t 108 (2007) 264–276www.elsevier.com/locate/rse

Remote Sensing of Environmen

Use of TRMM in determining the climatic characteristicsof rainfall over Bangladesh

Md. Nazrul Islam a,⁎, Hiroshi Uyeda b

a Department of Physics, Bangladesh University of Engineering and Technology, Dhaka, Bangladeshb Hydrospheric Atmospheric Research Center, Nagoya University, Japan

Received 11 May 2006; received in revised form 14 November 2006; accepted 15 November 2006

Abstract

Five years of data from 1998 to 2002 of TRMM-3B42 version 5 (V5), 3B43 V5, 3B42 version 6 (V6), 3B43 V6, and the BangladeshMeteorological Department rain-gauge network were analyzed to understand the climatic characteristics of rainfall over Bangladesh. TRMM-PR2A25 data were used to obtain the precipitation field of the convection events. Daily rainfall measured by TRMM V5 3B42 was compared to thatof rain-gauge values from pre-monsoon to post-monsoon months (March–November). The time sequence patterns of the daily rainfall determinedby the V5 3B42 and those from rain gauges were remarkably similar. The spatial and temporal averages of rainfall revealed good estimations ofrainfall: during March to November, the V5 3B42- and rain gauge-estimated daily rainfall was 8.12 and 8.34 mm, respectively. In annual scale,TRMM V5 3B42-, V5 3B43-, V6 3B42-, V6 3B43- and rain-gauge estimated rainfall was 6.9, 6.4, 6.6, 6.8 and 7.1 mm/day, respectively. Theaverage percentage of rainy days determined by V5 3B42 data with respect to the rain-gauge value was 96%. TRMM is useful for estimating theaverage values of rainfall in Bangladesh. The prominent difference between rainfall estimated by rain-gauge and V5 3B42 was found to be period-and location-dependent. The V5 3B42 overestimated the rainfall during the pre-monsoon period and in dry regions but underestimated it duringthe monsoon period and in wet regions. The reason for the differences according to season and locations is considered to be the vertical crosssection of convection obtained by TRMM-PR 2A25 data. The rainfall overestimation in pre-monsoon and underestimation in monsoon periodmeasured by V5 3B42 is reduced to reasonable amount by V6 3B42 and V6 3B43. In this manner, the merit of using TRMM data forclimatological studies of rainfall over Bangladesh is shown.© 2006 Elsevier Inc. All rights reserved.

Keywords: Precipitation; Monsoon; Tropical Rainfall Measuring Mission; Rain gauge

1. Introduction

The Tropical Rainfall Measuring Mission (TRMM), cospon-sored by the National Aeronautics and Space Administration(NASA) of the U.S.A. and the Japan Aerospace ExplorationAgency (JAXA, previously known as the National SpaceDevelopment Agency, or NASDA), has collected data sinceNovember 1997 (Kummerow et al., 2000). Tropical rainfall,which falls between 35°N and 35°S, comprises more than two-thirds of global rainfall. TRMM is a long-term research programdesigned to study the Earth's land, oceans, air, ice, and life as a

⁎ Corresponding author. Tel.: +880 2 9665650(7151); fax: +880 2 8613046.E-mail address: mnislam@phy.buet.ac.bd (M.N. Islam).

0034-4257/$ - see front matter © 2006 Elsevier Inc. All rights reserved.doi:10.1016/j.rse.2006.11.011

total system. Previous estimates of tropical precipitation wereusually made on the basis of climate prediction models and theoccasional inclusion of very sparse surface rain gauges (RNGs)and/or relatively few measurements from satellite sensors. TheTRMM satellite allows these measurements to be made in afocused manner. TRMM is NASA's first mission dedicated toobserving and understanding tropical rainfall and how it affectsthe global climate (Simpson et al., 1988; Wolff et al., 2005). TheTRMM spacecraft fills an enormous void in the ability toestimate worldwide precipitation because ground-based radarsthat measure precipitation cover a very small part of the planet.Ground-based radars cover only small percent of the areacovered by TRMM, said Kummerow (http://www.xs4all.nl/~carlkop/tropmis.html) and RNGs are limited to specificgeographic points. The TRMM Ground Validation (GV)

265M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

Program began in the late 1980s and has yielded a wealth ofdata and resources for validating TRMM satellite estimates byproviding rainfall products for four sites: Darwin, Australia(DARW); Houston, Texas (HSTN); Kwajalein, Republic of theMarshall Islands (KWAJ); and Melbourne, Florida (MELB).Wolff et al. (2005) provide extensive details on the TRMM GVprogram, site descriptions, algorithms, and data processing. Foryears, other groups studied various locations to validate TRMMdata, such as Ikai and Nakamura (2003), who calculated rainrates over the Ocean, Nicholson et al. (2003), who validatedTRMM rainfall for West Africa, and Barros et al. (2000), whostudied a monsoon case in Nepal. TRMM mission is officiallyextended up to 2009 and in the mean time focus has begun onplanning for the Global Precipitation Measurement (GPM) GVprogram. The GPM satellite will commence operations in 2010.The main goal of the TRMM GV program is to provide rainestimates at various sites throughout the globe in order tocompare with and hopefully help to improve GPM satelliteretrievals. Unfortunately, so far, there has been no research doneto validate TRMM data over Bangladesh. This paper describeshow accurately the TRMM satellite can determine surfacerainfall in Bangladesh. In Bangladesh, RNG data are the onlyrepresentation of precipitation throughout the country. Inade-quate RNG networks throughout the country sometimes provideincomplete information on the distribution of rainfall. The useof remote sensing data in estimating rainfall in Bangladesh,thus, offers an exciting opportunity. Neither RNGs nor satellite-based estimates are perfect indicators of rainfall (Nicholson etal., 2003). Morrissey and Greene (1993) and Xie and Arkin(1995, 1996) show in their examinations that all satelliteestimates have non-negligible biases when compared withconcurrent in situ observations. With RNGs, biases areintroduced by gauge type, maintenance, and placement (Legates& Willmott, 1990; Sevruk, 1982) as well as by spatial sampling(Huffman et al., 1995, 1997; Morrissey et al., 1995; Rudolf etal., 1994). Xie and Arkin (1995) concluded that these are smallwhen compared with the bias in satellite estimates. Even whenthese biases are included in the measurements, RNG rainfall isused as a ground-based rainfall measurement.

