Embed Size (px)
Citation preview
Available online at www.sciencedirect.com
www.elsevier.com/locate/asr
ScienceDirect
Advances in Space Research 54 (2014) 670–684
Evaluation of three high-resolution satellite precipitationestimates: Potential for monsoon monitoring over Pakistan
Sadiq Ibrahim Khan a,b,⇑, Yang Hong a,b, Jonathan J. Gourley c,Muhammad Umar Khan Khattak d, Bin Yong e, Humberto J. Vergara a,b
a School of Civil Engineering and Environmental Sciences, The University of Oklahoma, OK, USAb Advanced Radar Research Center, The University of Oklahoma, Norman, OK, USA
c National Oceanic and Atmospheric Administration/National Severe Storms Laboratory, National Weather Center, Norman, OK, USAd Institute of Geographical Information Systems, National University of Sciences and Technology, Islamabad 44000, Pakistane State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing 210098, China
Received 21 September 2012; received in revised form 14 December 2013; accepted 24 April 2014Available online 9 May 2014
Abstract
Multi-sensor precipitation datasets including two products from the Tropical Rainfall Measuring Mission (TRMM) Multi-satellitePrecipitation Analysis (TMPA) and estimates from Climate Prediction Center Morphing Technique (CMORPH) product were quanti-tatively evaluated to study the monsoon variability over Pakistan. Several statistical and graphical techniques are applied to illustrate thenonconformity of the three satellite products from the gauge observations. During the monsoon season (JAS), the three satellite precip-itation products captures the intense precipitation well, all showing high correlation for high rain rates (>30 mm/day). The spatial andtemporal satellite rainfall error variability shows a significant geo-topography dependent distribution, as all the three products overes-timate over mountain ranges in the north and coastal region in the south parts of Indus basin. The TMPA-RT product tends to over-estimate light rain rates (approximately 100%) and the bias is low for high rain rates (about ±20%). In general, daily comparisons from2005 to 2010 show the best agreement between the TMPA-V7 research product and gauge observations with correlation coefficient valuesranging from moderate (0.4) to high (0.8) over the spatial domain of Pakistan. The seasonal variation of rainfall frequency has largebiases (100–140%) over high latitudes (36N) with complex terrain for daily, monsoon, and pre-monsoon comparisons. Relatively lowuncertainties and errors (Bias ±25% and MAE 1–10 mm) were associated with the TMPA-RT product during the monsoon-dominatedregion (32–35N), thus demonstrating their potential use for developing an operational hydrological application of the satellite-based nearreal-time products in Pakistan for flood monitoring.� 2014 COSPAR. Published by Elsevier Ltd. All rights reserved.
Keywords: Satellite precipitation; TRMM; CMORPH; Pakistan monsoon
1. Introduction
In Pakistan major floods are caused by the summermonsoon, which have a long track stretching from the ocean
http://dx.doi.org/10.1016/j.asr.2014.04.017
0273-1177/� 2014 COSPAR. Published by Elsevier Ltd. All rights reserved.
⇑ Corresponding author at: School of Civil Engineering and Environ-mental Sciences, University of Oklahoma, Atmospheric Radar ResearchCenter, National Weather Center, 120 David L. Boren Blvd., Suite 3605,Norman, OK 73072, USA.
E-mail address: [email protected] (S.I. Khan).
to the convection near the Himalayan foothills. The 2010floods were the worst natural disasters in the history of Paki-stan killing and injuring nearly 5000 people directly, displac-ing as many as 20 million and inundating 20% of thecountry. In the scientific community, these events generatedan interesting debate on monitoring the extreme amount ofmonsoon precipitation that led to the flooding (Houze et al.,2011; Wang et al., 2011; Webster et al., 2011). It has beendiscussed that satellite precipitation and weather forecast
S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684 671
models can be used to monitor and predict the developmentand movement of such convective precipitation systems(Houze et al., 2011; Webster et al., 2011). However, theargument that the 2010 floods could have been predictedis contingent on the accuracy of the precipitation estimatesas well as the appropriate use of physically based distributedhydrologic models. An in-depth analysis has not beenperformed on the accuracy and uncertainties of near realtime satellite precipitation estimates to predict and monitorthe monsoon flooding in Pakistan. In this context, weattempt to investigate the error characteristics of threewidely-used high resolution satellite precipitation productsover Pakistan.
The monsoon system in Pakistan occurs mainly due toland-sea temperature differences inciting moisture-ladenfluxes from the Bay of Bengal and the Arabian Sea result-ing in low pressure and convection. The moist air movesnorthward, passing through the lower Indus plain andencounters the relief of the Hindu Kush-Himalayas result-ing in enhanced orographic effects and high rainfall rates.Pakistan receives precipitation during two seasons: (i)heavy summer rains in the northeastern part from mon-soon currents; and (ii) late winter and early spring rainsdue to disturbances in the mid-latitude westerlies (Khan,1993; Wang et al., 2011). Accurate precipitation estimatesduring monsoon season (JAS) is crucial in rainfall runoffmodeling for flood monitoring and prediction. Traditionaldense rain gauge networks have been used for flood moni-toring and other hydrologic studies but their distributionsare often sparse in developing countries and data availabil-ity over mountainous terrain is inadequate (Astin, 1997;Griffith et al., 1978; Huffman et al., 2001; Maddox et al.,2002; Margulis, 2001; Simpson et al., 1996; Steiner et al.,2003; Vicente et al., 1998). Contrary to ground based mea-surements, precipitation estimates from space withhigh-spatiotemporal resolution and large areal coveragecan supplement the existing precipitation estimates forhydrological models in regions where conventional in situprecipitation measurements are not readily available(Semire et al., 2012; Su et al., 2008; Yong et al., 2012).Thus, satellite-based precipitation estimates provide a goodrepresentation of spatial rainfall patterns and have beencomplementary to the ground-based rain gauge and radarmeasurements.
