• Anagnostou E. N., W. F. Krajewski & J. A. Smith 1999a Uncertainty quantification of mean-areal radar-rainfall estimates J. Atmos. Oceanic Technol. 16 206 215

    • Search Google Scholar
    • Export Citation
  • Anagnostou E. N., A. J. Negri & R. F. Adler 1999b Statistical adjustment of satellite microwave monthly rainfall estimates over Amazonia J. Appl. Meteor. 38 1590 1598

    • Search Google Scholar
    • Export Citation
  • Barrett E. C., J. Doodge, M. Goodman, J. Janowiak, E. Smith & C. Kidd 1994 The First WetNet Precipitation Intercomparison Project (PIP-1) Remote Sens. Rev. 11 49 60

    • Search Google Scholar
    • Export Citation
  • Bell T. L., A. Abdullah, R. L. Martin & G. R. North 1990 Sampling errors for satellite-derived tropical rainfall: Monte Carlo study using a space–time stochastic model J. Geophys. Res. 95 2195 2206

    • Search Google Scholar
    • Export Citation
  • Chang A. T. C. & L. S. Chiu 1999 Nonsystematic errors of monthly oceanic rainfall derived from SSM/I Mon. Wea. Rev. 127 1630 1638

  • Chiu L. S., D. Short, A. McConnel & G. North 1990 Rain estimation from satellites: Effect of finite field of view J. Geophys. Res. 95 2177 2185

  • Ha E. & G. R. North 1999 Error analysis for some ground validation designs for satellite observations of precipitation J. Atmos. Oceanic Technol. 16 1949 1957

    • Search Google Scholar
    • Export Citation
  • Huffman G. J. 1997 Estimates of root-mean-square random error for finite samples of estimated precipitation J. Appl. Meteor. 36 1191 1201

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

    • Search Google Scholar
    • Export Citation
  • Huffman G. J. Coauthors, 1997 The Global Precipitation Climatology Project (GPCP) combined precipitation dataset Bull. Amer. Meteor. Soc. 78 5 20

    • Search Google Scholar
    • Export Citation
  • Huffman G. J., M. M. Morrissey, S. Curtis, R. Joyce, B. McGavock & J. Susskind 2001 Global precipitation at one-degree daily resolution from multisatellite observations J. Hydrometeor. 2 36 50

    • Search Google Scholar
    • Export Citation
  • Iguchi T., T. Kozu, R. Meneghini, J. Awaka & K. Okamoto 2000 Rain-profiling algorithm for the TRMM precipitation radar J. Appl. Meteor. 39 2038 2052

  • Ilesanmi O. O. 1971 An empirical formulation of an ITD rainfall model for the Tropics: A case study of Nigeria J. Appl. Meteor. 10 882 891

  • Kedem B., L. S. Chiu & G. N. North 1990 Estimation of mean rain rate: Application to satellite observations J. Geophys. Res. 95 1965 1972

  • Kedem B., R. Pfeiffer & D. A. Short 1997 Variability of space–time mean rain rate J. Appl. Meteor. 36 443 451

  • Kousky V. E. 1980 Diurnal rainfall variation in northeast Brazil Mon. Wea. Rev. 108 488 498

  • Kummerow C. 1998 Beamfilling errors in passive microwave rainfall retrievals J. Appl. Meteor. 37 356 357

  • Kummerow C. Coauthors, 2000 The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit J. Appl. Meteor. 39 1965 1982

  • Morrissey M. L. & J. E. Janowiak 1996 Sampling-induced conditional biases in satellite climate-scale rainfall estimates J. Appl. Meteor. 35 541 548

    • Search Google Scholar
    • Export Citation
  • Philip G. 1999 New School Atlas and Gazetteer George Philip, 176 pp

  • Rudolf B., H. Hauschild, W. Rueth & U. Schneider 1994 Terrestrial precipitation analysis: Operational method and required density of point measurements Global Precipitations and Climate Change, M. Desbois and F. Desalmand, Eds., NATO ASI Series I, Vol. 26, Springer-Verlag, 173–186

    • Search Google Scholar
    • Export Citation
  • Rudolf B., H. Hauschild, W. Rueth & U. Schneider 1996 Comparison of raingauge analyses, satellite-based precipitation estimates and forecast model results Adv. Space Res. 18 7 53 62

    • Search Google Scholar
    • Export Citation
  • Rudolf B., T. Fuchs, W. Rueth & U. Schneider 1998 Precipitation data for verification of NWP model re-analyses: The accuracy of observational results Proc. Int. Conf. on Reanalyses, Washington, DC, WCRP, WCRP 104, WMO Tech. Doc. 876, 215–218

    • Search Google Scholar
    • Export Citation
  • Schumacher C. & R. A. Houze Jr. 2000 Comparison of radar data from the TRMM satellite and Kwajalein oceanic validation site J. Appl. Meteor. 39 2151 2164

    • Search Google Scholar
    • Export Citation
  • Shin D. & G. R. North 2000 Errors incurred in sampling a cyclostationary field J. Atmos. Oceanic Technol. 17 656 664

  • Shin D., G. R. North & K. P. Bowman 2000 A summary of reflectivity profiles from the first year of TRMM radar data J. Climate 13 4072 4086

  • Shin D., L. S. Chiu & M. Kafatos 2001 Comparison of monthly precipitation derived from the TRMM satellite Geophys. Res. Lett. 28 795 798

  • Short D. A. & G. R. North 1990 The beam filling error in ESMR-5 observations of GATE rainfall J. Geophys. Res. 95 2187 2194

  • Simpson J., C. Kummerow, W-K. Tao & R. F. Adler 1996 On the Tropical Rainfall Measuring Mission (TRMM) Meteor. Atmos. Phys. 60 19 36

  • Wilheit T. T., A. T. C. Chang & L. S. Chiu 1991 Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions J. Atmos. Oceanic Technol. 8 118 136

    • Search Google Scholar
    • Export Citation
  • Xie P. & P. A. Arkin 1995 An intercomparison of gauge observations and satellite estimates of monthly precipitation J. Appl. Meteor. 34 1143 1160

  • Xie P. & P. A. Arkin 1996 Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions J. Climate 9 840 858

    • Search Google Scholar
    • Export Citation
  • View in gallery

    GPCC rain gauge distribution over Africa (Jan 2000). The dots represent grid cells in which at least one rain gauge is present in a 1.0° lat × 1.0° lon box.

  • View in gallery

    Main climate zones of Africa used in this study. Their geographical coordinates are specified in the text.

  • View in gallery

    Seasonal variations of the zonal mean precipitation for the three algorithms (Dec 1997–Nov 2000).

