Extended
Abstracts, 12th Conf. on Satellite Meteorology and Oceanog-
raphy, Seattle, WA, Amer. Meteor. Soc., JP1.2. [Available
online at http://ams.confex.com/ams/annual2003/techprogram/
paper_56865.htm.]
——, G. V. Mostovoy, and V. G. Anantharaj, 2010: Soil moisture
sensitivity to NRL-blend high-resolution precipitation
products: Analysis of simulations with two land surface
models. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 3,
32–48.
Ushio, T., and Coauthors, 2009: A Kalman filter approach to the
Global Satellite Mapping of Precipitation (GSMaP) from
combined passive microwave and infrared radiometric data.
J. Meteor. Soc. Japan, 87, 137–151.
Vila, D., R. Ferraro, and R. Joyce, 2007: Evaluation and im-
provement of AMSU precipitation retrievals. J. Geophys.
Res., 112, D20119, doi:10.1029/2007JD008617.
Xie, P., and P. A. Arkin, 1995: An intercomparison of gauge
observations and satellite estimates of monthly precipitation.
J. Appl. Meteor., 34, 1143–1160.
——, and ——, 1996: Analyses of global monthly precipitation
using gauge observations, satellite estimates, and numerical
model predictions. J. Climate, 9, 840–858.
——, and ——, 1997: Global precipitation: A 17-year monthly
analysis based on gauge observations, satellite estimates, and
numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539–
2558.
——, A. Yatagai, M. Chen, T. Hayasaka, Y. Fukushima, C. Liu,
and S. Yang, 2007: A gauge-based analysis of daily precipi-
tation over East Asia. J. Hydrometeor., 8, 607–626.
——, M. Chen, and W. Shi, 2010: CPC unified gauge-based analysis
of global daily precipitation. Preprints, 24th Conf. on Hy-
drology, Atlanta, GA, Amer. Meteor. Soc., 2.3A. [Available
online at http://ams.confex.com/ams/90annual/techprogram/
paper_163676.htm.]
Zhang, J., and Coauthors, 2009: National Mosaic and QPE (NMQ)
system—Description, results and future plans. Preprints, 34th
Conf. on Radar Meteorology, Williamsburg, VA, Amer. Meteor.
Soc., 7A.1. [Available online at http://ams.confex.com/ams/
34Radar/techprogram/paper_155375.htm.]
DECEMBER 2011 J O Y C E A N D X I E 1563
Unauthenticated | Downloaded 04/22/22 11:18 PM UTC