The leap forward in precipitation estimation from spaceradar began in 1997 with the introduction of the TropicalRainfall Measuring Mission (TRMM). The Ku-bandTRMM Precipitation Radar (PR) is the first operationalspace borne radar that is exclusively used for precipitationestimation (Kawanishi et al., 2000; Kummerow et al., 1998;Simpson et al., 1996). The TRMM PR has enabled theglobal mapping of rainfall in the tropics and has contrib-uted to the increased physical understanding of stormcloud characteristics accompanying various forms andlevels of rainfall rates. Since then, numerous high resolu-tion, quasi-global and near real-time satellite precipitationalgorithms have been developed and satellite precipitation
products are readily available for use over the internet(Hong et al., 2007; Huffman et al., 2007; Joyce et al.,2004; Kubota et al., 2007; Sorooshian et al., 2000; Turkand Miller, 2005). Among the fine resolution satellite pre-cipitation data products comprises the Tropical RainfallMeasuring Mission (TRMM) Multi-satellite PrecipitationAnalysis (TMPA) products (Huffman et al., 2007), theNaval Research Laboratory (NRL) Global statisticalprecipitation product (Turk and Miller, 2005), the ClimatePrediction Center (CPC) morphing algorithm (CMORPH)(Joyce et al., 2004), the Passive Microwave CalibratedInfrared (PMIR) algorithm (Kidd et al., 2003), thePrecipitation Estimation from Remotely Sensed Informa-tion using Artificial Neural Networks (PERSIANN)(Sorooshian et al., 2000), and the PERSIANN Cloud Clas-sification System (Hong et al., 2004). The advancement ofthese estimation methods throughout the last decade is thecornerstone for the future Global Precipitation Measure-ment (GPM) mission (Smith et al., 2007), that will belaunched in June 2014.
The instantaneous availability and wide-rangingcoverage of satellites data offers invaluable precipitationestimates especially over complex terrain and ungaugedregions that lack adequate surface-based observations.However, these datasets with their global availability andspatiotemporal advantages might come with limitationson the accuracy of quantitative precipitation estimates.Moreover, the use of space and ground based multi-sensorrainfall estimates for flood monitoring and predictioncannot be achieved without understanding the errors char-acteristics of these estimates. Previous studies assessed theaccuracy of various satellite rainfall products by comparinggauge data around the world. At present, the mostextensive error studies focus on the continental UnitedStates (Ebert et al., 2007; Habib et al., 2009; Hossain andHuffman, 2008). Studies in other continental regionsinclude comparisons over several basins in South Africaby Su et al. (2008), over the La Plata basin in South Amer-ica, and (Dinku et al., 2011; Li et al., 2009) conducted acomprehensive evaluation of ten different satellite estimatesusing gauge observations over East Africa. Anotherdetailed comparison of several operational precipitationestimates has been performed by Andermann et al.(2011), Bookhagen and Burbank (2010) and Islam et al.(2010) while (Yin et al., 2008; Yong et al., 2010) lookedat complex terrains in Himalaya region in South Asiaand Tibetan Plateau in China. More recently, Tian andPeters-Lidard (2010) evaluated an ensemble of six satelliteprecipitation products including TRMM-V7, TRMM-RTand CMORPH and provided a global view of the errorcharacteristics of these estimates.
Realistic depiction of temporal and spatial rainfall esti-mates associated with intense rainfall is central for improve-ment of operational flood monitoring techniques and earlywarning strategies. Therefore, it is critical to evaluate theperformance of satellite precipitation products for any floodmodeling for monsoon systems. Several studies performed
672 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
evaluations of satellite precipitation for hydrologic applica-tions in different countries around the world (Barros et al.,2000; Behrangi et al., 2011; Gourley et al., 2010; Islamet al., 2010; Yong et al., 2012); and similar efforts are at leastequally needed for multi-sensor precipitation comparison inflood-prone Pakistan. In particular, information on theaccuracy related with satellite estimates is often lacking,especially for reliable monsoonal precipitation estimates inflood prone areas that are affected by intense precipitationat high latitudes and mountainous regions.
Consequently, specific objectives of this paper are to (i)assess the uncertainty of these estimates and quantify theerrors associated to the three high-resolution multi sensorproducts in Pakistan (Fig. 1) with varying precipitation cli-matology; (ii) offer an insight into the spatiotemporal errorstructure of these datasets and how much they differ duringmonsoon, pre-monsoon and daily scales; and (iii) investi-gate the geo-topography dependent distribution patternand variation of monsoonal rainfall for the catastrophic2010 flood. This study will provide insights into the futurepossibility to integrate these satellite precipitation productsinto rainfall runoff models for development of an opera-tional flood prediction system in disaster-prone regions ofPakistan. The manuscript is organized into the followingthree main sections: Section 2 provides a general back-ground of satellite precipitation observations, Section 3
Fig. 1. Map of Pakistan with topography (m) and the PMD stations (numbeproducts for 2005–2010 summer monsoon season.
focuses on the satellite precipitation comparison methodol-ogy, Section 4 discusses the results and the last section con-cludes the paper with a summary of the findings.
2. Datasets
2.1. Satellite based precipitation estimates
Satellite precipitation estimates were quantitatively eval-uated over Pakistan, located within the TMPA product lat-itude band ranging from 24�N to 37�N (Fig. 1). The threeprecipitation data sets used to assess the precipitation var-iability are the TMPA real-time version of 3B42 (hereafterTMPA-RT), post real time research product of 3B42 (here-after TMPA-V7) (Huffman et al., 2011) and the NationalOceanic and Atmospheric Administration (NOAA) Cli-mate Prediction Center Morphing Technique (CMORPH)precipitation product (Joyce et al., 2004). These standardhigh-resolution products have been used by the researchcommunity for numerous evaluation studies in different cli-matic conditions around the world. In this manuscript theabove mentioned products were quantitatively evaluatedagainst measurements from rain gauges over upper andlower Indus basin, Jhelum and Ravi basin in Pakistan(Fig. 1). In this study, the daily precipitation data at a0.25� � 0.25� latitude/longitude spatial scale are used.
red), used to compare the TRMM and CMORPH satellite precipitation
S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684 673
2.2. Synoptic gauge network data
Pakistan extends from approximately 61–77�E longitudeand 23.5–37�N latitude and endures diverse climaticregimes. The altitude ranges from sea level with very dryand hot weather in the south to very cold, snow-coveredterrain to the north on the world’s highest mountain peakswith a maximum elevation of about 7000 meters (Fig. 1).Due to diversified climates, the precipitation regime differssignificantly in time and space. Mean annual precipitationvaries from approximately 300 mm in the south to about1300 mm in the north. To measure these climatic condi-tions, ground-based instruments have been recording datasince the existence of Pakistan, while many other gaugedsites were installed later. However, the low spatial and tem-poral resolution of rain gauge networks in mountainouscatchments is inadequate for hydrologic applications. Inthis study we used the daily precipitation data of twenty-four gauged stations (Fig. 1) obtained from the PakistanMeteorological Department (PMD).
The gauge network data recordings are subject to signif-icant sources of error, both from the recording process andthe instruments themselves. First, the original data pro-vided by the PMD was subjected to manual and visualinspection for quality control, followed by corrections toany irregularities in the data record. Daily rainfall accumu-lation time series for each gauge was plotted to screen outthe corrupt data.
3. Methodology
3.1. Validation statistical indices
To quantitatively evaluate the three satellite precipita-tion datasets are compared to rain gauge network observa-tions, we used several statistical indices that are listed inTable 1 and described below. Pearson correlation coeffi-cient (CC) is used to assess the scale of agreement thatreflects the level of linear correlation. Moreover, four sta-tistical indices are used for validation to check for errorand bias between satellite precipitation and gauge observa-tions. The root mean square error (RMSE) evaluates
Table 1Validation statistical indices used to compare the TRMM-based precipitation
Statistical index/unit Equation
Correlation coefficient (CC)CC ¼ ffiffiPq
Root mean squared error (RMSE)/mm RMSE ¼
Mean error (ME)/mm ME ¼ 1n
P
Mean absolute error(MAE)/mm MAE ¼ 1n
Relative Bias (BIAS) /% BIAS ¼P
Notation. n: number of samples; PSati: satellite precipitation (e.g. TMPA-RT
magnitude of mean error and a more simplistic index themean error (ME) determines the mean variance relatingto the estimated and observed data whereas, the meanabsolute error (MAE) shows mean magnitude of the error.While, RMSE also calculates the magnitude of error butwith more emphasis on large errors as compare to meanabsolute error. The relative bias (BIAS) provides informa-tion on the magnitude of difference between the two data-sets. We used this index to quantify the systematic error inthe satellite precipitation data. The rainfall features forinstance the relative contribution to the total seasonal rain-fall and the percentage of rain occurrence in different inten-sity classes have been discussed in detail for pre-monsoonand monsoon seasons.
4. Quantitative analysis of satellite precipitation
Using the rain rate from daily satellite precipitationproducts, we studied the error structures of the three satel-lite products in relation to the elevation and latitude bands.
4.1. Comparisons at daily time scale
With the intention to understand the comparability ofthe space-based precipitation estimates with the rain gaugedata, the sub daily satellite rain rate is converted into dailyrainfall, which is the same temporal resolution as the gaugenetwork data. Daily observed rain gauge data for all sta-tions are pooled together for the entire study period(2005–2010) to find the overall correlation and error statis-tics of TMPA-V7, TMPA-RT and CMORPH precipitationin scatter plots (Fig. 2). The error statistics are shown at thelower right side of Fig. 2. It was found that TMPA-V7underestimated gauge observations by only 4% with amoderate correlation of 0.5 (Fig. 2(a)), while TMPA-RTand CMORPH systematically overestimated gauge obser-vations by 18% and 9% with a correlation of 0.46 and0.48 (Fig. 2(b) and (c)), respectively. Fig. 3 shows statisticalerror characteristics calculated from the daily comparisons.Over the monsoon region (32–35 N), the correlation coeffi-cient (CC) ranged from 0.5–0.6 (Fig. 3(a)) TMPA-RTtends to overestimate in the lower flat regions as well as
products and the gauged observations.
Perfect valuePn
i¼1ðPGaui�PGauÞðPSati�PSatÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n
i¼1ðPGaui�PGauÞ2�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPn
i¼1ðPSati�PSatÞ2
q 1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1n
Pni¼1ðPSati � PGauiÞ2
q0
ni¼1ðPSati � PGauiÞ 0Pn
i¼1jPSati � PGauij 0n
i¼1ðPSati�PGauiÞPn
i¼1PGaui
� 100% 0
and TMPA-V7); PGaui: gauged observation.
Fig. 2. Scatterplots of the grid-based precipitation comparison at the 24 selected stations between (a) daily TMPA-V7 and gauge, (b) daily TMPA-RT andgauge (c) daily CMORPH and gauge. Gauge Mean = 1.69 mm.
674 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
in the upper Indus basin (Fig. 3(b)). In general, TMPA-V7showed lower error than TMPA-RT and CMORPH(Fig. 3(c–e)).
The rainfall features expressed as the percentage ofrelative contribution to the total rainfall and the rainoccurrence in different intensity classes have been studiedin detail. The high intensity classes may be the leadingtrigger for the river floods in Pakistan. In the high intensityclasses (>25 mm day�1), the rainfall occurrence is about40%. Fig. 4(a) shows that TMPA-RT, TMPA-V7 andCMORPH overestimated for the low rainfall intensityrange (less than 1 mm day�1 to 10 mm day�1) Fig. 4(a).Interestingly, the satellite products slightly underestimatedat a higher rainfall classes (>30 mm day�1). The incidenceof satellite precipitation is approximately 45%, which isslightly lower than that indicated by the gaugeobservations.
Fig. 4(b) shows that TMPA-V7 matched with theground gauge observations as compared to TMPA-RTfor the high precipitation intensity range of more than45 mm day�1, while both the two products underestimate.All the three satellite products slightly underestimated
gauged observed precipitation, mainly associated withintense rainfall (>30 mm day�1) (Fig. 4(b)). The TMPA-RT coincides relatively well with the gauge observationsin relation to the probability of exceedance of differentprecipitation intensity classes. TMPA-V7, CMORPHestimates and gauge observations detect almost the samefrequency of rain events for thresholds from 16 mm to26 mm. Overall, TMPA-RT shows lower occurrence fre-quency in comparison to the reference gauge observations,especially at high threshold intensities (45–60 mm/day)(Fig. 4(b)). These findings are similar to studies conductedin other regions with similar conditions where TMPA-RThas a susceptibility to overestimate lower precipitationrates and missing higher ones (Islam et al., 2010; Tahiret al., 2011; Yong et al., 2010).
4.2. Monsoon and pre-monsoon comparisons
The monsoon precipitation that starts from July andlasts until September varies from low rainfall (100 mm) inthe south (24–28 N), higher (>700 mm) in the northeast(29–33 N), and again low (<100 mm) in the far north
0
500
1000
1500
2000
2500
3000
0.0
0.2
0.4
0.6
0.8
1.0
CC
Latitude (degrees)
(a)
% B
ias
Latitude (degrees)
RM
SE
Latitude (degrees)
Ele
vatio
n (m
)
MA
E
Latitude (degrees)
Ele
vatio
n (m
)
ME
Latitude (degrees)
Ele
vatio
n (m
)
0
500
1000
1500
2000
2500
3000
-50
-25
0
25
50
75
100
125
150
Ele
vati
on (m
)
(b)
0
500
1000
1500
2000
2500
3000
0
5
10
15
20
Ele
vati
on (m
)
(c)
0
500
1000
1500
2000
2500
3000
0
2
4
6
8
10
Ele
vatio
n (m
)
(d)
0
500
1000
1500
2000
2500
3000
-2
-1
0
1
2
Ele
vatio
n (m
)
(e)
Fig. 3. Statistical summary of the comparison of gauge station based daily precipitation estimates between TMPA-V7, TMPA-RT and CMORPH for2005–2010 time period.
S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684 675
0
2
4
6
8
10
12
14
16
18
Ocu
ranc
e fr
eque
ncy
(%)
Precipitation (mm/day)
TMPA-V7TMPA-RTCMORPHGauge Precipitation
(a)
0
1
2
3
4
5
6
7
8
Ocu
ranc
e fr
eque
ncy
(%)
Precipitation (mm/day)
(b)
Fig. 4. Precipitation occurrence frequency (%) of gauge observations, TMPA-V7, TMPA-RT, and CMORPH as a function of precipitation intensity(mm day�1) respectively.
Fig. 5. Monsoon climatology over Pakistan. Data sources. Pakistan Meteorological Department.
676 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
0
500
1000
1500
2000
2500
3000
0.0
0.2
0.4
0.6
0.8
1.0
Ele
vatio
n (m
)
CC
Latitude (degrees)
(a)
0
500
1000
1500
2000
2500
3000
-50
50
150
250
350
450
Ele
vati
on (m
)
% B
ias
Latitude (degrees)
Latitude (degrees)
(b)
0
500
1000
1500
2000
2500
3000
0
5
10
15
20
25
30
35
Ele
vati
on (m
)
RM
SE
(c)
0
500
1000
1500
2000
2500
3000
0
2
4
6
8
10
12
Ele
vati
on (m
)
MA
E
Latitude (degrees)
(d)
0
500
1000
1500
2000
2500
3000
-6
-4
-2
0
2
4
6
Ele
vati
on (m
)
ME
Latitude (degrees)
(e)
Fig. 6. Statistical comparison of gauge station based daily precipitation estimates between TMPA-V7, TMPA-RT and CMORPH for 2005–2010monsoon season (JAS).
S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684 677
0
500
1000
1500
2000
2500
3000
0.0
0.2
0.4
0.6
0.8
1.0
Ele
vati
on (m
)
CC
Latitude (degrees)
(a)
0
500
1000
1500
2000
2500
3000
-80
20
120
220
320
Ele
vati
on (m
)
% B
ias
Latitude (degrees)
(b)
0
500
1000
1500
2000
2500
3000
0
5
10
15
20
25
30
Ele
vati
on (m
)
RM
SE
Latitude (degrees)
(c)
0
500
1000
1500
2000
2500
3000
0
2
4
6
8
10
Ele
vati
on (m
)
MA
E
Latitude (degrees)
(d)
0
500
1000
1500
2000
2500
-2
0
2
4
6
8
10
Ele
vati
on (m
)
ME
Latitude (degrees)
(e)
Fig. 7. Same as Fig. 6, but for 2005–2010 pre-monsoon season (AMJ).
678 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684 679
(34–36 N) (Fig. 5). The rainfall during the monsoon seasonaccounts for 50–70% of the total annual rainfall as it inten-sifies during July and August.
During the monsoon season TMPA-V7, TMPA-RT andCMORPH performed relatively well from 30 N to 34 N inthe wettest region but relatively poor from 24 N to 31 Ndue to overestimation in the drier regions where themean annual precipitation is generally less than 200 mm(Fig. 6) TMPA-V7 performed better than TMPA-RT andCMORPH with an improved correlation and reduced biasratio and error statistics in the intense monsoon region(32–35 N), (Fig. 6(a–e). Interestingly, all satellite precipita-tion estimates had a tendency to overestimate particularlyat the higher latitude region (34–36 N) as well as at lowerlatitudes (24–29 N) (Fig. 6(b)). The reader is reminded thatprecipitation during the monsoon season is still low in thefoothills of the Hindu Kush-Himalayas as well as thecoastal region in the south in comparison to the intensemonsoon region (Fig. 5). The effect of topography is dis-cussed in detail in the next section.
TMPA- RT and CMORPH overestimated rainfallduring the “pre monsoon” months of April, May and Junewhen the rainfall is not intense as during the monsoon
Fig. 8. Spatial comparison of gauge precipitatio
season (Fig. 7(a–e)). Overall, TMPA-V7 gave the bestresults while CMORPH had the highest bias over themid-latitude region and TMPA-RT had the highest biasover the high-elevation regions. These findings coincidewith prior studies in similar topographic settings (Chiuet al., 2006; Houze et al., 2011; Li et al., 2009; Yonget al., 2010) that cautioned against uncertainties inTRMM-derived precipitation. Nonetheless, our compari-sons show a good correspondence with all satellite-basedestimates to rain gauge amounts over the most intense pre-cipitation zone during the monsoon season (Fig. 6(b))
4.3. Comparison as a function of latitude
Reliable estimation of rainfall distribution in mountain-ous regions is a challenge due to high precipitation variabil-ity from orographic enhancement and possibly due toterrain-induced errors on remote-sensing retrievals. Houzeet al. (2007), Medina et al. (2010) and Romatschke et al.(2010)) explained the relationship of the South Asianmonsoon with the unique terrain of the region. Pakistan’stopography in the north poses a challenge to observeprecipitation due to the complex terrain and climatological
n over upper, plain and lower Indus basin.
680 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
conditions of the region. Therefore, we compared the threesatellite products with the gauge data at different topo-graphical regions including the upper Indus, Indus plain,and lower Indus of Pakistan. It has been found that precip-itation climatology over the Karakorum and western Hima-layas is not influenced by monsoon regime, rather thewesterly circulations that bring the highest precipitationduring winter in the form of snow (Tahir et al., 2011).