  • View in gallery

    Zonal mean %MBE for the four seasons relative to 3B43 (Dec 1997–Nov 2000).

  • View in gallery

    Scattergrams of mean DJF precipitation of rain gauge vs (left) TRMM PR, (middle) 3B43, and (right) TMPI for the arid, semiarid, savanna, and tropical-wet regions

  • View in gallery

    Same as in Fig. 5, but for JJA.

  • View in gallery

    Comparisons of the mean DJF and JJA precipitation over the southern Atlantic Ocean: (left) 3B43 with TRMM-PR and (right) 3B43 with TMPI.

  • View in gallery

    Comparisons of the overall correlation coefficient obtained between the various algorithms and GPCC rain gauge data and also 3B43 for the different seasons and regions.

  • View in gallery

    Seasonal variations of %MBE for the seven regions in Africa.

  • View in gallery

    Same as in Fig. 9, but for %rmse.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 75 75 16
PDF Downloads 96 96 33

Validation of TRMM Radar Rainfall Data over Major Climatic Regions in Africa

View More View Less
  • 1 Hydrospheric Atmospheric Research Center, Nagoya University, Nagoya, Japan
© Get Permissions
Full access

Abstract

The Global Precipitation Climatology Center global-precipitation-analyses rain gauge data have been used to validate the Tropical Rainfall Measuring Mission (TRMM) precipitation radar (PR) data over the major climatic regions in Africa. The threshold-matched precipitation index and the TRMM-and-other-sources “best-estimate” data (3B43) have also been compared with rain gauge data and with the TRMM PR rainfall product. In the 36-month climatological assessment of the satellite products at a grid spacing of 1.0°, continental Africa has been categorized into five distinct climatic regions: arid, semiarid, savanna, tropical wet, and the South Atlantic Ocean. Zonal mean analysis shows that TRMM PR has a large overestimation in the tropical-rain-forest region of Africa in December–January–February and in March–April–May. The bias is minimized in June–July–August and September–October–November. Bias is generally high for all algorithms in the dry seasons when rainfall is minimal but is less pronounced in the dry seasons of southern African climatic regions. In general, 3B43 has the closest agreement with rain gauge data. The derived standard errors of the algorithms in the climatic regions of Africa are used as a measure of reliability and show that quantitative climatological estimates of TRMM PR are reliable only in the wettest season of the northern savanna and southern semiarid regions when the mean precipitation is greater than 120 mm month–1. The study also shows that TRMM PR has the closest agreement with 3B43 over the South Atlantic Ocean. Over land, all the satellite estimates showed significant seasonally and regionally dependent bias.

Current affiliation: Federal University of Technology, Akure, Nigeria

Corresponding author address: Z. Debo Adeyewa, Dept. of Meteorology, Federal University of Technology, PMB 704, Akure, Nigeria. adeyewa@yahoo.com

Abstract

The Global Precipitation Climatology Center global-precipitation-analyses rain gauge data have been used to validate the Tropical Rainfall Measuring Mission (TRMM) precipitation radar (PR) data over the major climatic regions in Africa. The threshold-matched precipitation index and the TRMM-and-other-sources “best-estimate” data (3B43) have also been compared with rain gauge data and with the TRMM PR rainfall product. In the 36-month climatological assessment of the satellite products at a grid spacing of 1.0°, continental Africa has been categorized into five distinct climatic regions: arid, semiarid, savanna, tropical wet, and the South Atlantic Ocean. Zonal mean analysis shows that TRMM PR has a large overestimation in the tropical-rain-forest region of Africa in December–January–February and in March–April–May. The bias is minimized in June–July–August and September–October–November. Bias is generally high for all algorithms in the dry seasons when rainfall is minimal but is less pronounced in the dry seasons of southern African climatic regions. In general, 3B43 has the closest agreement with rain gauge data. The derived standard errors of the algorithms in the climatic regions of Africa are used as a measure of reliability and show that quantitative climatological estimates of TRMM PR are reliable only in the wettest season of the northern savanna and southern semiarid regions when the mean precipitation is greater than 120 mm month–1. The study also shows that TRMM PR has the closest agreement with 3B43 over the South Atlantic Ocean. Over land, all the satellite estimates showed significant seasonally and regionally dependent bias.

Current affiliation: Federal University of Technology, Akure, Nigeria

Corresponding author address: Z. Debo Adeyewa, Dept. of Meteorology, Federal University of Technology, PMB 704, Akure, Nigeria. adeyewa@yahoo.com

Introduction

For climatological water-resource research studies and for the evaluation of regional model simulations, the reliability of the information on the monthly spatial distribution of precipitation is important. For Africa, rainfall is very crucial for the social and economic welfare of its people. The need for reliable rainfall data for scientific and economic purposes is therefore crucial because, in most parts of Africa, rain-fed agriculture is the method mostly practiced. Measured data from rain gauge networks are still conventionally the most reliable source of area-averaged precipitation for the land surface of the earth. However, rain gauge measurement networks are not as dense or regular as in other major continents. Also, large parts of the African continent are arid and semiarid regions with a sparse rainfall network, implying that satellite observations of rainfall are the best solution for adequate temporal and spatial coverage.

Satellite-based precipitation products could provide very high temporal (3 hourly) and spatial (0.5° latitude × 0.5° longitude grid size) resolution. Nevertheless, they are subject to larger biases and stochastic errors and need to be adjusted to in situ observations (Barrett et al. 1994; Rudolf et al. 1996). Satellites have biases and random errors that are caused by factors such as the sampling frequency, the diurnal cycle of rainfall, the nonuniform field of view of sensors, and the uncertainties in the rain retrieval algorithms (Bell et al. 1990; Kousky 1980; Kummerow 1998; Anagnostou et al. 1999a; Chiu et al. 1990; Chang and Chiu 1999). Huffman (1997) and Huffman et al. (1997) noted that often the random error due mainly to sparsely sampled datasets is the dominant component. Morrissey and Janowiak (1996) have shown that the effect of temporal sampling error in satellite estimates of climate-scale rainfall is to produce a “conditional” bias in which the algorithm overestimates high rainfall and underestimates low rainfall. Validations of satellite-derived rainfall products are, therefore, essential to quantify the direct usability of these products.

In general, rain gauge observations yield relatively accurate point measurements of precipitation but also suffer from sampling error in representing areal means. Also, they are not available over most oceanic and undeveloped land areas (Xie and Arkin 1996). The difference between area rainfall and rain gauge–point rainfall estimates imposes additional noise in radar–rain gauge difference statistics. Noting that this difference could sometimes be interpreted as radar error, Anagnostou et al. (1999a) have shown that the area–point difference in rain gauge rainfall contributes up to 60% of the variance observed in radar–rain gauge differences, depending on the radar grid size, the location of the sampling point in the grid, and the distance from the radar.

Although rain gauge estimates have their own intrinsic errors, Anagnostou et al. (1999b) have shown, using a densely gauged area in northeast Brazil, that the distortion in the rainfall distribution caused by the rain gauge sampling error is small when compared with the distortion caused by the Special Sensor Microwave Imager (SSM/I) rainfall estimation. Using hypothetical ground validation experiments, Ha and North (1999) have estimated that for a field-of-view width of 20 km and an autocorrelation length of a few kilometers, many months of data would be needed for a single-point rain gauge to detect a 10% bias in gauge–satellite differences.

The main objective of the Tropical Rainfall Measuring Mission (TRMM) satellite was to provide a better understanding of precipitation structure and heating in the tropical regions of the earth (Simpson et al. 1996). The TRMM satellite is operating on a non-sun-synchronous orbit that enables it to observe tropical rainfall. It completes one orbit around the earth in about 91 min, allowing for as much coverage of the Tropics and extraction of rainfall data over the 24-h day (16 orbits) as possible. TRMM's onboard instruments include the precipitation radar (PR), Microwave Imager (TMI), Visible and Infrared Scanner (VIRS), Clouds and the Earth's Radiant Energy System, and the Lightning Imaging Sensor. Of these, probably the most prominent is the PR, for obvious reasons. TRMM PR is the first spaceborne radar that was designed to capture a more comprehensive structure of rainfall than any spaceborne sensor before it. It has been producing three-dimensional rainfall data from space in a manner unprecedented by any previous scientific spacecraft.

Short and North (1990) and Shin et al. (2000) noted that one of the advantages of TRMM would be that sampling errors and related biases (e.g., beam-filling errors) would be reduced by its low altitude (350 km) and low inclination (35°). Shin and North (2000) also observed that the non-sun-synchronous orbit of TRMM enables it to sample at all local times over the course of few weeks, thereby reducing diurnal bias drastically. Using different sampling strategies and radar rain rates collected during the Global Atmospheric Research Program Atlantic Tropical Experiment and the Tropical Ocean and Global Atmosphere Coupled Ocean–Atmosphere Response Experiment, Kedem et al. (1990, 1997) estimated the sampling error of the TRMM satellite to be about 10%. A comparison of a 2-yr (1998 and 1999) monthly rainfall data derived from the TMI, PR, TRMM combined algorithm, and TMI emission algorithm were carried out by Shin et al. (2001). The results showed that, for the global and zonal means, the TMI rain rates were the largest and TRMM PR estimates were the lowest. Comparing the radar data from the TRMM satellite and the Kwajalein oceanic validation site, Schumacher and Houze (2000) observed that the temporal sampling of the TRMM radar accurately captures the Kwajelein radar's overall distribution of reflectivities and subdivisions into the different precipitation types. It was, however, noted that the diurnal and the latitudinal variations of the precipitation in the vicinity were not sampled well.

The African continent has diverse climatic regions and provides a suitable platform for validating the TRMM PR precipitation data. Huffman (1997) and Kummerow et al. (2000) have emphasized the need for adequate validations on a regional basis instead of the use of global approximations and the need for having calibration-validation datasets that are representative of various climate regimes around the globe.

The main objective of this paper is, therefore, to validate the TRMM PR data over the major climatic regions in Africa on a climatic scale. The threshold-matched precipitation index (TMPI) and the TRMM and other sources “best-estimate” data (3B43) algorithms have also been included in the validations as a means of comparing the performance of these multisourced best estimates with that of the single-sourced TRMM PR. A total of 36 months of data has been used for the validation.

In section 2, we give brief descriptions of the algorithms used and their sources. Section 3 is devoted to short descriptions of the major climatic regions used for the study in Africa and also to the processing method. Results of the zonal and regional characteristics of the validations are presented in section 4 before drawing some conclusions in section 5.

Data sources and structure

GPCC rain gauge data

Rain gauge analysis have been obtained from the Global Precipitation Climatology Center (GPCC). GPCC is one of the major components of the Global Precipitation Climatology Project (GPCP; Rudolf et al. 1994, 1998; Huffman et al. 1995, 1997). For each month of the year, it derives, on a regular basis, global analyses of monthly precipitation for the land surface of the earth on 1° latitude × 1° longitude grid boxes, based on conventionally measured precipitation data. The GPCC rain gauge distribution over Africa is shown in Fig. 1.

GPCC routinely receives meteorological “SYNOP” and “CLIMAT” reports that contain precipitation data through the numerous World Meteorological Organization Global Telecommunication System (GTS) stations worldwide. To improve the data coverage, the GTS data received at the Deutscher Wetterdienst (German National Meteorological Service) are supplemented by those obtained at the National Oceanic and Atmospheric Administration (United States) and are then merged at GPCC. In addition, GPCC collects additional monthly precipitation data from national meteorological and hydrological services or research projects. The total of the data used for Africa for each month is, therefore, typically about 2600 grid points with about 540 contributing stations. The data compiled from different sources at GPCC are methodically and rigorously checked, analyzed, and merged before being gridded.

We have compared some rain gauge data obtained from some national meteorological establishments in Africa and the African Center of Meteorological Applications for Development with GPCC data, and the results are in good agreement. Results similar to those presented in this study were also obtained when the independent rain gauge datasets were used in preliminary comparisons with the rainfall data derived from the satellite algorithms.

TRMM PR

For this study, the TRMM PR standard monthly product (3A25) at a spacing of 0.5° latitude × 0.5° longitude grid cells has been used. The data correspond to the version-5 algorithm provided by the TRMM science team. The 3A25 level-3 products were derived by the TRMM science team from level-2 instantaneous PR observations. The data cover tropical regions of the world within the 37°N–37°S orbit of TRMM. The original “fine-grid” products were regridded to 1.0° latitude × 1.0° longitude cells to conform to the resolution obtainable from the other satellite products used in this study.

Note that the 13.8-GHz frequency of the TRMM PR means that the beam could be subject to attenuation in heavy-precipitation areas. A first-order correction to such effects has been included in the algorithm (Iguchi et al. 2000). The authors noted, however, that the nonuniform rain distribution within the radar resolution cell might become a large source of error when the attenuation is severe.

TMPI

The threshold-matched precipitation index is part of the GPCP 1° daily (1DD) precipitation product. The GPCP 1DD global precipitation dataset consists of TMPI data between 40°N and 40°S and the Advanced Television and Infrared Observation Satellite Operational Vertical Sounder dataset outside this band (Huffman et al. 1997, 2001). For this study, the TMPI part of the GPCP 1DD product has been used. The TMPI approximate instantaneous precipitation data cover the latitudinal band within 40°N–40°S. The product is a gridded analysis based on satellite estimates of rainfall constrained by a monthly analysis of gauge and satellite observations. The TMPI is a Geostationary Operational Environmental Satellite (GOES) precipitation index–type algorithm, and its values have been calibrated to sum to the “Satellite Gauge” data for the same grid on a monthly basis.

3B43

The TRMM and other satellites/sources (3B43) precipitation estimate (algorithm 3B43) is one of the operational products of TRMM. The gridded estimates are on a calendar-month temporal resolution at a 1° latitude × 1° longitude spacing (40°N–40°S). The combined dataset is based on the concepts of Huffman et al. (1995) on combining precipitation datasets. The TRMM best estimates method is a combination of data from the TMI, PR, and VIRS with SSM/I, IR, and rain gauge data. The 3B43 dataset is a combined observation-only dataset based on gauge measurements and satellite estimates of rainfall. The algorithm was developed by the TRMM science team and the data were processed by the TRMM science data and information system.

Analytical method

Major climatic regions in Africa

For the analyses, continental Africa has been categorized into five distinct main climatic regions: arid, semiarid, savanna, tropical wet, and the South Atlantic Ocean. These regions represent the desert climate (sparse rainfall), steppe climate (low rainfall), hot dry season (moderate rainfall), wet climate (high, all-year rainfall), and oceanic climate, respectively. We have delineated these zones (Fig. 2) according to their observed main climatic features (e.g., Philip 1999).

The arid zone used in the study is the extensive dry region in northern Africa (Sahara Desert, 15°–30°N, 11.5°W–30°E). Based on geography, the northern African countries in the Sahara are mainly Tunisia, Libya, Egypt, Mauritania, Mali, Niger, Chad, and Sudan. Moisture is almost totally absent in some places except where a few oases exist. The climate is uniformly dry, with most areas having averages of less than 130 mm yr–1 of rain, some getting none at all for some years. The temperature range is also extreme, ranging from freezing to more than 54°C in the western and central portions of the desert.

Semiarid regions exist in both northern and southern parts of Africa and, for our study, they are defined as 12°–15°N, 11.5°W–30°E in northern Africa and 17°–22°S, 17°–30°E for southern Africa. In western Africa, this region is also known as the Sahel. The semiarid zone in northern Africa serves as a transition zone between the arid Sahara in the north and the wetter savanna region in the south. The semiarid zone has a steppe climate with low precipitation. Annual rainfall generally averages between 100 and 200 mm and is mostly confined to within June through September. The vegetation is relatively sparse, and grasses and shrubs predominate.

Savanna climate is also found in western Africa (8°–12°N, 11.5°W–30°E) and southern Africa (6.5°–17°S, 14°–40°E). The savanna climate zones occupy about one-fifth of Africa. The climate is characterized by a wet season during the summer months and a dry season during the winter months. Rainfall ranges between 100 and 400 mm yr–1. The region could vary in vegetation type from open-canopied forests with a grassy understory to real savanna regions for which grasses are dominant.

The tropical-rain-forest (tropical wet) climate (6.5°N–6.5°S, 11.5°W–30°E) lies between the northern and southern savanna regions. It is a wet and warm climate with high, all-year rainfall. The average annual rainfall can be as much as 1800 mm, and the climate resembles the equatorial climate. However, although rainfall is more concentrated in one season, no month is rainless. This zone has towering evergreen trees, oil palms, and numerous species of tropical hardwood trees under which one generally finds a dense surface covering of shrubs, ferns, and mosses.

The South Atlantic Ocean region used in this study is just below west Africa and adjacent to southern Africa. It is located at 3.5°N–39.5°S and 19.5°W–5.5°E and represents an oceanic climate.

Processing methods

In general, the data coverage for this study is 3 years (December 1997–November 2000). For this period, data were available for all the sources earlier mentioned except for 3B43 for December of 1997 because its processing began in January of 1998. As noted earlier, the main validation is for the TRMM PR, but similar measures have been applied to the TMPI and 3B43 algorithms to evaluate their performances relative to the single-source TRMM PR. The analyses were conducted on a monthly and seasonal basis, but the results presented here are on the seasonal basis for the sake of brevity. The seasonal analyses are based on 3-month averages: December–January–February (DJF), March–April–May (MAM), June–July–August (JJA), and September–October–November (SON).

For each region, a 3-yr mean bias error (MBE; mm) was computed for each algorithm. The reference is to GPCC rain gauge data, but 3B43 was also used as the reference in some comparative studies. The percentage or relative MBE (%MBE) was also estimated for each algorithm. In this analysis, %MBE is used to ascertain the systematic component of the error in an algorithm. The root-mean-square error (rmse; mm) was evaluated according to the conventional formula and the corresponding relative rmse (%rmse) was also determined. This parameter is also used to appraise the random component of the algorithms. A general evaluation of each algorithm in the different climatic regions relative to rain gauge measurements was conducted using the percentage standard error, or percentage standard deviation, as a yardstick.

For each season, the 3-yr mean precipitation estimates were compared with the mean GPCC rain gauge data for the same period. For an appreciably large number of grid points, the percentage standard error, or percentage standard deviation, reduces to relative rmse or %rmse. The latter has, therefore, been used to evaluate the reliability of each algorithm in the different zones under consideration and their geographical and seasonal variations. Where the rmse of an algorithm estimate is less than 50% of measured rainfall amount, such estimates are considered to be reliable in relative terms. In converse, where the magnitude of an algorithm error is equal to or greater than 50% of the magnitude of the reference estimates, the algorithm is considered to be unreliable for the region and for the particular season. For all of the analysis, the various components of the errors were estimated only when the mean data are available at a particular grid point for all of the algorithms. The number and location of grid points or observations used are therefore the same for all of the algorithms. Thus, the analysis was only for grid boxes in which there were gauges, except for oceanic regions where 3B43 has been used as reference.

Results and discussion

Zonal characteristics

Figure 3 shows the seasonal variations of the zonal mean precipitation for the three algorithms. For this analysis, the number of grid cells with rain gauge data does not have good longitudinal or latitudinal spread, particularly at 20°W–10°E and 40°–60°E and also in the Southern Hemisphere in Africa. The zonal and meridional means are, therefore, far from being representative and could lead to an undesired bias. Mean monthly and seasonal comparisons between the GPCC rain gauge precipitation and that of TRMM PR, TMPI, and 3B43 for the whole African continent showed that, in general, 3B43 has the best correlation with rain gauge data (as will also be seen later for the regional cases). The latter has therefore been used as reference for the following analysis. In all seasons, TMPI shows closer agreement with 3B43 (as expected) at all latitudes except in the southernmost parts of Africa (Fig. 3). In DJF, the departure of TRMM PR from 3B43 and TMPI is very apparent between 5°N and 15°S. A similar pattern of overestimation by TRMM PR but generally less and for a narrower latitudinal band is seen in MAM. Monthly mean zonal analysis for the 3 yr reveals that such features are persistent in the 3 months in DJF and also in March. This tendency is generally less in JJA and SON (least in August and September). Except in JJA, TMPI shows a remarkable departure from 3B43 between 27° and 40°S in all other seasons. This tendency of TMPI to overestimate is strongest in DJF (southern summer), when rainfall is highest for the longitudinal band of 12°–40°E, and also in MAM.

In terms of the zonal mean analysis of %MBE for the four seasons, TRMM PR has a large overestimation between 5°N and 15°S in DJF and MAM in comparison with 3B43 (Fig. 4). It is generally between 30% and 48% between 6°N and 15°S, with some underestimates between 10° and 18°N (Fig. 4). These overestimations occur in the high-rainband regions of Africa. Given that DJF corresponds to a period of rainfall maximum in southern Africa, the near 0% bias between 15° and 25°S is notable. The low bias within this region in DJF gradually changes to underestimation (−25%) by JJA (winter) and to overestimation (25%–50%) in SON. This seasonal pattern is repeated in the reverse order for 25°–37°S. It is interesting to note that in the other regions these seasonal changes from minimum to maximum (positive/negative) bias occur with unique characteristics (e.g., in the wet and dry latitudinal bands). For the dry regions of northern Africa (between 20° and 30°N), the positive bias is highest in MAM (up to 200%). TRMM PR overestimation is generally minimized in JJA and SON. Except for latitudes north of 25°N (arid regions) in JJA and SON (maximum of 140%), the bias is generally within ±25% and ±40%, respectively.

The relative bias for TMPI relative to 3B43 is generally much lower. For northern latitudes, its bias is within ±50% (Fig. 4). In the band of 5°–25°S, it has a significant negative bias in winter (JJA). For all seasons, TMPI has a consistent overestimation of at least 40% between 25° and 40°S.

Regional characteristics

In Figs. 5 and 6, mean seasonal (DJF and JJA) GPCC rain gauge estimates are plotted against the estimates derived from each of the three algorithms for the land regions in arid northern Africa, semiarid (north and south), savanna (north and south), and the tropical-wet regions for grid cells where rain gauge data are available. A 3-yr seasonal mean rainfall for the three algorithms including the rain gauge mean is shown in Table 1. In each region and season, 3B43 closely matches the rain gauge (except in the arid region in DJF) while TRMM PR has the worst agreement. In general, the best agreements are obtained in JJA (Fig. 6). Also, in comparison with other regions, the arid zone has the poorest agreement for all four seasons considered.

For arid northern Africa, the poorest agreements with rain gauge data are recorded in MAM for all the algorithms and the best are recorded in the peak of the wet season (JJA). In this dry zone, 3B43 (Fig. 6, middle panels) has the best correlation in all the seasons, although the overall agreements are just fair (except in JJA). TMPI also has relatively fair correlation as compared with TRMM PR, which has the largest scatter. It is apparent from these figures that in this region, relative to rain gauge estimations, all the algorithms tend particularly to underestimate in JJA and to overestimate in DJF.

For the northern and southern semiarid regions (Figs. 5 and 6), the correlations are generally better than for the drier arid zone. Again, 3B43 has the best agreements with rainfall data, followed by TMPI. The scattergrams for this zone can be understood better in the context of seasonal rainfall variations in the two hemispheres. In DJF, the upper data points correspond to rainfall data for the southern semiarid region where rainfall is maximum during this period, and the lowest points are for the dry season in northern semiarid zone. This is also true for JJA (northern summer; southern winter). Within this context, the best correlations are obtained in DJF and the next best are found in JJA. This result implies that, for all three algorithms, best agreements with rain gauge data are obtained for the southern semiarid zone. For TRMM PR, the random component of the error for the northern semiarid region at rainfall peak (JJA) is relatively larger than for its southern counterpart (DJF). Differences in the random component are therefore understandably more significant for TRMM PR than for the other two algorithms. In the two transitional seasons (MAM and SON, not shown), rainfall is gradually increasing (e.g., in MAM) in the northern zone but decreasing in the southern region. The data points are therefore more homogenous. The largest scatters are obtained in SON in the northern region, and for TRMM PR there is virtually no agreement with rain gauge data during this period.

For the general relationship between the various algorithms and rain gauge data for the northern and southern savanna regions of Africa, the earlier comments regarding the upper and lower data points in DJF, JJA, and the transition seasons are also applicable. In general, the correlations are not as high as in the semiarid regions but are better than those for arid northern Africa. This result is especially true for TRMM PR, which again has very poor correlations while 3B43 has the best. It is observed, however, that 3B43 has a strong tendency of underestimation in DJF for the southern savanna rainfall peak. In JJA, all of the algorithms show strong tendencies toward overestimation for the wet season in the northern savanna. Note also that, unlike in the semiarid regions, the agreement between rain gauge data and the algorithms are better in the northern savanna region than in its southern counterpart.

For the tropical-wet region of Africa, which lies in the equatorial belt between the northern and southern savannas, the scatterplot of the DJF and JJA rainfall intercomparisons shows that 3B43 has better agreements than the others. Note also that, except for 3B43, agreement with the rain gauge data appears to be better in DJF than in JJA in this region. The correlation of TRMM PR with rain gauge data is also generally poor like that of the arid zone. Though not shown, it is interesting to note that, unlike for other zones, the scatterplots for MAM reveal that the lowest rainfall in MAM within this zone is generally above 80 mm. For TRMM PR, however, the upper and lower limits are significantly different. Table 1 shows that, in this region, the precipitation in JJA is only slightly higher than that in MAM. This result is explained by the fact that in some parts of the tropical rain forest of Africa near the Gulf of Guinea, JJA corresponds to a brief period of cessation of rainfall, which is referred to in West Africa as the “little dry season” (Ilesanmi 1971).

Over the South Atlantic Ocean, TRMM PR and TMPI have been compared with 3B43 because rain gauge data are not available for this oceanic region. The close agreement of 3B43 with rain gauge data (much lower bias and random errors) justifies this approach. It is apparent from Fig. 7 (left) that the agreement between TRMM PR and 3B43 is very close. In general, our analyses have shown that TRMM PR has better agreement with 3B43 than with TMPI for all the seasons over this oceanic environment. However, in Fig. 7, TMPI has better agreement with 3B43 (right) except in DJF. The differences between TMPI and 3B43 generally reflect the discrepancies between the underlying algorithms (e.g., Wilheit et al. 1991).

Figure 8 compares the overall correlation coefficient obtained between the various algorithms and GPCC rain gauge data and also 3B43 for the different seasons and regions. In comparison with other regions, the correlations are generally low in the arid region, especially in MAM and SON. The correlation between TRMM PR and rain gauge and between TRMM PR and 3B43 is very low for all the land regions in SON and MAM. One case in particular is the negative correlations (−0.6 for PR–rain gauge and −0.3 for PR–3B43) for the tropical-wet region in SON. The implication of these negative correlations is that the PR algorithm in this season and region is totally unreliable. A cursory look at Fig. 8 reveals that PR–rain gauge has the lowest correlations in the rain-forest region in all seasons of the year. The very low correlations in MAM and SON are explained by the fact that these are transitional rainfall periods for these zones and so errors from undersampling are greatly amplified, in particular for TRMM PR, which is a single-source algorithm. The figure also shows that the TRMM PR–rain gauge pair has very low correlations for all regions. On the other hand, TMPI–3B43 has the highest correlations, as expected. Also (as deduced from Figs. 5 and 6), of all of the algorithms under consideration, 3B43 has the best agreement with rain gauge data in all four seasons.

The correlation between 3B43 and TRMM PR is generally high (≥0.8) and consistent for all four seasons of the year for the South Atlantic Ocean. This is especially remarkable in MAM and SON, when the correlations between TRMM PR and rain gauge and 3B43 are very low (<0.4) for all of the land regions. The figure also shows that both TRMM PR and TMPI generally have better agreement with 3B43 than with rain gauge data for all of the land regions. For TMPI, the improvement varies between 1% in DJF (semiarid) to almost 200% in MAM (arid). On the other hand, the relative improvement is more remarkable for TRMM PR, especially in MAM and SON in the arid and tropical-wet regions. However, its correlation with both rain gauge and 3B43 is still very low in these land regions. It therefore can be concluded that TRMM PR has a better performance over the oceanic environment.

Regional bias and rmse characteristics

The seasonal variations of %MBE for the different land and oceanic regions relative to rain gauge data are depicted in Fig. 9. In a similar way, the %rmse for each region and season is shown in Fig. 10. In general, the TRMM PR has a higher bias and rmse than the other multisourced algorithms; 3B43 has the lowest bias and rmse for each region and season. In comparing DJF and JJA, one can conclude that both random and systematic errors are higher in the northern regions for the winter conditions (DJF) than in the southern regions for a similar season (JJA). Note, however, that the reliability of the analysis in some regions (e.g., tropical rain forest) may be affected by paucity of data in the region.

Arid region

For arid northern Africa in JJA and SON, the negative bias of TMPI exceeds those of TRMM PR and 3B43. TRMM PR has a higher bias in DJF and MAM when the mean precipitation per month (Table 1) is extremely low (<2 mm). This result accounts for the reason the %MBE of TRMM PR is exceptionally high during this period, although, as mentioned earlier, the undersampling of TRMM PR could also be implicated.

In the arid zone, the rmse for all algorithms is largest in JJA and least in DJF and MAM. In each season, TRMM PR has the highest rmse, varying from 4 mm in DJF to 22 mm in JJA, which translates to a relative rmse of more than 200% in DJF and 116% in JJA. In MAM, the rmse of 7 mm becomes more than 400% for the small mean rainfall during this period. Because the %rmse is greater than 50% in each season for TRMM PR, it means that climatologically its rainfall estimates cannot be safely used in a qualitative sense in the arid northern Africa region for all seasons of the year. The rmse of TMPI in this region is somewhat lower but is more than 50% in all four seasons. In a similar way, 3B43 has the lowest rmse except in DJF, and the inherent relative error is more than 50% of the reference estimates in all seasons.

Semiarid regions

For the mean seasonal variation of precipitation for the northern and southern semiarid regions shown in Table 1, it is interesting to note that in both regions the mean rainfall estimates by rain gauge and 3B43 for the peak of rainfall are about the same. However, the estimates of TRMM PR and TMPI are considerably different. It is also apparent that all of the algorithms overestimate during the period of rainfall peak. Bias analysis corroborates this fact, with TRMM PR having the highest bias for the northern (33 mm) and southern (25 mm) semiarid regions. For the insignificant rainfall in the northern semiarid zone in DJF, the bias is correspondingly small but translates to extremely high %MBE in this season, especially for TMPI and 3B43 (Fig. 9). The rmse is also very small (<1 mm) for all algorithms. The %MBE and %rmse can therefore be said to be exaggerative in the arid zones because of the amplification effect of the small mean rainfall (divisor). On the other hand, the high bias in JJA for TRMM PR in this region becomes only 27% MBE because of the high rainfall in the season. The rmse is also high (78 mm) but approximates to about 64% because of the high precipitation. The %MBE of 3B43 is very small (3%) in JJA and the %rmse is also very small (13%). In all seasons in the northern semiarid region, TMPI occupies the intermediate position while TRMM PR has the largest bias and rmse. The %rmse in 3B43 and TMPI is generally below 50% in all seasons, except in DJF, and the rainfall estimates of TRMM PR are not reliable in any season in the zone.

In the southern semiarid region, similar results are obtained (Figs. 9 and 10) but with generally much lower %MBE and %rmse in the comparative seasons. However, one observes the negative bias (about 7%) of TMPI and 3B43 in SON. In practical terms, the TRMM PR estimate is valid only in the wettest season whereas the estimates of the other two algorithms are climatologically valid in all seasons of the year except for the driest season (JJA). In comparison with the northern semiarid region, the overall performance of all of the algorithms is better here.

Savanna regions

As expected, Table 1 indicates that in comparison with the previously considered semiarid regions, rainfall is appreciably higher in the northern and southern savanna regions in all seasons of the year. It also shows that TRMM PR has a comparatively high overestimation in the southern savanna summer. Comparisons of the bias by TRMM PR in the wet seasons of the northern (JJA) and southern (DJF) savanna regions shows that TRMM PR has a larger bias in the southern part (67 mm, 51%) than in the northern zone (26 mm, 15%). The random error is also higher in the south, with an rmse of 98 mm (74%) as compared with 88 mm (50%) in the north (Figs. 9 and 10). In the dry season, all algorithms underestimate in the southern savanna, with TMPI having the largest (−2 mm, −32%). Also, only TMPI underestimates in the northern savanna in DJF. In terms of %rmse, the situation is reversed in the dry season, with all algorithms having more than 130% in the northern zone. A cursory look at Fig. 10 shows that the estimates of TMPI and 3B43 are relatively reliable in the three wetter seasons in the southern savanna (except TMPI in SON) and similar seasons in the northern savanna. TRMM PR has less than 50% relative rmse in only one season in the north and in none in the south. Hence, unlike the results obtained for the semiarid region, the performance of TRMM PR is better in the northern savanna than in the southern part.

Tropical-wet regions

The overestimation of TRMM PR is apparent in the tropical-wet zone for all of the seasons except JJA (Table 1). Figure 9 gives further details of the large bias in DJF (57 mm; 79%), which also appeared in the mean zonal analysis. Though somewhat lower, the overestimation of TRMM PR in this wet zone is also conspicuous in MAM (41 mm; 35%), but it is almost nonexistent in JJA.

The rmse for TRMM PR is also high in DJF for which in relative terms it is over 140%, making the TRMM PR estimate in the season unreliable in this zone. In the other seasons, the random component of TRMM PR error reduces to about 67%. From Fig. 9, it can be seen that TMPI generally overestimates in this region but underestimates in JJA (−5 mm; −4%). Its rmse is also relatively high in JJA (50 mm; 37%) when compared with that of 3B43 (33 mm; 25%).

South Atlantic Ocean

As shown in the seasonal variations of the mean precipitation for the three algorithms for the South Atlantic Ocean (Table 1), the highest rainfall period is in MAM and the lowest is in SON. Comparison with other zones shows that in general the rainfall in the oceanic region is lower than in the tropical-wet region but that it receives more than 10 mm month–1 in any particular season. The mean seasonal bias of TMPI and TRMM PR (both relative to 3B43) is also depicted in Fig. 9. Unlike other regions, the bias by TRMM PR is generally low (<10 mm or 20%). Note that TMPI has a higher bias than TRMM PR in JJA and SON. Also, as compared with the relative overestimation of 15% by TRMM PR in SON, TMPI overestimated by 64% and also has the highest relative rmse (112%) during this period. The rmse estimates indicate that the estimation of both TMPI and TRMM PR is not reliable in all seasons, except in MAM for TMPI. In general, the performance of the two algorithms is much better for the oceanic region than for the other zones previously considered. This result is in agreement with Xie and Arkin (1995) who compared three sets of gauge observations and eight different satellite estimates. Results of their examination of several sources of climatic-scale precipitation data showed that over the oceans all satellite estimates showed a high correlation and stable bias when compared with atoll-based gauges. On the other hand, they exhibited high pattern correlation over land but with large seasonally and regionally dependent bias.

Regional summary

Using the relative standard error of each algorithm as an index of reliability, a regional summary of the reliability of the precipitation data of the three algorithms is given in Table 2. The valid seasons refer to 3-month periods for which the relative standard error of each algorithm is less than 50% in each climatic region. The observed minimum seasonal mean (“safe” threshold) rainfall is also indicated. The threshold rainfall is the observed minimum measurable amount of rainfall by the rain gauge for which the algorithm can be safely considered reliable. For TRMM PR, the table shows that its precipitation data are not reliable in most seasons in all delineated regions of Africa. This is due to the fact that the inherent error in its estimation generally exceeds 50% of the actual or reference rainfall amount. The table shows that the rainfall product of TRMM PR is only reliable in the wettest season of the northern savanna and southern semiarid regions (JJA and DJF, respectively). Note also that the PR also fails to meet the requirement over the Atlantic Ocean. The lowest threshold of TRMM PR (121 mm month–1) is over the southern semiarid region. When a less stringent measure is applied (relative standard error of each algorithm less than 100% in each climatic region; not shown), it is observed that the TRMM PR precipitation product is still not reliable in any season in the northern Africa arid region because the inherent error in its estimation generally exceeds the actual or reference rainfall amount. In the northern semiarid region, only two seasons (JJA and SON) are valid. However, its reliability increases in the wetter regions and seasons of the tropical-wet and northern and southern savanna regions. The lowest threshold of TRMM PR (21 mm month–1) for the less stringent case is observed over the South Atlantic Ocean.

It is observed that, although the TMPI and 3B43 algorithms both include rain gauge data, neither of them qualifies as being reliable in the present context in arid northern Africa. Except in this region, the precipitation products of the TMPI and 3B43 are reliable in the wet seasons and transition seasons for the climatic regions of Africa. The threshold rainfall is also very low. In the southern savanna region, however, only 3B43 is reliable in the transition from the dry to wet season (SON).

Further analysis of the relative error (not shown) indicated that, for each algorithm, higher rainfall amounts than those shown in Table 2 would provide better reliability.

The larger biases of the TRMM PR in most of the climatic regions may be attributed to the effect of temporal sampling error. They may also be due to the fact that the TRMM PR algorithm cannot adequately account for inherent errors resulting from the contamination of the PR reflectivities by elevated ground clutter.

Summary and conclusions

At a 1.0° grid spacing, we have conducted a 36-month climatological assessment of the TRMM PR, 3B43, and GPCP TMPI satellite products over the major climatic regions in Africa with characteristically different rainfall regimes. Zonal mean analysis shows that TRMM PR has a large overestimation in the tropical-rain-forest region of Africa in DJF and in MAM, but it is minimal in JJA and SON. Bias is generally high for all algorithms in the dry seasons when rainfall is minimal, but it is less pronounced in the dry seasons of southern African climatic regions. In general, the study shows that 3B43 closely matches rain gauge data, suggesting that the goal of the algorithm was largely achieved.

For all algorithms, the random and systematic components of the errors show significant sensitivity to seasonal and regional differences. This sensitivity is also noticed in similar climatic zones that are located in geographically different regions of the continent. For instance, the best agreements with rain gauge data are obtained for the southern semiarid zone rather than for its northern counterpart for all three algorithms. For TRMM PR, the random component of the error for the northern semiarid region at rainfall peak (JJA) is relatively larger than for its southern counterpart (DJF). In this zone, differences in the random component are more significant for TRMM PR than for the other two algorithms. On the other hand, the agreement between rain gauge data and the algorithms is better in the northern savanna region than in its southern counterpart.

This study shows that TRMM PR precipitation data at 1.0° (or lower) resolution are only reliable in the wettest seasons of the northern savanna (JJA) and southern semiarid (DJF) regions when the minimum seasonal mean rainfall is about 170 and 120 mm, respectively. The performance of TRMM PR is generally better in the southern semiarid region. Its bias is lowest over the South Atlantic Ocean. The 3B43 product has better reliability than TMPI. It also has the lowest threshold in each region. Apart from the fact that both 3B43 and TMPI include other satellite information and rain gauge data, we note that the apparent superiority of the two products to the TRMM PR may be due mainly to disparities in sampling. Also, the observed large biases of TRMM PR may be due to inadequacies of the algorithm to account for contamination of the PR reflectivities by elevated ground clutter.

Given that the TRMM PR is the only single-source rainfall algorithm in this analysis and that the 3B43 and TMPI precipitation estimates are both based on the combination of multiple satellite datasets and rain gauge analyses, the results of these analyses should be understood from the viewpoint of possible climatological uses of TRMM PR data over the African region.

One major conclusion that can be drawn from this study is that the TRMM PR data do not appear to be adequate in serving as a stand-alone product. However, the results should be very useful and reasonable when combined with other satellite and rain gauge data as is done in the 3B43 algorithm. The TRMM PR data are still unique in terms of their three-dimensional structure, and this feature is perhaps most valued on the African continent because in most cases only rain gauge data are available.

Acknowledgments

The support of Nagoya University, Japan, and its Hydrospheric Atmospheric Research Center to one of the authors (ZDA) through a visiting research fellowship is gratefully acknowledged. TRMM data were provided by the TRMM Science Data and Information System (TSDIS) at the NASA Goddard Space Flight Center Distributed Active Archive Center (DAAC). Precipitation analysis data based on rain gauge observations were obtained from the GPCC operated by the Deutscher Wetterdienst (DWD, National Meteorological Services), Germany. The GPCP TMPI combined precipitation data were provided by the NASA Goddard Space Flight Center's Laboratory for Atmospheres, which develops and computes the dataset as a contribution to the GEWEX Global Precipitation Climatology Project.

REFERENCES

  • Anagnostou E. N., W. F. Krajewski & J. A. Smith 1999a Uncertainty quantification of mean-areal radar-rainfall estimates J. Atmos. Oceanic Technol. 16 206 215

    • Search Google Scholar
    • Export Citation
  • Anagnostou E. N., A. J. Negri & R. F. Adler 1999b Statistical adjustment of satellite microwave monthly rainfall estimates over Amazonia J. Appl. Meteor. 38 1590 1598

    • Search Google Scholar
    • Export Citation
  • Barrett E. C., J. Doodge, M. Goodman, J. Janowiak, E. Smith & C. Kidd 1994 The First WetNet Precipitation Intercomparison Project (PIP-1) Remote Sens. Rev. 11 49 60

    • Search Google Scholar
    • Export Citation
  • Bell T. L., A. Abdullah, R. L. Martin & G. R. North 1990 Sampling errors for satellite-derived tropical rainfall: Monte Carlo study using a space–time stochastic model J. Geophys. Res. 95 2195 2206

    • Search Google Scholar
    • Export Citation
  • Chang A. T. C. & L. S. Chiu 1999 Nonsystematic errors of monthly oceanic rainfall derived from SSM/I Mon. Wea. Rev. 127 1630 1638

  • Chiu L. S., D. Short, A. McConnel & G. North 1990 Rain estimation from satellites: Effect of finite field of view J. Geophys. Res. 95 2177 2185

  • Ha E. & G. R. North 1999 Error analysis for some ground validation designs for satellite observations of precipitation J. Atmos. Oceanic Technol. 16 1949 1957

    • Search Google Scholar
    • Export Citation
  • Huffman G. J. 1997 Estimates of root-mean-square random error for finite samples of estimated precipitation J. Appl. Meteor. 36 1191 1201

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

    • Search Google Scholar
    • Export Citation
  • Huffman G. J. Coauthors, 1997 The Global Precipitation Climatology Project (GPCP) combined precipitation dataset Bull. Amer. Meteor. Soc. 78 5 20

    • Search Google Scholar
    • Export Citation
  • Huffman G. J., M. M. Morrissey, S. Curtis, R. Joyce, B. McGavock & J. Susskind 2001 Global precipitation at one-degree daily resolution from multisatellite observations J. Hydrometeor. 2 36 50

    • Search Google Scholar
    • Export Citation
  • Iguchi T., T. Kozu, R. Meneghini, J. Awaka & K. Okamoto 2000 Rain-profiling algorithm for the TRMM precipitation radar J. Appl. Meteor. 39 2038 2052

  • Ilesanmi O. O. 1971 An empirical formulation of an ITD rainfall model for the Tropics: A case study of Nigeria J. Appl. Meteor. 10 882 891

  • Kedem B., L. S. Chiu & G. N. North 1990 Estimation of mean rain rate: Application to satellite observations J. Geophys. Res. 95 1965 1972

  • Kedem B., R. Pfeiffer & D. A. Short 1997 Variability of space–time mean rain rate J. Appl. Meteor. 36 443 451

  • Kousky V. E. 1980 Diurnal rainfall variation in northeast Brazil Mon. Wea. Rev. 108 488 498

  • Kummerow C. 1998 Beamfilling errors in passive microwave rainfall retrievals J. Appl. Meteor. 37 356 357

  • Kummerow C. Coauthors, 2000 The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit J. Appl. Meteor. 39 1965 1982

  • Morrissey M. L. & J. E. Janowiak 1996 Sampling-induced conditional biases in satellite climate-scale rainfall estimates J. Appl. Meteor. 35 541 548

    • Search Google Scholar
    • Export Citation
  • Philip G. 1999 New School Atlas and Gazetteer George Philip, 176 pp

  • Rudolf B., H. Hauschild, W. Rueth & U. Schneider 1994 Terrestrial precipitation analysis: Operational method and required density of point measurements Global Precipitations and Climate Change, M. Desbois and F. Desalmand, Eds., NATO ASI Series I, Vol. 26, Springer-Verlag, 173–186

    • Search Google Scholar
    • Export Citation
  • Rudolf B., H. Hauschild, W. Rueth & U. Schneider 1996 Comparison of raingauge analyses, satellite-based precipitation estimates and forecast model results Adv. Space Res. 18 7 53 62

    • Search Google Scholar
    • Export Citation
  • Rudolf B., T. Fuchs, W. Rueth & U. Schneider 1998 Precipitation data for verification of NWP model re-analyses: The accuracy of observational results Proc. Int. Conf. on Reanalyses, Washington, DC, WCRP, WCRP 104, WMO Tech. Doc. 876, 215–218

    • Search Google Scholar
    • Export Citation
  • