Journal of Hydrology 513 (2014) 446–456

Contents lists available at ScienceDirect

Journal of Hydrology

journal homepage: www.elsevier .com/locate / jhydrol

Performance assessment of the successive Version 6 and Version 7 TMPAproducts over the climate-transitional zone in the southern Great Plains,USA

http://dx.doi.org/10.1016/j.jhydrol.2014.03.0400022-1694/� 2014 Elsevier B.V. All rights reserved.

⇑ Corresponding author at: National Weather Center, Atmospheric RadarResearch Center, Suite 4610, 120 David L. Boren Blvd., Norman, OK 73072-7303,United States. Tel.: +1 405 325 3644.

E-mail address: yanghong@ou.edu (Y. Hong).

Lei Qiao a,b,c, Yang Hong a,b,f,⇑, Sheng Chen a,b, Chris B. Zou c, Jonathan J. Gourley d, Bin Yong e

a School of Civil Engineering and Environmental Science, University of Oklahoma, United Statesb Advanced Radar Research Center, National Weather Center, Norman, OK 73072, United Statesc Department of Natural Resource Ecology and Management, Oklahoma State University, Stillwater, OK 74078, United Statesd NOAA/National Severe Storms Laboratory, Norman, OK 73072, United Statese State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing 210098, Chinaf Department of Hydraulic Engineering, Tsinghua University, Beijing 100084, China

a r t i c l e i n f o

Article history:Received 20 November 2013Received in revised form 30 January 2014Accepted 17 March 2014Available online 1 April 2014This manuscript was handled by AndrasBardossy, Editor-in-Chief, with theassistance of Uwe Haberlandt, AssociateEditor

Keywords:TRMMQuantitative Precipitation EstimationVerdigris River BasinUpper Washita BasinGlobal Precipitation MeasurementTMPA

s u m m a r y

This study assesses the latest version, Version 7 (V7) Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) rainfall estimates by comparison with the previous version, Ver-sion 6 (V6), for both near-real-time product (3B42RT) and post-real-time research products (3B42) overthe climate-transitional zone in the southern Great Plains, USA. Two basins, the Verdigris River Basin(VRB) in the east and the Upper Washita Basin (UWB) in the west, with distinctive precipitation but sim-ilar vegetation and elevation, were selected to evaluate the TMPA products using rain gauge-blendedproducts with WSR-88D NEXRAD Stage IV. This study sheds important insights into the detailed spatio-temporal precipitation errors, and also reveals algorithm performance during extreme events over thetwo low-relief basins within a high precipitation gradient zone. Based on nine years of measurements(2002–2010), this study shows that: (1) 3B42V7 corrects the widespread rainfall underestimation fromresearch product 3B42V6, especially for the drier UWB with relative bias (RB) improvement from�23.24% to 2.24%. (2) 3B42RTV7 reduces the widespread, notable overestimation from the real-timeproduct 3B42RTV6, with minor overestimation in the wet VRB and underestimation in the dry UWB.(3) For both versions of TMPA products, larger root mean square error (RMSE) but higher correlation coef-ficients (CCs) tend to appear for the wet VRB, while lower RMSE and CC mostly occur in the dry UWB.3B42RTV7 shows a drawback that the CC declines significantly, especially in the dry region where it dropsbelow 0.5. (4) Seasonally, autumn rainfall estimations in both versions and basins have the least bias. The3B42RTV6 overestimation and 3B42V6 underestimation of spring and summer rainfall, which dominatethe annual total bias, are significantly reduced for both basins in the V7 products. Winter precipitationestimation improvement is also noticeable with significant RB and RMSE reductions. However, consider-able overestimation in summer rainfall still exists for the wet basin. (5) Although V7 has the overall bestperformance, it still shows deficiency in detecting extreme rainfall events in low-relief regions, tending tounderestimate peak rainfall intensity and to misrepresent timing and locations. Results from this studycan be used for reference in the algorithm development of the next generation of Integrated Multi-Satellite Retrievals for Global Precipitation Measurement (GPM) scheduled to launch in 2014.

� 2014 Elsevier B.V. All rights reserved.

1. Introduction

Quantitative Precipitation Estimation (QPE) products based onsatellite- and ground-based microwave radar retrievals with high

spatiotemporal resolution play an important role in climatic, atmo-spheric and hydrological studies, especially for regions with insuf-ficient rain gauge coverage (Huffman et al., 2001; Nijssen andLettenmaier, 2004; Sorooshian et al., 2011). Prevailing QPEproducts mainly include the Precipitation Estimation fromRemotely Sensed Information Using Artificial Neural Networks(PERSIANN) (Sorooshian et al., 2000), the Climate Prediction Center(CPC) Morphing Technique (CMORPH) (Joyce et al., 2004),

L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456 447

PERSIANN-CCS (Cloud Classification System) (Hong et al., 2004),the Naval Research Laboratory Global Blended-Statistical Precipita-tion Analysis (Turk and Miller, 2005), the Global Satellite Mappingof Precipitation (Kubota et al., 2007; Okamoto et al., 2005), and theTropical Rainfall Measuring Mission (TRMM) Multi-satellite Pre-cipitation Analysis (Huffman et al., 2007).