All the satellite precipitation products showed elevation-dependent biases that are characterized by overestimationat highest as well as the lowest altitude regions (Fig. 3).The overestimation by the three satellite products, particu-larly TMPA-RT and CMORPH during monsoon andpre-monsoon depends on the topography of the region.Although the satellite estimates were well correlated withthe gauge observations, they tended to overestimate lightprecipitation at both high and low elevation regions duringthe monsoon season (Fig. 6(a) and (b)). Daily precipitationcomparisons showed better correlation (Fig. 6(a)) forgauges within the elevation range of 500–1000 m with asignificantly reduced bias ratio (Fig. 6(b)) and error statis-tics (Fig. 3(c–e)). Statistical indices of the three satellitedatasets seemed to associate with the geo-topographic dis-tribution of the gauge stations as well as the precipitationamount. For example, worse statistics are generally for
Fig. 9. Spatial comparison of TMPA-V7 (red), TMPA-RT (blue) and CMORPin this figure legend, the reader is referred to the web version of this article.)
gauge stations located to the far north at higher elevationand far south, with both regions receiving relatively lessrainfall throughout the year (Fig. 5). Better performancewas observed to the northeast region that receives highintensity of rainfall during monsoon season (Fig. 6).
The overestimation TMPA-RT, TMPA-V7 andCMORPH with light rainfall intensities in the mountain-ous northern areas and southern coastal region is notableand consistent for daily and seasonal (monsoon andpre-monsoon) comparison. These findings are similar withother satellite precipitation evaluation studies over theRocky Mountains and the Sierra Nevada (Gottschalcket al., 2005), the Appalachian Mountains, Tibetan Plateau(Yin et al., 2008), Himalayas (Barros et al., 2000), and aglobal evaluation of TRMM (Gopalan et al., 2010). More-over, these spatial features and error statistics from thisstudy are consistent with some of the recent evaluations(Tahir et al., 2011; Webster et al., 2011) over the Indusbasin that showed that TMPA tends to overestimateprecipitation rates in the northwestern hilly tracts andsouthern coastal areas.
The comparison of the multi-sensor products to gaugeobservations during the monsoon at higher elevation showsa high positive bias. The overestimation can be attributedto the microwave data that might be seriously affected by
H (green) with gauge (black). (For interpretation of the references to color
Table 2Statistical comparison of the daily gauge precipitation with TMPA-V7 and TMPA-RT and CMORPH precipitation estimates for 2010 monsoon season (JAS).
Station Elevation Lat. Long. TMPA-V7 vs. Gauge TMPA-RT vs. Gauge CMORPH vs. Gauge
CC RMSE ME MAE % Bias CC RMSE ME MAE % Bias CC RMSE ME MAE % Bias
1 Karachi 22 24.89 67.05 0.8 9.2 0.04 3.0 1.4 0.7 15.1 1.54 4.3 51 0.7 8.0 1.23 2.2 412 Jiwani 56 25.04 61.74 0.6 15.2 0.01 0.2 23 0.5 16.9 0.01 0.3 48 0.5 17.0 0.04 7.7 633 Pasni 9 25.26 63.46 0.6 24.5 0.26 3.2 8 0.6 22.2 1.13 10.8 37 0.4 21.2 0.88 9.6 294 Hyderabad 28 25.38 68.36 0.6 18.8 1.23 9.8 85 0.5 25.8 2.77 12.5 191 0.6 20.3 2.26 9.7 1565 Lasbella 87 25.83 66.58 0.3 21.5 1.90 8.9 182 0.4 21.5 3.73 11.0 358 0.4 20.0 2.36 9.9 2276 Nawabshah 37 26.15 68.22 0.7 18.2 -1.19 11.0 -13 0.7 14.1 -3.45 9.5 -38 0.6 25.6 -1.35 11.6 -157 Rohri 60 27.69 68.85 0.4 3.8 0.59 1.4 81 0.5 11.2 0.79 3.8 108 0.5 14.7 2.03 4.7 2798 Jacobabad 55 28.27 68.45 0.4 21.0 -1.28 7.8 -43 0.6 19.5 -1.27 7.6 -43 0.5 20.2 0.03 8.2 1.19 Kalat 370 29.02 66.58 0.0 1.2 0.26 0.0 79 0.1 15.2 0.20 0.2 66 0.2 15.7 0.88 6.3 2910 Barkhan 1372 29.89 69.52 0.5 8.9 -0.86 3.9 -23 0.5 8.9 -0.19 4.0 -5 0.6 16.1 0.60 5.4 1611 Multan 121 30.19 71.46 0.8 6.9 0.43 2.8 17 0.8 21.8 1.49 8.5 59 0.9 22.5 1.49 9.1 5912 Lahore 214 31.54 74.34 0.7 20.2 -0.45 9.3 -8 0.6 20.1 -1.08 10.4 -20 0.6 19.1 -0.92 11.1 -1713 Sialkot 255 32.50 74.53 0.3 11.3 0.04 11.8 29 0.3 24.0 0.02 8.2 13 0.3 24.9 0.22 11.2 15614 Jhelum 287 32.93 73.72 0.4 16.9 0.90 7.6 16 0.5 19.3 -1.15 8.0 -20 0.5 21.3 -0.89 8.9 -1615 Kotli 614 33.52 73.90 0.4 18.1 0.78 10.0 11 0.4 17.3 0.32 9.5 5 0.4 11.2 -0.49 5.8 -716 Islamabad 508 33.66 73.06 0.6 19.4 -0.83 7.8 -11 0.5 16.2 -1.86 9.1 -24 0.5 20.5 -0.83 11.2 -1117 Cherat 1250 33.82 71.88 0.8 18.2 -1.66 5.2 -24 0.6 24.6 -3.37 5.0 -48 0.6 33.4 -2.07 9.0 -2918 Murree 1900 33.91 73.39 0.7 22.5 -2.03 9.4 -21 0.7 17.3 -2.43 8.1 -25 0.5 12.2 -2.01 5.9 -2019 Garidopatta 813 34.13 73.37 0.7 18.0 -1.54 8.2 -16 0.7 16.0 -3.79 6.0 -40 0.7 9.2 -1.69 3.9 -1820 Muzaffarabad 838 34.22 73.29 0.7 22.5 -2.03 7.7 -21 0.7 17.3 -2.43 5.5 -25 0.5 12.2 -2.01 5.9 -2021 Balakot 995 34.55 73.35 0.8 11.0 0.46 5.1 7 0.7 10.3 -1.93 4.8 -31 0.8 11.3 -0.96 7.7 -1522 Astore 2168 35.35 74.86 0.6 7.4 1.51 2.3 119 0.6 5.6 0.79 2.1 62 0.2 3.7 0.66 2.3 5223 Drosh 1463 35.55 71.79 0.9 10.2 4.19 3.2 105 1.0 2.3 0.25 1.2 4 0.9 8.1 0.90 3.3 4324 Chitral 1497 35.83 71.78 0.9 11.4 2.37 1.6 175 0.9 2.4 0.20 2.1 15 0.9 5.2 0.78 2.4 58
S.I.
Kh
an
eta
l./Ad
van
cesin
Sp
ace
Resea
rch5
4(
20
14
)6
70
–6
84
681
682 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
the ice particles in the air and snow cover over land; Yonget al. (2012) describes the overestimation at higherelevation as misclassification of cold surface and icecovered mountaintops as rain clouds by passive MWsensors. Therefore rainfall retrieval algorithms that useinformation from a combination of IR, passive MW, andactive MW sensors are subject to all of these error sourcesin mountainous regions (Yong et al., 2010). As a result, thesatellite estimates suffer serious bias over the higheraltitudes.
In terms of error statistics, all three products show anRMSE ranging from 2–20 mm/day. Of the three datasets,TMPA-V7 exhibits the lowest RMSE and MAE over mostregions, which is also consistent with the bias pattern formonsoon season (Fig. 6) and pre-monsoon season(Fig. 7). CMORPH does not present similar advantagesin RMSE, with larger RMSE than TMPA-RT over severaltopographic regions.
4.4. Comparison for the 2010 flood case
The monsoon weather system that developed on 24 July,2010 in the Bay of Bengal, penetrated westward reachingPakistan on 27 July, which produced extreme rainfall overthe northwestern region. The unusual intense precipitationleading to the floods in August 2010 began with a series ofintense downpours at the end of July through early August(Fig. 8). The downpour from 28–30 July amounted to morethan 300 mm of rainfall in the northwest Pakistan at a sin-gle gauge station (Fig. 9), which led to the devastatingfloods. We evaluated the spatial and temporal patterns ofthe biases of intense rainfall estimates based on the threesatellite products over the study area during the 2010intense monsoon event. It can be seen that the rainfallwas unusually high and deviated strongly from the normalmonsoon pattern, with high intensity into the northern part(35–37 N) (Fig. 8).
A station-based comparison of the TMPA-V7, TMPA-RT and CMORPH datasets illustrates considerable simi-larities and differences in the precipitation estimates duringthe third precipitation spell of July 2010 that occurred overthe monsoon region (32–35 N). Several statistical andgraphical techniques are applied to illustrate the noncon-formity of satellite estimates from the gauge observations.Error statistics calculated from the daily rainfall duringthe 2010 monsoon season are tabulated in Table 2. Inter-estingly, for this intense rainfall event, all the productsperformed well and the best performance was again seenover the northern part (monsoon) region where mostrainfall occurred (Fig. 9). TMPA-RT, TMPA-V7 andCMORPH exhibit better performance for the monsoonregion (32–35 N) with high correlation coefficients andlow error statistics (Table 2). Analysis of the error charac-teristics directed that, overall, the three satellite productshave a tendency to underestimate on storm scale. Duringthat particular event, satellite precipitation estimates aremarked by underestimation that was typically within
25%, 50% and 30% for TMPA-RT, TMPA-V7 andCMORPH, respectively.
Similarly, during late July 2010, the satellite algorithmshave distinct regional patterns with the error statistics andbiases for the largest rainstorm. In the southern coast andnorthern mountainous terrain the three products showed ahigh correlation with gauge precipitation but also overesti-mate the observed values (Table 2). TMPA-V7 shows verygood agreement with high to moderate correlation values(CC = 0.8–0.5) respectively and low systematic bias (Bias±25%). Meanwhile, both TMPA-RT and TMPA-V7estimates and CMORPH products captures the temporalevolution and fluctuations of gauge network observationsreasonably well throughout the 2010 monsoon (Fig. 9).All three products show positive biases for daily rain ratesover the lower latitudes at elevations lower than 100 m.The same trend was seen for the high altitude region at ele-vations higher than 1000 m (Table 2).
5. Summary and conclusions
The presented study focused on the performance ofthree high-resolution satellite rainfall products TMPA-V7, TMPA-RT and CMORPH over Pakistan with diversetopographical features. The inter-comparison and valida-tion of these satellite products is performed both quantita-tively and qualitatively. Several statistical validations areshown for quantitative analysis, whereas the qualitativeanalysis provides information about the performance ofthe satellite products relative to the rainfall rate andground elevation. Moreover, the topographical andmonsoonal influences on the performance of the satelliteproducts over different climatological regions in Pakistanare also investigated. Our method consists of inter compar-ison of satellite rainfall estimates and daily rain gauge dataduring the northeast monsoon and pre-monsoon seasonsfor 6 years from 2005 to 2010. In addition, a quantitativeassessment of the 2010 flooding event as a function ofelevation is discussed. The main findings of this study aresummarized below:
1. For the overall compassion that used all the 24 selectedrain gauge stations through the country for the studyperiod from 2005–2010, TRMM-v7 and CMORPH,tended to slightly overestimate rainfall Relative toTMPA-RT. For the comparison at different latitudes,the three satellite products showed a significant positivebias in comparison to surface gauge observations andthe accuracy was influenced by the diverse topographyof the area. For example, there are large biases (100%)at lower latitudes and southern coastal areas. andTMPA-RT. similar overestimates over high latitudes(35N) in complex terrain.
2. In terms of capturing different precipitation intensities,TMPA-V7 had better agreement with gauge observa-tions than TMPA-RT for the high precipitationintensity range (45–60 mm/day). TMPA-V7 performed
S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684 683
better at for the entire time series (2005–2010) as well asmonsoon and pre-monsoon seasons. High relativeuncertainties in terms of error statistics persisted overlower latitudes for the daily, monsoon as well as thepre-monsoon seasons.
3. All satellite algorithms slightly underestimated rainfall inthe monsoon region. They indicated a negative bias ofapproximately 25% for the monsoon region, while bothTMPA products grossly overestimated in the far northmountainous region and low coastal areas to the south.
4. All satellite estimates showed larger correlations whenthe comparison was conducted for the monsoon seasonand 2010 flooding event. Correlations improved from0.5 for entire time period (2005–2010) to 0.7 for the2010 monsoon season for the three estimates. Duringthe pre-monsoon season, TMPA-V7 slightly underesti-mated the rainfall amounts, especially along the mon-soon center and over higher elevations, but its overallperformance was less variable spatially and temporarilythan CMORPH. For the 2010 monsoon season, TMPA-V7 performed the best over the monsoon center with abias of �20% showing slight underestimation of stationprecipitation.
Synthesis of these results supports the conclusion thatcurrent satellite-based precipitation products are reliableover the monsoon region areas such as over the Indusplains. It is noteworthy that we found several shortcomingsof the satellite rainfall estimates, but these errors weremainly confined to regions and times where the rainfallintensity is relatively light. Fortunately, the remote-sensingretrievals seemed to have the best performance during theintense phases of the monsoon when they are needed most.From this study we can conclude that satellite-derived pre-cipitation products, such as TMPA can be used as inputforcing variable into hydrological models to derive riverrunoff estimates. This is especially true for applicationsduring the summer monsoon season over the Indus basin,where in situ precipitation gauge networks are sparse.The future strategy will be utilize these unconventionalsatellite precipitation products, within a hydrologic model-ing framework for runoff simulations for flood monitoringand water resource management.
Acknowledgments
This research work was supported by the ongoingproject under the Pakistan-US Science and TechnologyCooperation Program jointly funded by the United StatesAgency for International Development (USAID) with theMinistry of Science and Technology (MOST) and theHigher Education Commission of Pakistan (HEC). Thisprogram is being implemented by the U.S. NationalAcademy of Sciences. The authors would like toacknowledge Climate Data Processing Centre of PakistanMeteorological Department for providing rainfall data.
References
Andermann, C., Bonnet, S., Gloaguen, R., 2011. Evaluation of precip-itation data sets along the Himalayan front. Geochem. Geophys.Geosyst. 12, Q07023.
Astin, I., 1997. A survey of studies into errors in large scale space-timeaverages of rainfall, cloud cover, sea surface processes and the earth’sradiation budget as derived from low earth orbit satellite instrumentsbecause of their incomplete temporal and spatial coverage. Surv.Geophys. 18, 385–403.
Barros, A., Joshi, M., Putkonen, J., Burbank, D., 2000. A study of the1999 monsoon rainfall in a mountainous region in central Nepal usingTRMM products and rain gauge observations. Geophys. Res. Lett. 27,3683–3686.
Behrangi, A., Khakbaz, B., Jaw, T.C., AghaKouchak, A., Hsu, K.,Sorooshian, S., 2011. Hydrologic evaluation of satellite precipitationproducts over a mid-size basin. J. Hydrol. 397, 225–237.
Bookhagen, B., Burbank, D.W., 2010. Toward a complete Himalayanhydrological budget: spatiotemporal distribution of snowmelt andrainfall and their impact on river discharge. J. Geophys. Res. 115,F03019.
Chiu, L.S., Liu, Z., Vongsaard, J., et al., 2006. Comparison of TRMM andwater district rain rates over New Mexico. Adv. Atmos. Sci. 23, 1–13.
Dinku, T., Ceccato, P., Connor, S.J., 2011. Challenges of satellite rainfallestimation over mountainous and arid parts of east Africa. Int. J.Remote Sens. 32, 5965–5979.
Ebert, E.E., Janowiak, J.E., Kidd, C., 2007. Comparison of near-real-timeprecipitation estimates from satellite observations and numericalmodels. Bull. Am. Meteorol. Soc. 88, 47–64.
Gopalan, K., Wang, N.-Y., Ferraro, R., Liu, C., 2010. Status of theTRMM 2A12 land precipitation algorithm. J. Atmos. OceanicTechnol. 27, 1343–1354.
Gottschalck, J., Meng, J., Rodell, M., Houser, P., 2005. Analysis ofmultiple precipitation products and preliminary assessment of theirimpact on global land data assimilation system land surface states. J.Hydrometeorol. 6, 573–598.
Gourley, J.J., Hong, Y., Flamig, Z.L., Li, L., Wang, J., 2010. Inter-comparison of rainfall estimates from radar, satellite, gauge, andcombinations for a season of record rainfall. J. Appl. Meteorol.Climatol. 49, 437–452.
Griffith, C.G., Woodley, W.L., Grube, P.G., Martin, D.W., Stout, J.,Sikdar, D.N., 1978. Rain estimation from geosynchronous satelliteimagery—visible and infrared studies. Mon. Weather Rev. 106, 1153.
Habib, E., Henschke, A., Adler, R.F., 2009. Evaluation of TMPAsatellite-based research and real-time rainfall estimates during sixtropical-related heavy rainfall events over Louisiana. USA Atmos.Res. 94, 373–388.
Hong, Y., Hsu, K.L., Sorooshian, S., Gao, X., 2004. Precipitationestimation from remotely sensed imagery using an artificial neuralnetwork cloud classification system. J. Appl. Meteorol. 43, 1834–1853.
Hong, Y., Adler, R.F., Hossain, F., Curtis, S., Huffman, G.J., 2007. A firstapproach to global runoff simulation using satellite rainfall estimation.Water Resour. Res. 43, W08502.
Hossain, F., Huffman, G.J., 2008. Investigating error metrics for satelliterainfall data at hydrologically relevant scales. J. Hydrometeorol. 9,563–575.
Houze Jr, R.A., Wilton, D.C., Smull, B.F., 2007. Monsoon convection inthe Himalayan region as seen by the TRMM precipitation radar. Q. J.R. Meteorol. Soc. 133, 1389–1411.
Houze, R.A., Rasmussen, K.L., Medina, S., Brodzik, S.R., Romatschke,U., 2011. Anomalous atmospheric events leading to the summer 2010floods in Pakistan. Bull. Am. Meteorol. Soc. 92, 291–298.
Huffman, G.J., Adler, R.F., Morrissey, M.M., et al., 2001. Globalprecipitation at one-degree daily resolution from multisatellite obser-vations. J. Hydrometeorol. 2, 36–50.
Huffman, G.J., Bolvin, D.T., Nelkin, E.J., et al., 2007. The TRMMmultisatellite precipitation analysis (TMPA): quasi-global, multiyear,
684 S.I. Khan et al. / Advances in Space Research 54 (2014) 670–684
combined-sensor precipitation estimates at fine scales. J. Hydromete-orol. 8, 38–55.