In and around Bangladesh, the rainy season is divided intothree periods: (a) pre-monsoon (March–May), (b) monsoon(June–September), and (c) post-monsoon (October–Novem-ber). In Bangladesh, about 20%, 62.5%, 15.5%, and 2% of theannual rainfall (∼2700 mm) occurs during pre-monsoon,monsoon, post-monsoon, and winter periods, respectively(Islam & Uyeda, 2005). Research conducted on the estimationof rainfall in Bangladesh using remote sensing data remainsinadequate. Recently, Ohsawa et al. (2001) studied therelationship of infrared brightness temperature (TBB) of cloudtop heights from GMS (Geostationary Meteorological Satellite)data to RNG rainfall in Bangladesh. As mentioned above, verylittle validation work has been conducted on the rainfallestimated by TRMM over Bangladesh, and the GPM GV is anon-going project. A number of researchers are working on thedevelopment of instrumentation and algorithms for GPM.Information about the distribution of rainfall and the structure ofprecipitation systems from a heavy-rainfall region, such as

Bangladesh, is important for these developers. In this work,attempts have been made to compare the rainfall determined byTRMM 3B42 products, which is the combination of TRMMPrecipitation Radar (PR) and TRMM Microwave Imager(TMI), with the values of ground-based RNGs throughoutBangladesh. Pre-monsoon, monsoon, and post-monsoon rainrates have distinctive features in Bangladesh as well as in partsof Asia that experience monsoons. These are also obtained fromTRMM 3B42, TRMM-PR 2A25, and RNG data. Radar PPIscans data from the Bangladesh Meteorological Department(BMD) are used to support the TRMM-PR horizontal images.

2. Data and methods

Rainfall data for 3-h periods from 1998 to 2002, measuredand collected by the Bangladesh Meteorological Department(BMD), were used to prepare the RNG rainfall data. In thispurpose, the metadata chart of BMD was utilized to avoid therandom error such as mistake in data transfer through radio link.Even though, RNG data were missing at WMO station #41960(90.36E, 22.33N) in 1998 and WMO station #41963 (91.07E,22.30N) in 2000. Area of Bangladesh (88.05°–92.74°E and20.67°–26.63°N) is 147570 km2, where BMD placed 31 raingauges throughout the country. TRMM produced a daily 1°×1°microwave-calibrated IR rain estimate (TRMM Science Dataand Information System (TSDIS) 3B42). The daily data fromTRMM V5 3B42 products estimated gridded rainfall of 1°×1°resolution for the same analysis period. The V5 3B43 product isproduced by merging the monthly accumulation of daily V53B42 with the monthly accumulated Climate Assessment andMonitoring System (CAMS) or Global Precipitation Climatol-ogy Centre (GPCC) rain-gauge analysis (3A45) (Huffman et al.,1995). The V6 3B42 is a 3 hourly 0.25° product based on multi-satellite precipitation analysis (Huffman et al., 2004). The multi-satellite and gauge analysis are merged (Huffman et al., 1997) tocreate a post-real-time satellite–gauge monthly product, whichis the TRMM product V6 3B43. The RNG values were notuniformly gridded because the RNGs of BMD are notpositioned at uniform grid distances. Therefore, the comparisonbetween rainfall estimated by TRMM V5 3B42 and thatmeasured by RNG was performed on a point-to-point basis, i.e.,one RNG must be in a TRMM data grid box. The analysisperiod of 1998–2002 was chosen because it was the only onewith a full complement of data for the TRMM and the collectedRNGs rainfall in Bangladesh. In the analysis, the entire rainyperiod of March–November has been accounted for, which isnearly the complete rain period of Bangladesh, with 98% of thetotal annual precipitation. Except for 2A25 and 3B43, 3B42products are available for each day of the 5-year analysis period.Five types of analyses were carried out. First, the mean rainfallfield for March to November (MAMJJASON) was prepared forthe V5 3B42 dataset. Second, day-to-day comparisons ofrainfall for five well-separated and selected stations throughoutBangladesh were conducted. The 0-mm rainfall within3-h periods has been accounted to calculate the daily rainfallamount. Third, point-to-point comparisons of rain climates wereobtained from the TRMM V5 3B42 and RNG datasets for the

Fig. 1. Daily rainfall (mm) determined by RNG (left of plus mark) and TRMM(right of plus mark) at different rain-gauge sites (plus mark) throughoutBangladesh. Area of Bangladesh is 147570 km2. Data averages for 1 March to30 November from 1998 to 2002. The shaded arrow represents the route ofmonsoon progression.

266 M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

three rainy periods. Fourth, rainy days for each month, eachyear, and the overall 5-year period were detected with the RNGdataset as a reference. Fifth, comparative analysis wasperformed by using TRMM V5 3B42-, V5 3B43-, V6 3B42-,V6 3B43- and RNG rainfall to see the performance of TRMMalgorithms. Any amount of rainfall recorded by an RNG within3-h was defined as a rainy day, whereas zero rainfall within a24-h period was defined as a rain-free day. Rainy dayssimultaneously detected by both TRMM 3B42 and RNG weredeclared as match days. The dry and wet regions were definedas areas in which the relative humidity was below and above theaveraged relative humidity value of the country, respectively.The monsoon onset date in a particular region was determinedby a change of the surface wind field from mainly NW to S/SEand recorded precipitation during at least 2–3 consecutive days.Anomalies were defined by deviations in a country-averaged(50-year data from 1950) value from individual ones. Rainfallclimatology was defined normal, dry and wet for rainfallanomalies within ±10%, below −10% and above +10%respectively. Performance of TRMM 3B42 and 3B43 algo-rithms was characterized into three basic categories (Brown,2006): underestimation, overestimation and approximatelyequal (within ±10%). The false detection of daily rainy dayswas defined by the number of rainy days determined by TRMMminus the number of rainy days determined by RNG and thendivided by the number of days under calculation.

3. Results

This section will show several comparisons of RNG rainintensities and satellite retrievals from the TRMM MicrowaveImager (TMI) and Precipitation Radar (PR) algorithms.

3.1. Point-to-point comparison of TRMM V5 3B42 and RNGrain rates

Daily rainfall (mm) estimated from TRMM V5 3B42 (rightof plus mark, Fig. 1) and RNG data (left of plus mark, Fig. 1)averaged for the period of 1 March to 30 November 1998–2002at the location of each RNG (plus mark, Fig. 1) throughoutBangladesh are shown in Fig. 1. The V5 3B42 underestimated(11–37%) the rain rate in the north–northeastern and south–southeastern parts of Bangladesh, which are the heavy-rainfallregions of the country (Islam et al., 2005), except overestima-tion (8–16%) at 3 stations. The determination of the rain rate inthe remaining parts of the country by V5 3B42 is overestimated(12–42%).

3.2. Day-to-day comparison of TRMM V5 3B42 and RNG rainrates

For the day-to-day comparison, five stations, Teknaf(southeast), Dhaka (center), Sylhet (northeast), Rangpur(northwest), and Khulna (southwest), were selected, as shownin Fig. 1. The mentioned stations are selected as therepresentative of the respective regions and detail statisticalanalysis of 31 stations is discussed later. A comparison of the

daily rainfall measured by TRMM V5 3B42 and RNG at thefive selected stations in 2001 is shown in Fig. 2. Vertical lines inFig. 2 were used to divide the entire rainy season into pre-monsoon, monsoon, and post-monsoon periods. To representthe date of the monsoon onset in a particular region, a downarrow is used. Teknaf is located in the southeastern coastalregion of the country (see Fig. 1), and the pre-monsoon systemdoes not generally reach that region. Therefore, the lowest rateswere recorded in this region during the first 2 months (Marchand April) of the pre-monsoon period. At the end of the firstweek or in the beginning of the second week of May, rainfallwas first observed in this region when the monsoon circulationwas established (not shown) in the South Bay and adjoiningareas (south of 13(N)). Generally, the monsoon circulationbegins to have an impact in the last week of May. Historically,the monsoon onset is 31 May in this region (Das, 1995). Themean monsoon onset dates in Bangladesh have been clarified byAhmed and Karmakar (1993). However, the rainfall in the lastweek of May is sometimes indistinguishable from the monsoononset. The set-up of surface wind fields from northeasterly tosouthwesterly with a consecutive rainfall for about 2 to 3 ormore days determined the monsoon onset in this region, andBMD used this criterion for the definition of the onset. BMDconcluded that the monsoon onset in the southern parts ofBangladesh (Teknaf) had occurred on 7 June in 1998, 25 May in1999, 29 May in 2000, 26 May in 2001, and 03 June in 2002(the onset dates for 1998–2000 and 2002 are not shown). Anearly onset of the monsoon season usually reaches here.However, in this region, TRMM V5 3B42 underestimated therainfall, especially during monsoon months. Dhaka is located inthe center of the country. Some rainfall events occurred during

Fig. 2. Daily rainfall (DRF) measured by RNG and TRMM at five selected stations (a) Teknaf, (b) Dhaka, (c) Sylhet, (d) Rangpur, and (e) Khulna in 2001. Verticallines have been used to divide the entire rainy season into three periods.

267M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

the pre-monsoon months during the observed period, andrainfall measured by TRMM V5 3B42 compared well to that ofthe RNG values. Sylhet is located in the northeastern part of thecountry, and heavy rainfall was recorded in this region from thepre-monsoon to monsoon periods. The pre-monsoon rainfallmeasured by TRMM V5 3B42 is almost comparable with thatof the RNG values of Sylhet, but, during the monsoon period,TRMM V5 3B42 underestimated the rainfall. In the northwestRangpur, the V5 3B42 calculation of rainfall was quitecomparable with the RNG values, whereas, in the monsoonperiod, the V5 3B42 estimation did not compare well with thepoint values. In some cases of 2001 (also 1999 and 2002, notshown), the estimation patterns of TRMM V5 3B42 showedgood similarity with the RNG values, but the amount of

deviation was very high, which may be due to the short life ofthe cloud cells and the single pass of TRMM V5 3B42 duringthe 24-h period. In the southwest Khulna, the average rainfalldetected by both TRMM V5 3B42 and RNG in 2001 was quitecomparable. In some cases in 1999 and 2000 and many cases in2002, the TRMM V5 3B42 estimation was very low incomparison to the RNG values (not shown), supporting theconclusion that the TRMM V5 3B42 estimation is very low forheavy-rainfall events.

Daily rainfall (DRF) measured by TRMM V5 3B42 andRNG at Teknaf in 1999 and 2002 is shown in Fig. 3 as anexample. Rainfall and rainfall biases (bias=TRMM−RNG) inall analyzed years (1998–2002) and at all 5 stations aresummarized in Table 1. As in Fig. 3, the TRMM V5 3B42 and

Fig. 3. Same as Fig. 2 but at Teknaf in (a) 1999 and (b) 2002.

268 M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

RNG patterns are quite similar, but rainfall measured by TRMMV5 3B42 differs substantially from the RNG values. TRMMV53B42 fails to detect rain intensities with biases of about −6.5and −4.8 mm/day in 1999 and 2002, respectively (Table 1). Thebiases at Teknaf are −6.8, −8.7, and −5.3 mm/day in 1998,2000, and 2001, respectively. On average, for the 5-yearanalysis, the bias was about 1.7 mm/day at Dhaka, 1.9 mm/dayat Khulna, −1.7 mm/day at Rangpur, −3.2 mm/day at Sylhet,and −6.4 mm/day at Teknaf. Therefore, of the 5 locations, V53B42 fails to determine rainfall at 3 (negative bias) andsucceeded or overrated at 2 (positive bias). In 1999, dailyrainfall measured by V5 3B42 and RNG was 10.4 and 16.9 mm/day, respectively, at Teknaf, whereas the values were 9.2 and14 mm/day in 2002. Therefore, in 1999 and 2002, TRMM V53B42 underestimated 6.5 and 4.8 mm/day, respectively. Hence,it is obvious that the magnitude of the errors in rainfallmeasurement from TRMMV5 3B42 is largely dependent on thecalculation site and period.

Table 1Rainfall and rainfall biases calculated from TRMM V5 3B42 and RNG rainfall avBangladesh

Rainfall and rainfall biases (mm/day)

Dhaka Khulna Rangpur

RNG TRMM Bias RNG TRMM Bias RNG T

1998 8.0 10.1 2.0 6.7 9.1 2.4 8.8 81999 8.8 9.1 0.3 6.7 9.2 2.5 10.7 72000 7.7 8.7 1.0 6.0 8.5 2.5 6.0 72001 6.0 8.8 2.8 5.9 8.4 2.5 9.0 62002 6.8 9.0 2.1 9.4 9.1 −0.3 11.5 7Ave. 7.5 9.1 1.7 6.9 8.8 1.9 9.2 7

The comparative bar diagram of Fig. 4 shows the dailyrainfall measured from TRMM V5 3B42 and RNG data at fiveselected stations for 5 years. The daily rainfall from TRMM V53B42 data is almost always higher than the RNG value at Dhakaand Khulna. In the case of the three other sites, the opposite wasobserved. Therefore, TRMM V5 3B42 underestimated therainfall at locations that lie within the eastern and northern partsof the country and overestimated the rainfall at locations that liewithin the central and southwestern parts of the country. Sametype of variability for day-to-day TRMM data versus rain-gaugedata over India is found by Brown (2006). Therefore, careshould be taken for the application of V5 3B42 data to short-term hydrological problems such as flash flood.

Fig. 5 shows the comparison of daily average rainfallthroughout the country in 1998 and 2001 determined fromTRMM V5 3B42 and RNG. Table 2 shows the date of themonsoon onset, available for 3 stations as determined by theBMD, and monsoon rainfall climatology. The date of the

eraged from 1 March to 30 November in each year at different stations over

Sylhet Teknaf

RMM Bias RNG TRMM Bias RNG TRMM Bias

.7 0.0 14.9 10.8 −4.1 15.3 8.5 −6.8

.9 −2.8 12.7 11.1 −1.6 16.9 10.4 −6.5

.3 1.3 16.2 10.1 −6.2 16.8 8.1 −8.7

.7 −2.4 11.9 10.4 −1.5 16.5 11.2 −5.3

.0 −4.5 13.2 10.9 −2.3 14.0 9.2 −4.8

.5 −1.7 13.8 10.7 −3.2 15.9 9.5 −6.4

Fig. 4. Rainfall (mm/day) measured from TRMM 3B42 and rain-gauge (RNG) data for a 5-year period at each selected station.

269M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

monsoon onset was delayed in 1998 and early in 2001. Pre-monsoon rainfall prevailed in 1998, while rainfall was deficientduring the peak monsoon months, particularly, in July and

Fig. 5. Same as Fig. 2 except that rainfall is averaged from 31 stations throughout Bmonsoon onset throughout the country as determined by the BMD.

August, 2001. The DRF patterns between the RNG and TRMMV5 3B42 values have good similarity during the observedperiod, but there is a slight deviation in calculating the average

angladesh in (a) 1998 and (b) 2001. The down arrow indicates the date of the

Table 2Dates of monsoon onset determined by BMDwith monsoon rainfall climatology

Year Onset date Remarks

Teknaf Sylhet Dhaka Onset Monsoonrainfall anomaly

Rainfallclimatology

1998 7 June 9 June 14 June delayed −2% Normalmonsoon year

1999 25 May 27 May Missing early 0% Normalmonsoon year

2000 29 May 03 June 8 June on time −23% Dry monsoonyear

2001 26 May 03 June 31 May early −12% Dry monsoonyear

2002 03 June 06 June 8 June on time −7% Normalmonsoon year

Fig. 6. Comparison of the rainfall estimated by TRMM and RNG in differentyears averaged for 31 stations throughout the country. Rainy days (in %)determined from RNG and TRMM in different years and averaged for 1998–2002. A rainy day is defined as any day that has a measurable amount of rainfall,and a match day is any day detected as rainy by both TRMM and RNG.

270 M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

rainfall; particularly, during the pre-monsoon and post-monsoonmonths, the V5 3B42 estimates are higher than those of RNG,whereas, during the monsoon months, the V5 3B42 estimatesare lower. Hence, it is again noticeable that the magnitude of theerrors in rainfall determination from TRMMV5 3B42 is largelydependent on the calculation period or on the season of theoccurrence of precipitation.

Fig. 6 shows the comparison of rainfall (mm/day) estimatedby TRMM V5 3B42 and RNG and the percentage of rainy days(in %) and match days (in %) in different years averaged for the31 stations throughout the country. The V5 3B42 estimation isvery close to the RNG values for country-scale averaged dailyrainfall, and the variation in the rainfall lies between 0.08 and0.51 mm. In 2000, measurements between these two systemsare very close, but, in 1998, 1999, and 2002, TRMM V5 3B42slightly underestimates the rainfall, while, in 2001, it slightlyoverestimates it. As explained in Fig. 5, 1998 was a normalmonsoon year, and, in 2001, there was lack of rainfall during thepeak monsoon months. TRMM V5 3B42 underestimates therainfall in the normal monsoon year and overestimates it in anunusual monsoon year during which there is a lack of rainfallduring the monsoon months. The rainfall anomaly (individualannual average rainfall minus the annual average value) liesbetween −0.61 and 0.52 mm/day for V5 3B42 values andbetween −0.41 and 0.52 mm/day for RNG values. On average,from the 5-year analysis, the daily rainfall calculated by TRMMV5 3B42 and RNG is 8.12 and 8.34 mm, respectively.Therefore, the TRMM V5 3B42 estimation is very close tothe RNG values for daily rainfall. The number of rainy daysdetermined by TRMM V5 3B42 in 1998 and 2002 was higherthan that for RNG and lower in 1999 and 2000. However, thenumber of rainy days determined by RNG and TRMMV5 3B42obtained on average for 5 years (1998–2002) was almost thesame as and very close to that for match days, respectively.There are always fewer match days than rainy days in individualyears. The average percentage of rainy days determined byTRMM V5 3B42 data with respect to the RNG value is 96%.Hence, TRMM V5 3B42 succeeds in detecting rainy days verywell. Therefore, TRMM V5 3B42 can be used as a tool fordetecting rainy days in the parts of Asia, such as Bangladesh,that experience monsoons.

3.3. Rain climatology obtained from TRMM V5 3B42 and RNG

3.3.1. Rain climatology at five selected stations overBangladesh

Average daily rainfalls (DRFs) for the 5-year period (1998–2002) at Khulna and Teknaf are shown in Fig. 7. From theanalysis of the average daily rainfall, the estimated value byTRMM V5 3B42 is comparable to the RNG value at Khulna(Fig. 7(a)) and Dhaka (not shown) except on some days inwhich TRMM V5 3B42 underestimated it. However, at Teknaf(Fig. 7(b)), TRMM V5 3B42 underestimated the rainfall duringthe monsoon period. The same situation was observed at Sylhetand Rangpur (not shown). Thus, the calculation site and periodare shown again to be important factors in estimating rainfallfrom TRMM V5 3B42 data.

3.3.2. Rain climatology in different rainy periods inBangladesh

The daily average rainfall determined by TRMM V5 3B42and RNG for 31 locations during the three observed periods ofthe rainy season (pre-monsoon, monsoon, and post-monsoonperiods) is described below.

3.3.2.1. Pre-monsoon. Daily rainfall measured by TRMM V53B42 together with that of the RNG value at 31 locations duringthe pre-monsoon period showed that V5 3B42 overestimated(0.6 to 2.6 mm/day) the rainfall at 22 locations, had the almostsame (−0.2 to 0.4 mm/day) measurement at 7 locations, andunderestimated (1.9 to 1.8 mm/day) it at 2 locations (notshown). In this season, the rainfall determined by bothprocesses at different locations throughout the country had adeviation within −1.9 to 2.6 mm/day in some isolated places.

3.3.2.2. Monsoon. Daily average monsoon rainfall valuesmeasured by TRMM V5 3B42 and RNG at 31 locationsthroughout Bangladesh are shown in Fig. 8. During this period,

Fig. 7. Same as Fig. 2 except that the averages are for 5 years (1998–2002) and at (a) Khulna and (b) Teknaf.

Fig. 8. Daily rainfall determined from RNG (left of plus mark) and TRMM (rightof plus mark) at different stations (plus mark). Data averaged for 1 June to 30September from 1998–2002. The shadowed region represents the region wherethe rainfall was underestimated by TRMM.

271M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

the V5 3B42- and RNG-estimated rainfall values in the centraland western parts of the country are almost the same, with onlya small variation. However, TRMM V5 3B42 underestimatedthe rainfall in the eastern, southern, and northern parts, with avariation of up to ∼13 mm (the shadowed region), and theseare the regions through which monsoon progression occurredas indicated in Fig. 1. In this period, TRMM V5 3B42overestimated (0.7 to 4.7 mm/day) the rainfall at 14 locations,had the almost same measurement (0 to 0.6 mm/day) at 2locations, and underestimated (0.6 to 13.3 mm/day) it at 15locations.

3.3.2.3. Post-monsoon. During the post-monsoon period,daily average rainfall measured by TRMM V5 3B42 andRNG throughout the country showed that both techniquesdetermined rainfall with almost the same accuracy. In thisperiod, V5 3B42 overestimated (0.7 to 1.2 mm/day) the rainfallat 10 locations, had the almost same (− .05 to 0.3 mm/day)measurement at 13 locations, and underestimated (0.7 to1.6 mm/day) it at 8 locations (not shown). The deviation ofthe rainfall measurement was within −1.6 to 1.2 mm/day insome isolated places.

The rain climatology obtained from TRMM V5 3B42 andRNG rainfall averaged for 5 well-separated stations and for 31stations throughout the country for the years 1998–2002 istabulated in Table 3.

The rain climatology in different rainy periods showsthat the daily average rainfall calculated by using the V53B42 and RNG data has good similarity in the pre-monsoon and post-monsoon period and the entire rainy

period (March–November) for 5 well-separated stations and31 stations of Bangladesh. The overall estimation by V53B42 is about 97.36% of the RNG rainfall. Hence, TRMMmay be used as a successful tool for rainfall estimation inBangladesh.

Table 3Average daily rainfall calculated from TRMM V5 3B42 and RNG data in thepre-monsoon, monsoon, and post-monsoon period and the entire rainy period(March–November)

Average daily rainfall (mm/day) for 1998–2002

RNG-5st5Yr

TRMM-5st5Yr

RNG-31st5Yr

TRMM-31st5Yr

Pre-monsoon 6.50 6.47 5.30 6.40Monsoon 17.01 13.55 15.14 13.43Post-monsoon 4.29 4.32 4.56 4.54March–November 9.27 8.11 8.34 8.12

Averages from 5 well-separated stations (5st5Yr) and 31 stations (31st5Yr) overBangladesh.

272 M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

3.3.3. Long-term rain climatology derived from TRMM inBangladesh

The time sequences of the average DRF for five stations(Fig. 9(a)) show that there is a good comparison between theTRMM V5 3B42 values and the RNG values, with theexception of the mid-June to mid-August period, when V53B42 fails to obtain precise estimates. Therefore, TRMM V53B42 data can be used for long periodic rainfall estimation aswell as to prepare rainfall climatology. The average dailyrainfall determined by the TRMM V5 3B42 and RNG datasetat 31 locations during the observed period (1998–2002) isshown in Fig. 9(b). The patterns of DRF determined by bothprocesses depict good agreement in the case of rainfall

Fig. 9. Daily rainfall measured by rain-gauge (RNG) and TRMM 3B42. Averages

amounts as well as the variation of average rainfall. Hence,the comparison is much better for the average from a largenumber of sites and over the long term.

3.3.4. Accuracy of TRMM estimates in BangladeshThere are several potential problems that may cause the

TRMM V5 3B42 algorithms to fail, many of which are relatedto precipitation characteristics in a particular region and aparticular period, i.e., the measurement depends on the site andthe period. There are many possible reasons for the discrepan-cies between the RNG and TRMM measurements, such as thefact that TRMM-PR measures the exact rain rate, whilemeasurements from the TMI depend on the rain characteristicsin pre-monsoon, monsoon, and post-monsoon periods, and thecalculation sites, such as whether they are in heavy- or low-rainfall regions. TRMM V5 3B42 overestimated 20.73% andunderestimated 11.3% and 0.54% of the surface rain during thepre-monsoon, monsoon, and post-monsoon periods, respective-ly. The percentages of TRMM V5 3B42 to RNG rainfall are98.27%, 97.54%, 100.4%, 100.92%, and 94.59% in 1998,1999, 2000, 2001, and 2002, respectively, as explained in Fig.6. It is reasonable to say that, on an annual scale, the RNG andTRMM V5 3B42 estimates agree to within 94.59–100.92%with an average of 98.34% for the 5-year period. In knownheavy-rainfall regions, such as the northeastern and southeast-ern parts of Bangladesh, for example, V5 3B42 data result inlarge differences, and V5 3B42 fails to accurately estimate the

from 1998–2002 for (a) 5 stations and (b) 31 stations throughout Bangladesh.

Fig. 11. Rainfall (mm/day) averaged for the wet region (high humidity) and thedry region (low humidity). Data averages for 17 stations in the wet region and 14stations in the dry region throughout Bangladesh from 1998 to 2002.

273M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

rainfall. In the central and western parts of the country, TRMMV5 3B42 overestimates the rainfall. The drop size distributionsmay play a large role in these regional/temporal differences(Berg, 2004). A good review of error sources and performancecharacteristics of microwave sensors can be found in Petty(1995). Additional studies will be necessary to refine theseresults using radar rain rates and detailed characteristics ofprecipitation structures using TRMM-PR 2A25 and V6 3B42records in Bangladesh.

4. Discussion

Daily average rain rates estimated by TRMM V5 3B42 dataare lower than the RNG values in the regions of northeasternand southeastern Bangladesh (shown in Fig. 1), which are theregions with the heaviest rainfall (Islam et al., 2005). Ananalysis of humidity anomalies (individual station averageminus the average from all stations), as shown in Fig. 10, showsthat the eastern, southern, and northern parts of the country aremore humid than the central and western parts. The less humidarea lies on the southeastern boundary and is hilly. Based onexcess (positive anomaly) and deficit (negative anomaly)average humidity, Bangladesh can be divided into two regions:wet and dry. The sites located in the regions of positive andnegative anomalies can be applied to determine the wet and dryregions, respectively. The wet region is also in the path of themonsoon progression over the country.

The average DRF measured by both TRMM and RNGtechniques in the wet and dry region is shown in Fig. 11.Averages are shown from 17 sites located in excessively humidregions in the wet region and from 14 sites located in deficiently

Fig. 10. Anomaly of the average daily humidity (in %) over Bangladesh.Averages from 1 March to 30 November in each year for 1998–2002. Positiveand negative values represent surpluses and deficits of humidity, respectively.Positive and negative values also represent wet and dry regions, respectively.

humid regions in the dry region during 1998–2002. TRMM V53B42 underestimated 0.97 mm/day in the wet region butoverestimated 0.67 mm/day in the dry region. In other words,V5 3B42 underestimated 10.08% of the surface rain in the wetregion and overestimated 9.82% in the dry region. On average,TRMM V5 3B42 can determine 97.36% of the surface rain inBangladesh.

The structure of the precipitation fields and the strength ofthe rain rate in Bangladesh have been observed using TRMM2A25 data records. The available data from the Plan PositionIndicator (PPI) scans of BMD radar were used to see the realsituation of precipitation field. BMD installed an S-bandweather radar (wavelength: ∼10 cm) at Dhaka (see Fig. 1) onthe roof of a building with a height of 60 m, which provides acoverage of 600 km by 600 km rectangular area. BMD radarestimated rainfall was calibrated with surface observation byIslam et al. (2005). The typical structure of pre-monsoon andmonsoon period precipitation fields has been analyzed and isshown in Fig. 12. The precipitation fields had intensities of up to65 mm/h in the southeastern part of Bangladesh on 23 May2002 at 19:14 LST in the pre-monsoon season. The PPI scans ofBMD radar at 18:04 LST indicated the development ofconvection on the same day at the same location. The verysouthern part of the domain is outside of the radar coverage. Thevertical profile of the convective cloud along line AB hadvertical extensions of about 17.5 km and a maximum intensityof 45 mm/h. An example of a similar case of a cloud thatdeveloped with vertical extensions of 19 km and rain ratesbeyond 30 mm/h was observed in May 1998, as reported in theTRMM report (2002). During the monsoon period, the rain ratehad a lower value of about 13 mm/h on 31 July 2001 at 08:44LST. The vertical extension was about 11.5 km. The BMD radarPPI scan at 08:46 LST on the same date confirmed thedevelopment of this event. Hence, it is clear that tall and intenseconvective clouds developed during the pre-monsoon seasonwhereas, in the monsoon period, less intense short convectiveclouds developed in Bangladesh. Detail of the vertical structureof precipitation developed over Bangladesh in different seasonsis explained in Islam and Uyeda (2006). Kodama et al. (2005)also showed that the mean echo top height in areas with rain ishigher during the pre-monsoon than during the monsoon. Hence

Fig. 12. Typical structure of the precipitation field of (a) pre-monsoon: 23 May 2002 at 19:14 LST and (b) monsoon: 31 July 2001 at 08:44 LST. The available datafrom the PPI scan of BMD at 18:04 LSTand the vertical profile of the precipitation field along line AB are also shown in panel (a). The available data from the PPI scanof BMD at 08:46 LST and the vertical cross section of the precipitation field along line AB are shown in panel (b). The dashed lines show the TRMM pass.

274 M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

it is clear that convection is deeper and more intense during thepre-monsoon. Pre-monsoon rainfall is characterized by moreconvective rain and after monsoon onset, stratiform and shallowrains become more common (Kodama et al., 2005). Theshallow, isolated echoes seen by the TRMM PR probablyrepresent warm rain processes and, as such, should therefore beclassified as convective (Schumacher & Houze, 2003). In fact,TRMM V5 3B42 could successfully detect the development oftall convection during the pre-monsoon period, whereas it failedto detect the development of low-level short convection duringthe monsoon period. It is reasonable to say that, during themonsoon period, the low-level monsoon flow (Zuidema, 2003)carrying water vapor from the Bay of Bengal (Bookhagen &Burbank, 2006) assisted in the development of low-level shortconvections over the monsoon progression region of thecountry. The indication of the westward propagation ofmonsoon depression is also described in Chen et al. (2005).

Fig. 13 shows the seasonal cycle of rainfall in Bangladesh forall rain estimates. All estimates show single peak in July exceptfor V5 3B42 which shows the peak in June. This single peak isthe historical rainfall climatology in Bangladesh that differsfrom the double peak obtained by Chokngamwong and Chiu(2006) for Thailand. However, the difference of rainfallestimated by V5 3B42 in June and July is very little(0.43 mm/day). The V6 3B43 shows low biases compared tothe other products. This is consistent with Chokngamwong andChiu (2006) for Thailand. During all seasons, V6 3B43estimates are close to RNG estimates. The V5 3B42 over-estimates the gauge measurement in pre-monsoon and post-monsoon months whereas it underestimates during monsoonmonths. On the other hand, V5 3B43 underestimates duringApril to October and overestimates in other months. The V63B42 follows the same trend of V6 3B43 with a little negativebias. It underestimates in all months except overestimates in

Table 4bTRMM 3B42 False detection of rainy days for 1998–2002

V5 3B42_98-02 V6 3B42_98-02

Approximately equal (±10%) 14 142Under-detection 11–20% 0 10

21–30% 0 0>30% 0 0Total 0 10

Over-detection 11–20% 113 121–30% 22 0>30% 4 0Total 139 1

Missing stations 2 2Total stations 155 155

Fig. 13. Comparison of seasonal rainfall climatology in Bangladesh. RNG,TRMMV5 3B42, V5 3B43 and V6 3B43 are computed from respective datasetsfor1998–2002.

275M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

pre-monsoon month March, post-monsoon month Novemberand dry month December.

The performance of the TRMM product is summarized inTable 4a,b for the year 1998–2002. To find the distribution of31 stations in 5 years, 155 (=31×5) individual stations are to beconsidered in Bangladesh. The values in each row under year inTable 4a shows the number of stations either underestimated,overestimated, or approximately equal of the stations observedvalues for daily rainfall. These results are consistent with Brown(2006) who analyzed TRMM data for India/Sri Lanka. TheTRMMV5 3B42-, V5 3B43-, V6 3B42- and V6 3B43 productswere accurate to within 10% of the analyzed stations only about18%, 34%, 38% and 43%, respectively of the time. Byexpanding definition of acceptable accuracy to be ±25% ofobserved precipitation (Brown, 2006), the TRMM V5 3B42-,V5 3B43-, V6 3B42 and V6 3B43 products were accurate for52%, 68%, 72% and 84%, respectively of station locations. Thevalues in each row under year in Table 4b show the number ofstations either under-detected, over-detected, or approximatelyequal of the stations observed values for false detection of dailyrainfall by TRMM 3B42. The TRMM V5 3B42 and V6 3B42

Table 4aTRMM 3B42 and 3B43 Performances for 1998–2002

V53B42_98-02

V53B43_98-02

V63B42_98-02

V63B43_98-02

Approximatelyequal

(±10%) 27 52 58 66

Underestimation 11–20% 16 22 38 3721–30% 17 17 25 1831–40% 19 18 8 5>40% 6 8 0 0Total 58 65 71 60

Overestimation 11–20% 23 17 14 1421–30% 15 6 7 831–40% 10 6 1 4>40% 20 7 2 1Total 68 36 24 27

Missing stations 2 2 2 2Total stations 155 155 155 155

products were accurate to within 10% of the analyzed stationsabout 9% and 93%, respectively of the time. For V6 3B42under-detection was 7% that is much better than over-detection91% for V5 3B42. Hence, it is clear that the performance of V63B43 is better than that of other TRMM datasets. However, forthe estimation of daily rainfall TRMM 3B42 data products areessential, especially for flood monitoring. The V6 3B42 isfound better than V5 3B42. This is consistent with the result ofChokngamwong and Chiu (2006) for Thailand. Therefore, theTRMM V6 data products are suggested for further study tosettle on TRMM data as a tool in estimating rainfall inBangladesh.

5. Conclusions

In this analysis, the mean fields for rainfall for the pre-monsoon, monsoon, post-monsoon, and entire rainy season overBangladesh are presented on the basis of a 31-station rain-gauge(RNG) dataset and TRMM products for the period of 1998–2002. On a seasonal scale (March–November), the averagevalues of rainfalls determined by TRMMV5 3B42 and RNG are8.12 and 8.34 mm/day, respectively. Therefore, the TRMMsatellite was able to measure 97.36% of the surface rain inBangladesh. TRMMV5 3B42 overestimated the rain rates in thepre-monsoon period but underestimated them in the monsoonperiod. In the post-monsoon period, the rain rates measured byV5 3B42 and RNG were nearly identical. The V6 3B42 productreduces the overestimation and underestimation. It under-estimates in all months except overestimates in April, Novemberand December. This analysis revealed that V5 3B42 under-estimated the calculation of average rainfall in the humid andheavy-rainfall regions (the eastern and northeastern parts) ofBangladesh but overestimated the rainfall in the dry and light-rainfall regions (the central and western parts of the country).Rainy days counted from the TRMM3B42 data matched 96% ofthe RNG data. The TRMM 2A25 precipitation field confirmedthat the TRMM Microwave Imager (TMI) in the 3B42 productaccurately detected the rain rate from the strong and tall pre-monsoon convections but failed to detect precise rain rates fromweak and short monsoon convections. The cloud height of about17.5 km was observed during the pre-monsoon period and wasabout 11.5 km during the monsoon period. The verticalextension of clouds measured by TRMM 2A25 records showed

276 M.N. Islam, H. Uyeda / Remote Sensing of Environment 108 (2007) 264–276

good agreement with the real situation. Overall, from all TRMMproducts analyzed in this study, the performance of V6 3B43 isfound better compared to other. Long periodic analysis showedthat TRMM data can be used to better understand the rainclimatology of tropical regions, such as Bangladesh. Therefore,in Bangladesh, TRMM is good for long-term climatology withapplication to strategic water resource management and it is notso good for shorter term hydrological applications such as floodforecasting.

Acknowledgments

The authors would like to thank the BMD for providing radarand RNG data. Mr. Yamamoto and Shingo Shimizu of HyARC,Nagoya University, Japan, are gratefully acknowledged for theirassistance with copying TRMM 2A25 data. Thanks are alsogiven to Dr. A.K.M. Saiful Islam, IWFM, BUET, for hisinvaluable help with the TRMM 2A25 data processing of Fig.12. The TRMM 2A25 data were provided by JAXA fromHyARC, Nagoya University, Japan. The TRMM 3B42 datawere acquired from their website. TRMM is an internationalproject jointly sponsored by the Japan National SpaceDevelopment Agency (NASDA) and the U.S. NationalAeronautics Space Administration (NASA) Office of EarthScience. This work was supported by TRMM-RA4 of JAXA.

References

Ahmed, R., & Karmakar, S. (1993). Arrival and the withdrawal dates of thesummer monsoon in Bangladesh. International Journal of Climatology,Royal Meteorological Society, 13, 727−740.

Barros, A. P., Joshi, M., Putkonen, J., & Burbank, D. W. (2000). A study of the1999 monsoon rainfall in a mountainous region of central Nepal usingTRMM products and rain gauge observations. Geophysical ResearchLetters, 27, 3683−3686.

Berg, W. (2004). The Relationship of TRMM Rainfall Biases to the SynopticEnvironment. 2nd International TRMM Science Conference, Nara Prefec-ture, Japan, Sep. 6–10, 2004.

Bookhagen, B., & Burbank, D. W. (2006). Topography, relief, and TRMM-derived rainfall variations along the Himalaya. Geophysical ResearchLetters, 33, L08405. doi:10.1029/2006GL026037

Brown, J. E. M. (2006). An analysis of the performance of hybrid infrared andmicrowave satellite precipitation algorithms over India and adjacent regions.Remote Sensing of Environment, 101, 63−81.

Chen, T. C., Yoon, J. H., & Wang, S. Y. (2005). Westward propagation of theIndian monsoon depression. Tellus, 57(A), 758−769.

Chokngamwong, R., & Chiu, L. (2006). TRMM and Thailand daily gaugerainfall comparison, P1.2. The 86th AMS annual meeting, 28 Jan.–2 Feb,Atlanta, GA. H23A-10 (http://ams.confex.com/ams/pdfpapers/103039.pdf )

Das, P. K. (1995). The monsoons (pp. 61)., 3rd ed. New Delhi, India: NationalBook Trust published by the Director.

Huffman, G., Adler, R., Arkin, P., Chang, A., Ferraro, R., Gruber, A., et al. (1997).The Global Precipitation Climatology Project (GPCP) combined precipitationdataset. Bulletin of the American Meteorological Society, 78, 5−20.

Huffman, G., Adler, R., Bolvin, D., & Nelkin, E. (2004). Uncertainty in fine-scale MPA precipitation estimates and implications for hydrometeorologicalanalysis and forecasting. 18th conf. on hydrology, 11–18 January, 2004,Seattle, W.A.

Huffman, G. J., Adler, R. F., Rudolf, B., Schneider, U., & Keehn, P. R. (1995).Global precipitation estimates based on a technique for combining satellite-based estimates, rain gauge analysis and NWP model precipitationinformation. Journal of Climate, 8, 1284−1295.

Ikai, J., & Nakamura, K. (2003). Comparison of rain rates over the oceanderived from TRMMMicrowave Imager and Precipitation Radar. Journal ofAtmospheric and Oceanic Technology, 20(12), 1709−1726.

Islam, M. N., Terao, T., Uyeda, H., Hayashi, T., & Kikuchi, K. (2005). Spatialand temporal variations of precipitation in and around Bangladesh. Journalof the Meteorological Society of Japan, 83(1), 23−41.

Islam, M. N., & Uyeda, H. (2005). Comparison of TRMM 3B42 products withsurface rainfall over Bangladesh. Preprints in the proceedings ofInternational Geoscience and Remote Sensing Symposium (IGARSS)2005, Seoul, South Korea, July 25–29 (pp. 4112−4115). 4TH127A_05.

Islam, M. N., & Uyeda, H. (2006). TRMM observed vertical structure anddiurnal variation of precipitation in south Asia. Preprints in the proceedingsof International Geoscience and Remote Sensing Symposium (IGARSS)2006, Denver, USA, 31 July-4 August (pp. 4).

Kodama, Y. -M., Ohta, A., Katsumata, M., Mori, S., Satoh, S., & Ueda, H.(2005). Seasonal transition of predominant precipitation type and lightningactivity over tropical monsoon areas derived from TRMM observations.Geophysical Research Letters, 32, L14710. doi:10.1029/2005GL022986

Kummerow, C., Simpson, J., Thiele, O., Barnes, W., Chang, A. T. C., Stocker,E., et al. (2000). The status of the Tropical Rainfall Measuring Mission(TRMM) after two years in orbit. Journal of Applied Meteorology, 39,1965−1982.

Legates, D. R., &Willmott, C. J. (1990). Mean seasonal and spatial variability ingauge-corrected, global precipitation. Interantional Journal of Climate, 10,111−127.

Morrissey, M. L., & Greene, J. S. (1993). Comparison of two satellite-basedrainfall algorithms using atoll rain gauge data. Journal of AppliedMeteorology, 32, 411−425.

Morrissey, M. L., Shafer, M. A., Postawko, S., & Gibson, B. (1995). Pacific raingauge data. Water Resources Research, 31, 2111−2113.

Nicholson, S. E., Some, B., McCollum, J., Nelkin, E., Klotter, D., Berte, Y., et al.(2003). Validation of TRMM and other rainfall estimates with a high-densitygauge dataset for West Africa: Part I. Validation of GPCC rainfall productand pre-TRMM satellite and blended products. Journal of AppliedMeteorology, 42, 1354−1377.

Ohsawa, T., Ueda, H., Hayashi, T., Watanabe, A., & Matsumoto, J. (2001).Diurnal variations of convective activity and rainfall in tropical Asia.Journal of the Meteorological Society of Japan, 79, 333−352.

Petty, G. W. (1995). The status of satellite-based rainfall estimation over land.Remote Sensing of Environment, 51, 125−137.

Rudolf, B., Hauschild, H., Rüth, W., & Schneider, U. (1994). Terrestrialprecipitation analysis: Operational method and required density of pointmeasurements. In M. Dubois, & F. Desalmand (Eds.), Global precipitationsand climate change (pp. 173−186). Springer Verlag.

Schumacher, C., & Houze, R. A., Jr. (2003). The TRMM precipitation radar'sview of shallow, isolated rain. Journal of Applied Meteorology, 42(10),1519−1524.

Sevruk, B. (1982). Methods of correction for systematic error in pointprecipitation measurement for operational use. WMO Operational Hydrol-ogy Rep. 21, WMO Rep. No. 589, 91, 1982.

Simpson, J., Adler, R. F., & North, G. R. (1988). A proposed Tropical RainfallMeasuring Mission (TRMM) satellite. Bulletin of the American Meteoro-logical Society, 69, 278−295.

TRMM report (2002). In “Rain as seen from space”, production committee(Eds.), Rain as seen from space (1st ed.). Japan Advanced Plan, 31 May2002, pp. 26.

Wolff, D. B., Marks, D. A., Amitai, E., Silberstein, D. S., Fisher, B. L., Tokay,A., et al. (2005). Ground validation for the Tropical Rainfall MeasuringMission (TRMM). Journal of Atmospheric and Oceanic Technology, 22,365−379.

Xie, P., & Arkin, P. A. (1995). An intercomparison of gauge observations andsatellite estimates of monthly precipitation. Journal of Applied Meteorology,34, 1143−1160.

Xie, P., & Arkin, P. A. (1996). Analysis of global monthly precipitation usinggauge observations, satellite estimates, and numerical model predictions.Journal of Climate, 9, 840−858.

Zuidema, P. (2003). Convective clouds over the Bay of Bengal. MonthlyWeather Review, 131, 780−798.