Uncertainty analyses of satellite precipitation are vital forunderstanding their characteristics for applications in waterresources research and management (Hossain and Anagnostou,2004, 2006; Steiner et al., 2003). A series of studies have been con-ducted recently to evaluate various rainfall products and relatedhydrological applications (Anagnostou, 2007; Ebert et al., 2007;Gourley et al., 2011; Hong et al., 2006; Hossain and Huffman,2008; Huang et al., 2013; Li et al., 2012; Pan et al., 2010;Serpetzoglou et al., 2010; Yong et al., 2010; Zeweldi andGebremichael, 2009). Hossain and Anagnostou (2004), using aprobabilistic error model, investigated the uncertainty of passivemicrowave (PM) and infrared-based (IR) rainfall data andthe potential for flood prediction over small-size basins (at50–500 km2). Their study described the complexity and highuncertainty of peak rainfall across small basin scales and suggestedpotential methods for uncertainty reduction of IR and PM rainfallretrievals. Hong et al. (2006) quantified satellite-based precipita-tion error as a nonlinear function of space–time scale, rainintensity, and sampling frequency (from hourly to daily). Hossain

98°0'0"W98°30'0"W99°0'0"W

36°0'0"N

35°30'0"N

3 5°0'0"N

34°30'0"N

DEM range (m)

0 -269

269 -603

603 -4,280

Value

Fig. 1. Study areas of Upper Washita Basin (a) and Verdigris River Basin (b) with featureare designated as polygons in the elevation (c) and annual precipitation (d) maps.

and Huffman (2008) focused on hydrologically relevant scales(ranging from 0.04� to 1.0� of latitude/longitude) to assess errorsand optimize rainfall estimates based on a two dimensional satel-lite rainfall error model (SREM2D) they developed (Hossain andAnagnostou, 2006). Their study showed that conventional metricssuch as CC was less sensitive to finer spatial scales and proposedan error metric framework to better indentify seasonal and regionaldifferences of satellite rainfall products. Stephens and Kummerow(2007) presented a review about error characteristics from satelliteradiance measurements and highlighted the error sources of radi-ance assimilation and unconstrained atmospheric parameters.

It has been reported that satellite rainfall estimations havemuch lower uncertainty in low-relief terrain than in mountainousregions (Gebregiorgis and Hossain, 2013; Pan et al., 2010;Stampoulis and Anagnostou, 2012; Yong et al., 2010). In otherwords, the high spatiotemporal resolution of satellite rainfall datain relatively flat plains is potentially more applicable for hydrolog-ical modeling and extreme hazard monitoring. TMPA productsinclude the three-hour real-time product 3B42RT and researchproduct 3B42, which are the most commonly used satellite rainfallretrievals and have evolved from Version 6 (V6) to Version 7 (V7)in 2012. Many studies have been carried out to validate 3B42V6and make comparions with other rainfall products (e.g. Gourleyet al., 2011; Huffman et al., 2007; Sapiano and Arkin, 2009).Gourley et al. (2011) evaluated several real-time ground radar

95°15'0"W96°0'0"W96°45'0"W

38°15'0"N

37° 30'0"N

36° 45'0"N

Water

Deciduous tree

Evergree tree

Shrub

Grassland

Pasture/hay

Crop land

Annual Precipitation (mm)

0 -635

635 -889

889 -1143

RANGE

s of National Land Cover Dataset (NLCD) 2006 classification. Geographical locations

448 L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456

and remote sensing algorithms including Next Generation WeatherRadar (NEXRAD), 3B42RTV6 and PERSIANN-CCS. They concludedthat the 3B42RTV6 was better in producing hydrological variablesthan the other two types of rainfall estimation. However their studyonly focused on one small watershed, the Ft. Cobb watershed. Thus,a comprehensive evaluation over a wider spatial domain and longertime period is necessary for determining the proper application ofthe TMPA products in the climate-transitional zone in the southernGreat Plains of the United States.

This study includes and extends the Ft. Cobb watershed byincluding the Verdigris River Basin which has higher annualprecipitation in order to assess performance of the successive V6and V7 TMPA products. The newly available V7 products consistof new parameters for improving accuracy across spatiotemporalscales. In this study, we reveal the systematic and random errorsof both real-time (RT) and research products for two low-relief,intermediate-sized basins located in the climate-transitional zone

Fig. 2. Mean daily rainfall (mm/day) for stage IV (a) and V6/V7 TMPA products (b–e) in th

in the south-central Great Plains. The error characteristics of V6and V7 products are identified and evaluated at a daily temporalscale and 0.125� spatial resolution for both long-term averagesand for individual, rare extreme events. Multiple climate changemodels project that annual precipitation will decrease for Ft. Cobbwatershed but increase for Verdigris River Basin (USGCRP, 2009). Arecent study by Liu et al. (2012) suggested a very good agreementbetween the Coupled Model Intercomparison Project phase 5(CMIP5) simulations and the Global Precipitation ClimatologyProject (GPCP) and TRMM satellite observations. In addition, thisregion is primary grasslands but experiencing rapid land usechange and increase in woody plant cover (Wine and Zou, 2012).The combination of rapid changes in land surface condition andprecipitation regime impose great challenges in understandingand managing responses of water resources in this region (Zouet al., 2013). Therefore knowing the errors of TMPA and reductionof uncertainty of precipitation estimation is pivotal for application

e southern Great Plains. The arrow points out the low estimation patches in 3B42V6.

Fig. 3. Bias (mm) distribution for V6/V7 TMPA products in the southern Great Plains.

L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456 449

of hydrological models and subsequent management of waterresources under increasing climate variability in the southernGreat Plains of the U.S.

2. Research areas, data and methods

2.1. Research area 1 – the Verdigris River Basin

The Verdigris River Basin (VRB) is a trans-state watershed withan area of 11,250 km2, extending northwestward from Oklahomaalong the Verdigris River to Kansas with approximately two thirdsof the basin is in Kansas (Fig. 1a). The Verdigris River flows gener-ally southeast for about 563 km and drains into Oologah Lake, adammed reservoir providing major drinking water for the city ofTulsa, Oklahoma. Elevations within the basin are low and rangefrom 502 m to 207 m from the headwaters to the Oologah Lake(Fig. 1c). The land cover type is dominated by grassland (37%)and pasture/hay (46%) according to the USGS National Land CoverDatabase (NLCD) (Fig. 1a). Annual precipitation in the basin variesfrom 800 mm in the northwest to 1100 mm in the southeast(Fig. 1d). Approximately 70% of this precipitation is accumulatedfrom April to September, which causes streamflow to peak in Julyand then declining through August and September due to acceler-ating ET and human water usage. This study area is relativelywarm and humid, and snowmelt plays a negligible role.

2.2. Research area 2 – the Upper Washita Basin

The Upper Washita Basin (UWB) is located in the semiaridsouthwest of Oklahoma (Fig. 1b). The Ft. Cobb watershed is withinthis basin and it is a USDA – Agricultural Research Service (ARS)designated research site with intensive rain gauge coverage andmonitoring instruments. In addition, initial comparison of different

QPE products was conducted on the Ft. Cobb Basin (Gourley et al.,2011). In this study we extend the study to the whole UWB (anarea of 8310 km2) making its spatial size similar to the VRB. Thiswatershed is quite flat with elevation difference less than 200 m(ranging from 564 to 379 m) (Fig. 1c). The land use and cover isdominated by grassland (41%) and cropland (40%) (Fig. 1b). Annualprecipitation in the basin ranges from 700–800 mm (Fig. 1d) andmost rainfall occurs in spring with around 19% of rainfall accumu-lated in May.

2.3. TMPA precipitation and Stage IV reference

The TMPA includes both near-real-time and post-real-time prod-ucts computed with the V6 and V7 TMPA algorithms. In this paper,the near-real-time products are abbreviated as 3B42RTV6 and3B42RTV7 for the two versions, and the research products areabbreviated as 3B42V6 and 3B42V7 following the nomenclature ofChen et al. (2013). The near-real-time product is solely from satelliteretrievals so that it can be available several hours after the observa-tion. A lower degree of accuracy is expected compared to theresearch products, which have been bias-corrected based onmonthly gauge observations. Multiple sources of passive micro-wave signals are incorporated for the precipitation estimates, suchas the TRMM Microwave Imager (TMI), Special Sensor MicrowaveImager (SSMI), Special Sensor Microwave Imager/Sounder (SSMIS)(3B42V7 only), Advanced Microwave Scanning Radiometer-EOS(AMSR-E), Advanced Microwave Sounding Unit-B (AMSU-B), andMicrowave Humidity Sounder (MHS). For the period of 1998–1999, 3B42V7 also includes the 0.07� Grisat-B1 infrared data, whichrepresents a significant improvement in resolution and areal cover-age over the 1� 24-class histogram infrared data used in V6. Addi-tionally, 3B42V7 incorporates the new Global PrecipitationClimatology Center (GPCC) ‘‘full’’ gauge analysis and the GPCC

Fig. 4. RMSE (mm) distribution for V6/V7 TMPA products in the southern Great Plains.

Fig. 5. Correlation coefficient (CC) distribution for V6/V7 TMPA products in the southern Great Plains. All correlations are significant at the confidence level of 99% (p < 0.01).

450 L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456

Fig. 6. Scatter plots of mean daily precipitation over the Verdigris River Basin (a–d) and Upper Washita Basin (e–h) for V6/V7 TMPA vs. Stage IV. Note: the CC here is spatialcorrelation coefficient of mean daily precipitation of the grids within the two basins.

Fig. 7. Probability density function by occurrence (PDFc) for Verdigris River Basin(a) and Upper Washita Basin (b).

L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456 451

‘‘monitoring’’ gauge analysis since 2010; whereas, the previousmonitoring product covered up to April 2005 and used the ClimateAssessment and Monitoring System (CAMS) analysis thereafter(Huffman et al., 2011). AMSU precipitation has a low bias that isvery important for the oceans. The upgrade of AMSU in 2007included data reprocessing which was used by V7 throughout thewhole period, but V6 did not use the reprocessed data before May,2007 (Prakash et al., 2013). More details about the TMPA improve-ment can be seen in Huffman et al. (2011) and Huffman and Bolvin(2013). All of the TMPA products have a spatiotemporal resolutionof 0.25�/3-h, which have been resampled to 0.125�/daily in thisstudy.

Stage IV is the hourly gauge-adjusted radar (WSR-88D NEXRAD)product developed by the US National Weather Service with a spa-tial resolution of 4 km. This dataset, available since 2002, representsthe highest accuracy for precipitation estimation at this particularscale over the continental United States. Gourley et al. (2011) com-pared the Stage IV and the ARS Micronet (multiple stations acrossthe Ft. Cobb watershed) gauge products and concluded that theStage IV had minimal bias and even outperformed the gauge productin simulating streamflow responses for Ft. Cobb watershed. In thisstudy, we used the Stage IV product as the rainfall reference afterresampling it to the spatiotemporal resolution mentioned above.

2.4. Statistical metrics

Bias (the difference between the TRMM QPE and the Stage IVreference), relative bias (RB, the Bias divided by the reference), rootmean square error (RMSE), and the Pearson linear correlation coef-ficient (CC) are statistics used for evaluation. These metrics aredefined as follows.

Bias ¼ QPE� gauge ð1Þ

RB ¼PðQPE� gaugeÞP

gaugeð2Þ

RMSE ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPðQPE� gaugeÞ2

N

sð3Þ

CC ¼ CovðQPE; gaugeÞrQPErgauge

ð4Þ

where RB and CC are dimensionless, RMSE is in mm. Here, ‘gauge’ isthe Stage IV reference. In Eq. (3), ‘‘Cov ()’’ refers to the covariance,and r indicates the standard deviation. RB, when multiplied by

452 L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456

100, denotes the degree of overestimation or underestimation inpercentage. All the above statistics were computed on a pixel-by-pixel basis.

3. Results and discussion

3.1. Spatial comparison of TMPA rainfall products

Fig. 2 shows the distribution of mean annual rainfall of theTMPA and Stage IV in the southern Great Plains. The overall over-estimation existing in the RT rainfall of 3B42RTV6 has been largelyreduced in 3B42RTV7, especially for the upper VRB (Fig. 2b and d).Apparently, the 3B42V7 rainfall pattern is more similar to the StageIV compared to 3B42V6 (Fig. 2a, c and e). Also the distinct under-estimation zones by 3B42V6, as pointed out with an arrow inFig. 2c, are removed in the latest algorithm of 3B42V7. The spatialerror distribution of TMPA products, in terms of Bias (Fig. 3), RMSE(Fig. 4) and CC (Fig. 5), shows the improvement in V7 over V6 ofTMPA products in the southern Great Plains. 3B42V7 corrects thewidespread rainfall underestimation from 3B42V6, especially forthe relatively dry UWB. 3B42RTV7 corrects the widespread rainfall

Fig. 8. Scatter plots of seasonal mean daily precipitation ove

overestimation from 3B42RTV6. However, minor overestimation inthe wet VRB and underestimation in the dry UWB are still present.Greater RMSE and CC tend to appear for the wet region, whilelower RMSE and CC appear for the dry region. RMSE and CC arebetter for 3B42V7 compared to the 3B42V6 product especially forthe CC in the wet region. Although the error and bias inherent inRT V6 are greatly decreased in RT V7 in most areas, 3B42RTV7shows a clear drawback of significant decrease of CC, especiallyin the dry region where it drops down to less than 0.5.

The scatter plots of TMPA vs. Stage IV for mean daily precipita-tion over the Verdigris River Basin and Upper Washita Basin areshown in Fig. 6a and b respectively. Here, the CC on the plot is spa-tial correlation coefficient of mean daily precipitation of the gridswithin the two basins. From the scatter plot and statistics, wecan see that the RB by 3B42V6 is reduced slightly from �6.8% to3.16% (Fig. 6b and d) for the VRB, but the enhancement for theUWB is very significant, from �23.24% to 2.24% (Fig. 6f and h).The improvement of 3B42RTV7 over 3B42RTV6 is notable for bothbasins but with more substantial improvement for the VRB. Thespatial CC increased from 0.25 to 0.48, RB and RMSE reduced from22.92% to 10.09% and 0.78 mm to 0.36 mm, respectively, for the

r the Verdigris River Basin for V6/V7 TMPA vs. Stage IV.

L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456 453

VRB (Fig. 6a and c). Fig. 7 shows the daily precipitation probabilitydensity function by occurrence (PDFc) for the TMPA products com-pared with the reference precipitation of Stage IV. Here, only gridcells where both the reference and the TMPA precipitation esti-mates are nonzero were selected for computation. It is shown that3B42V7 has the better agreement with Stage IV in daily precipita-tion occurrence, while 3B42V6 underestimated precipitationoccurrences for rates greater than 6 mm/day and overestimatedoccurrences for rates less than 6 mm/day in both basins. This prob-ably is the major reason of the widespread underestimation ofrainfall in the 3B42V6.

3.2. Seasonal comparison of TMPA rainfall products

Seasonal rainfall scatter plots of TMPA vs. Stage IV over the wetVRB are shown in Fig. 8. The improvement of real-time product of3B42RTV7 is more evident, especially for spring, summer and win-ter, whereas the research product of 3B42V7 has the greatestimprovement in autumn. The overestimation during spring

Fig. 9. Scatter plots of seasonal mean daily precipitation ove

(22.5%) and summer (57.03%) in 3B42RTV6 is reduced to 8.42%and 20.14% in 3B42RTV7, respectively. More importantly, the win-ter precipitation is corrected considerably by 3B42RTV7 (CCincreased from �0.3 to 0.62). The research products for springand summer tend to have a trade-off in improvement where biasreduction will also reduce the CC, or vice versa. However, theimprovement for autumn and winter are consistent with respectto all statistical indices of RB, CC and RMSE.

Fig. 9 shows the seasonal rainfall scatter plots of TMPA vs. StageIV over the dry UWB. Generally, the change pattern is similar withthe wet basin; the improvement is mostly evident in the spring,summer and winter precipitation in the V7 products. However,bias reductions in spring and summer rainfall for both researchand real-time products are very notable for the UWB. The overes-timation during spring (47.3%) and summer (36.7%) in 3B42RTV6 isreduced to 3.49% and �1.27% in 3B42RTV7, respectively. Theunderestimation during spring (�25%) and summer (�22%) in3B42V6 is reduced to 0.61% and 1.69% in 3B42V7, respectively.As mentioned above, most rainfall in this basin occurs in spring

r the Upper Washita Basin for V6/V7 TMPA vs. Stage IV.

454 L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456

with around 19% accumulated in May, therefore, properly estimat-ing spring rainfall definitely increases the accuracy of total annualrainfall. Relatively speaking, the improvement in V7 for theautumn and winter precipitation is not very significant, perhapsbecause of the lower precipitation amounts and intensity(�2 mm/day) compared to the larger spring and summer rainfall(3–4 mm/day).

3.3. Extreme-rainfall retrievals with TMPA

Oklahoma and nearby states experienced a very wet year in2007. From June to August there were a series of heavy rainfallevents setting records for consecutive rainfall days and wettestmonth on record. During June 26–30, 2007, the Verdigris RiverBasin experienced very heavy rainfall conditions, which lastedalmost one week and culminated in more than 180 mm/day ofrainfall locally at the upper-most part of the basin (Fig. 10). Recordstreamflows were set following the rainfall. River water breachedlevees and flowed to the floodplain, forcing up to 3000 people toevacuate their homes. Economic damages were nearly 40 millionU.S. dollars. In Kansas, crude oil tanks were displaced from therefinery and caused environmental pollution. On August 19,2007, Tropical Storm Erin’s path crossed the UWB watershedproducing 187 mm of rainfall in three hours, which caused

Fig. 10. Spatial patterns of the storm rainfall accumulation (mm) during June 26–30, 200(e), and the variation of basin-averaged daily rainfall from 06/23/2007 to 07/03/2007 (f

extreme flooding, loss of 5 lives, and extensive property damage.In this section, we focus on these two abnormal events to examinethe efficacy of the TMPA rainfall algorithms for extreme floodingevents in the low-relief basins.

Fig. 10(a–e) shows the spatial patterns of the storm rainfall inthe VRB during June 26–30, 2007 measured by Stage IV,3B42RTV6, 3B42V6, 3B42RTV7 and 3B42V7, respectively. BothTMPA and Stage IV captured the precipitation relative maximumin the upper part of the basin, but there was significant underesti-mation by the TMPA products. The daily variation of basin-aver-aged rainfall throughout the event (Fig. 10f) shows that the StageIV reveals the heaviest rainfall (nearly 85 mm basin-wide) on June29, 2007, whereas the TMPA products yield peak rainfall one day inadvance of Stage IV with an underestimation by 20 mm or more(Fig. 10f). Although the 3B42V7 rainfall amount is closest to theStage IV, its real-time product (3B42RTV7) has significant underes-timation. In this case, rainfall estimates of 3B42RTV6 and 3B42V6are almost equal, which are better than 3B42RTV7.

Fig. 11(a–e) shows the spatial patterns of the storm rainfallaccumulation in the UWB from August 15–22, 2007 measured byStage IV, 3B42RTV6, 3B42V6, 3B42RTV7 and 3B42V7, respectively.3B42RTV7 underestimated this extreme rainfall intensity remark-ably, which makes 3B42RTV6 perform better in precipitationintensity estimation. However the centroid of the maximum rain-

7, measured by Stage IV (a), 3B42RTV6 (b), 3B42V6 (c), 3B42RTV7 (d), and 3B42V7) in the Verdigris River Basin.

Fig. 11. Spatial patterns of the storm rainfall accumulation (mm) during August 15–22, 2007, measured by Stage IV (a), 3B42RTV6 (b), 3B42V6 (c), 3B42RTV7 (d), and 3B42V7(e), and the variation of basin-averaged daily rainfall from 08/15/2007 to 08/25/2007 (f) in the Upper Washita Basin.

L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456 455

fall is spatially offset according to both versions of real-time prod-ucts. On the other hand, both 3B42RTV6 and 3B42RTV7 productsdetected the right timing for the maximum rainfall day in this case(Fig. 11f), in contrast to the rainfall event that took place on theVRB. For research products, the 3B42V7 outperforms the 3B42V6in spite of still underestimating the maximum daily rainfall by30%. The research products estimate the peak rainfall on August18, which is again one day earlier than the Stage IV peak on August19 (Fig. 11f).

4. Conclusion

In this study, we assessed the successive V6 and V7 TMPA rain-fall products over the climate-transitional zone in the southernGreat Plains, USA. The systematic and random errors for bothreal-time and research rainfall products were evaluated for thetime period from 2002 to 2010 in terms of multi-year mean, sea-sonal variation and extreme, rare events.

Our results suggest that satellite rainfall retrieval uncertaintiesdepend on seasons and climate regimes. Specifically, the studyshows that (1) 3B42V7 corrects the widespread rainfall underesti-mation from 3B42V6, and this improvement is more pronouncedfor the relatively dry basin. (2) 3B42RTV7 corrects the widespreadoverestimation from 3B42RTV6. However, minor overestimation inthe wet region and underestimation in the dry region tends to stillbe present. (3) For both versions of TMPA products, greater rootmean square error (RMSE) but higher correlation coefficients(CCs) tend to appear for the wet region, while lower RMSE andCC appear for the dry region. 3B42RTV7 shows a clear drawbackin that the CC declines significantly, especially in the dry regionwhere it drops down less than 0.5. (4) Seasonally, autumn rainfallestimation in both versions and basins has the least bias. The3B42V6RT overestimation and 3B42V6 underestimation of springand summer rainfall, dominating the annual total bias, are signifi-cantly reduced for both areas in the V7 versions. Winter precipita-tion estimation improvement is also noticeable with significant RB

and RMSE reductions. However, considerable overestimation insummer rainfall is still present in the wet basin. (5) Substantialimprovement is still needed for the real-time TMPA rainfall prod-ucts before they can be used to identify extreme rainfall centersand their intensities, even for a basin with relatively low reliefand elevation.

The South-central Great Plains is located in the climate-transi-tional zone with drastic changes in hydroclimatology and vegeta-tion cover from the arid west towards much more humidconditions in the east. Improvement in TMPA precipitation prod-ucts and further evaluation of those products and their perfor-mance in hydrological models are needed before they can beused for management of water resources, flash flood warning, floodcontrol, and assessment of changes in ecosystem goods and ser-vices under changing precipitation regimes.

Acknowledgements

We acknowledge the TRMM mission scientists and associatedNASA personnel for the production of the data used in this researcheffort, and are very much indebted to the team responsible for theTMPA products, especially Dr. George J. Huffman. Thanks also go toDr. Huilin Gao of the Texas A&M University. This work was finan-cially supported by the NOAA Multi-function Phased-Array RadarProject and the Responses to Climate Change program, U.S. ArmyCorps of Engineers Institute for Water Resources and the SouthCentral Climate Science Center.

References

Anagnostou, E.N., 2007. Assessment of satellite rain retrieval error propagation inthe prediction of land surface hydrology. In: Levizzani, V., Bauer, P., Turk, F.J.(Eds.), Measuring Precipitation From Space: EURAINSAT and the Future.Springer-Verlag, Heidelberg, pp. 357–368.

Chen, S., Hong, Y., Gourley, J.J., Huffman, G.J., Tian, Y., Cao, Q., Kirstetter, P.E., Hu, J.,Hardy, J., Khan, S., Xue, X., Yong, B., 2013. Evaluation of the successive V6 andV7TRMM multisatellite precipitation analysis over the Continental UnitedStates. Water Resour. Res. 49, 8174–8186.

456 L. Qiao et al. / Journal of Hydrology 513 (2014) 446–456

Ebert, E.E., Janowiak, J.E., Kidd, C., 2007. Comparison of near-real-time precipitationestimates from satellite observations and numerical models. Bull. Am.Meteorol. Soc. 88 (1), 47–64.

Gebregiorgis, A.S., Hossain, F., 2013. Understanding the dependence of satelliterainfall uncertainty on topography and climate for hydrologic modelsimulation. IEEE Trans. Geosci. Remote Sens. 51 (1), 704–718.

Gourley, J.J., Hong, Y., Flamig, Z.L., Wang, J., Vergara, H., Anagnostou, E.N., 2011.Hydrologic evaluation of rainfall estimates from radar, satellite, gauge, andcombinations on Ft. Cobb Basin, Oklahoma. J. Hydrometeorol. 12 (5), 973–988.

Hong, Y., Hsu, K.L., Sorooshian, S., Gao, X., 2004. Precipitation estimation fromremotely sensed imagery using an artificial neural network cloud classificationsystem. J. Appl. Meteorol. 43 (12), 1834–1853.

Hong, Y., Hsu, K.L., Moradkhani, H., Sorooshian, S., 2006. Uncertainty quantificationof satellite precipitation estimation and Monte Carlo assessment of the errorpropagation into hydrologic response. Water Resour. Res. 42 (8), W08421.

Hossain, F., Anagnostou, E.N., 2004. Assessment of current passive-microwave-andinfrared-based satellite rainfall remote sensing for flood prediction. J. Geophys.Res. 109 (D7), D07102.

Hossain, F., Anagnostou, E.N., 2006. A two-dimensional satellite rainfall error model.IEEE Trans. Geosci. Remote Sens. 44 (6), 1511–1522.

Hossain, F., Huffman, G.J., 2008. Investigating error metrics for satellite rainfall dataat hydrologically relevant scales. J. Hydrometeorol. 9 (3), 563–575.

Huang, Y., Chen, S., Cao, Q., Hong, Y., Wu, B., Huang, M., Qiao, L., Zhang, Z., Li, Z., Li,W., Yang, X., 2013. Evaluation of version-7 TRMM multi-satellite precipitationanalysis product during the Beijing extreme heavy rainfall event of 21 July2012. Water 6 (1), 32–44.

Huffman, G.J., Bolvin, D.T., 2013. TRMM and other data precipitation data setdocumentation. Laboratory for Atmospheres, NASA Goddard Space Flight Centerand Science Systems and Applications.

Huffman, G.J., Adler, R.F., Morrissey, M.M., Bolvin, D.T., Curtis, S., Joyce, R.,McGavock, B., Susskind, J., 2001. Global precipitation at one-degree dailyresolution from multisatellite observations. J. Hydrometeorol. 2 (1), 36–50.

Huffman, G.J., Bolvin, D.T., Nelkin, E.J., Wolff, D.B., Adler, R.F., Gu, G., Hong, Y.,Bowman, K.P., Stocker, E.F., 2007. The TRMM multisatellite precipitationanalysis (TMPA): quasi-global, multiyear, combined-sensor precipitationestimates at fine scales. J. Hydrometeorol. 8 (1), 38–55.

Huffman, G.J., Bolvin, D.T., Nelkin, E.J., Adler, R.F., 2011. Highlights of version 7TRMM multi-satellite precipitation analysis (TMPA). In: Klepp, C., Huffman, G.J.,(Eds.), 5th Internat. Precip. Working Group Workshop, Workshop Program andProceedings, 11–15 October 2010, Hamburg, Germany. Reports on Earth Sys.Sci., 100/2011, Max-Planck-Institut für Meteorologie. pp. 109–110.

Joyce, R.J., Janowiak, J.E., Arkin, P.A., Xie, P., 2004. CMORPH: a method thatproduces global precipitation estimates from passive microwave and infrareddata at high spatial and temporal resolution. J. Hydrometeorol. 5 (3), 487–503.

Kubota, T., Shige, S., Hashizume, H., Aonashi, K., Takahashi, N., Seto, S., Takayabu,Y.N., Ushio, T., Nakagawa, K., Iwanami, K., Kachi, M., Okamoto, K., 2007. Globalprecipitation map using satellite-borne microwave radiometers by the GSMaPproject: production and validation. IEEE Trans. Geosci. Remote Sens. 45 (7),2259–2275.

Li, X.H., Zhang, Q., Xu, C.Y., 2012. Suitability of the TRMM satellite rainfalls indriving a distributed hydrological model for water balance computations inXinjiang catchment, Poyang lake basin. J. Hydrol. 426, 28–38.

Liu, C., Allan, R.P., Huffman, G.J., 2012. Co-variation of temperature and precipitationin CMIP5 models and satellite observations. Geophys. Res. Lett. 39 (13), L13803.

Nijssen, B., Lettenmaier, D.P., 2004. Effect of precipitation sampling error onsimulated hydrological fluxes and states: anticipating the global precipitationmeasurement satellites. J. Geophys. Res.: Atmos. 109 (D2), D02103.

Okamoto, K., Iguchi, T., Takahashi, N., Iwanami, K., Ushio, T., 2005. The globalsatellite mapping of precipitation (GSMaP) project. In: Proc. 25th Int.Geoscience and Remote Sensing Symp., Seoul, South Korea, Institute ofElectrical and Electronics Engineers. pp. 3414–3416.

Pan, M., Li, H., Wood, E., 2010. Assessing the skill of satellite-based precipitationestimates in hydrologic applications. Water Resour. Res. 46 (9), W09535.

Prakash, S., Mahesh, C., Gairola, R.M., 2013. Comparison of TRMM multi-satelliteprecipitation analysis (TMPA)-3B43 version 6 and 7 products with rain gaugedata from ocean buoys. Remote Sens. Lett. 4 (7), 677–685.

Sapiano, M.R.P., Arkin, P.A., 2009. An intercomparison and validation of high-resolution satellite precipitation estimates with 3-hourly gauge data. J.Hydrometeorol. 10 (1), 149–166.

Serpetzoglou, E., Anagnostou, E.N., Papadopoulos, A., Nikolopoulos, E.I., Maggioni,V., 2010. Error propagation of remote sensing rainfall estimates in soil moistureprediction from a land surface model. J. Hydrometeorol. 11 (3), 705–720.

Sorooshian, S., Hsu, K.L., Gao, X.G., Gupta, H.V., Imam, B., Braithwaite, D., 2000.Evaluation of PERSIANN system satellite-based estimates of tropical rainfall.Bull. Am. Meteorol. Soc. 81 (9), 2035–2046.

Sorooshian, S., AghaKouchak, A., Arkin, P., Eylander, J., Foufoula-Georgiou, E.,Harmon, R., Hendrickx, J.M.H., Imam, B., Kuligowski, R., Skahill, B., Skofronick-Jackson, G., 2011. Advanced concepts on remote sensing of precipitation atmultiple scales. Bull. Am. Meteorol. Soc. 92 (10), 1353–1357.

Stampoulis, D., Anagnostou, E.N., 2012. Evaluation of global satellite rainfallproducts over continental europe. J. Hydrometeorol. 13 (2), 588–603.

Steiner, M., Bell, T.L., Zhang, Y., Wood, E.F., 2003. Comparison of two methods forestimating the sampling-related uncertainty of satellite rainfall averages basedon a large radar dataset. J. Climate 16 (22), 3759–3778.

Stephens, G.L., Kummerow, C.D., 2007. The remote sensing of clouds andprecipitation from space: a review. J. Atmos. Sci. 64 (11), 3742–3765.

Turk, F.J., Miller, S.D., 2005. Toward improved characterization of remotely sensedprecipitation regimes with MODIS/AMSR-E blended data techniques. IEEETrans. Geosci. Remote Sens. 43 (5), 1059–1069.

United States Global Change Research Program (USGCRP), 2009. Global ClimateChange Impacts in the United States. In: Karl, T.R., Melillo, J.M., Peterson T.C.,(Eds.). Cambridge University Press, New York, NY, USA.

Wine, M.L., Zou, C.B., 2012. Long-term streamflow relations with riparian galleryforest expansion into tallgrass prairie in the Southern Great Plains, USA. ForestEcol. Manage. 266, 170–179.

Yong, B., Ren, L.L., Hong, Y., Wang, J.H., Gourley, J.J., Jiang, S.H., Chen, X., Wang, W.,2010. Hydrologic evaluation of Multisatellite Precipitation Analysis standardprecipitation products in basins beyond its inclined latitude band: a case studyin Laohahe basin, China. Water Resour. Res. 46 (7), W07542.

Zeweldi, D.A., Gebremichael, M., 2009. Evaluation of CMORPH precipitationproducts at fine space-time scales. J. Hydrometeorol. 10 (1), 300–307.

Zou, C.B., Turton, D.J., Will, R.E., Engle, D.M., Fulendorf, S.D., 2013. Alteration ofhydrological processes and streamflow with juniper (Juniperus virginiana)encroachment in a mesic grassland catchment. Hydrol. Process.. http://dx.doi.org/10.1002/hyp.10102.