Huffman, G.J., Bolvin, D.T., Nelkin, E.J., Adler, R.F. Highlights ofVersion 7 TRMM Multi-satellite Precipitation Analysis (TMPA). In:Klepp, C., Huffman, G. (Eds.), 5th Internat. Precip. Working GroupWorkshop, Workshop Program and Proceedings, 11–15 October 2010,Hamburg, Germany. Reports on Earth Sys. Sci., 100/2011, Max-Planck-Institut fur Meteorologie. ISSN 1614-1199, 2011, pp. 109–110.
Islam, M.N., Das, S., Uyeda, H., 2010. Calibration of TRMM derivedrainfall over Nepal during 1998–2007. Open Atmos. Sci. J. 4, 12–23.
Joyce, R.J., Janowiak, J.E., Arkin, P.A., Xie, P., 2004. CMORPH: amethod that produces global precipitation estimates from passivemicrowave and infrared data at high spatial and temporal resolution.J. Hydrometeorol. 5, 487–503.
Kawanishi, T., Kuroiwa, H., Kojima, M., et al., 2000. TRMM Precip-itation Radar. Adv. Space Res. 25, 969–972.
Khan, J.A., 1993. The Climate of Pakistan. Rehbar Publishers.Kidd, C., Kniveton, D.R., Todd, M.C., Bellerby, T.J., 2003. Satellite
rainfall estimation using combined passive microwave and infraredalgorithms. J. Hydrometeorol. 4, 1088–1104.
Kubota, T., Shige, S., Hashizume, H., et al., 2007. Global precipitationmap using satellite-borne microwave radiometers by the GSMaPProject: production and validation. IEEE Trans. Geosci. Remote Sens.45, 2259–2275.
Kummerow, C., Barnes, W., Kozu, T., Shiue, J., Simpson, J., 1998. Thetropical rainfall measuring mission (TRMM) sensor package. J.Atmos. Oceanic Technol. 15, 809–817.
Li, L., Hong, Y., Wang, J., et al., 2009. Evaluation of the real-timeTRMM-based multi-satellite precipitation analysis for an operationalflood prediction system in Nzoia Basin, Lake Victoria. Africa. Nat.Hazards 50, 109–123.
Maddox, R.A., Zhang, J., Gourley, J.J., Howard, K.W., 2002. Weatherradar coverage over the contiguous United States. Weather Forecast.17, 927–934.
Margulis, S.A., 2001. Temporal disaggregation of satellite-derivedmonthly precipitation estimates and the resulting propagation of errorin partitioning of water at the land surface. In: Hydrology and EarthSystem, Science 5.
Medina, S., Houze Jr, R.A., Kumar, A., Niyogi, D., 2010. Summermonsoon convection in the Himalayan region: terrain and land covereffects. Q. J. R. Meteorol. Soc. 136, 593–616.
Romatschke, U., Medina, S., Houze Jr, R.A., 2010. Regional, seasonal,and diurnal variations of extreme convection in the South Asianregion. J. Clim. 23, 419–439.
Semire, F.A., Mohd-Mokhtar, R., Ismail, W., Mohamad, N., Mandeep,J.S., 2012. Ground validation of space-borne satellite rainfall productsin Malaysia. Adv. Space Res. 50, 1241–1249.
Simpson, J., Kummerow, C., Tao, W.K., Adler, R., 1996. On the tropicalrainfall measuring mission (TRMM). Meteorol. Atmos. Phys. 60, 19–36.
Smith, E., Asrar, G., Furuhama, Y., et al., 2007. International GlobalPrecipitation Measurement (GPM) Program and Mission: An Over-view. In: Levizzani, V., Bauer, P., Turk, F.J. (Eds.), MeasuringPrecipitation From Space. Springer, Netherlands, pp. 611–653.
Sorooshian, S., Hsu, K.L., Xiaogang, G., Gupta, H.V., Imam, B.,Braithwaite, D., 2000. Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. In: Bull. Am. Meteorol. Soc. 81,2035–2046.
Steiner, M., Bell, T.L., Zhang, Y., Wood, E.F., 2003. Comparison of twomethods for estimating the sampling-related uncertainty of satelliterainfall averages based on a large radar dataset. J. Clim. 16, 3759–3778.
Su, F., Hong, Y., Lettenmaier, D.P., 2008. Evaluation of TRMMmultisatellite precipitation analysis (TMPA) and its utility in hydro-logic prediction in the La Plata Basin. J. Hydrometeorol. 9, 622–640.
Tahir, A.A., Chevallier, P., Arnaud, Y., Neppel, L., Ahmad, B., 2011.Modeling snowmelt-runoff under climate scenarios in the Hunza Riverbasin, Karakoram Range, Northern Pakistan. J. Hydrol. 409, 104–117.
Tian, Y., Peters-Lidard, C.D., 2010. A global map of uncertainties insatellite-based precipitation measurements. Geophys. Res. Lett. 37,L24407.
Turk, F.J., Miller, S.D., 2005. Toward improved characterization ofremotely sensed precipitation regimes with MODIS/AMSR-E blendeddata techniques. IEEE Trans. Geosci. Remote Sens. 43, 1059–1069.
Vicente, G.A., Scofield, R.A., Menzel, W.P., 1998. The operational GOESinfrared rainfall estimation technique. Bull. Am. Meteorol. Soc. 79,1883–1898.
Wang, S.Y., Davies, R.E., Huang, W.R., Gillies, R.R., 2011. Pakistan’stwo-stage monsoon and links with the recent climate change. J.Geophys. Res. Atmos. 116.
Webster, P.J., Toma, V.E., Kim, H.M., 2011. Were the 2010 Pakistanfloods predictable? Geophys. Res. Lett. 38.
Yin, Z.Y., Zhang, X., Liu, X., Colella, M., Chen, X., 2008. An assessmentof the biases of satellite rainfall estimates over the Tibetan Plateau andcorrection methods based on topographic analysis. J. Hydrometeorol.9, 301–326.
Yong, B., Ren, L.-L., Hong, Y., et al., 2010. Hydrologic evaluation ofmultisatellite precipitation analysis standard precipitation products inbasins beyond its inclined latitude band: a case study in Laohahe basin.China. Water Resour. Res. 46, W07542.
Yong, B., Hong, Y., Ren, L.-L., et al., 2012. Assessment of evolvingTRMM-based multi-satellite real-time precipitation estimation meth-ods and their impacts on hydrologic prediction in a high latitude basin.J. Geophys. Res. Atmos.
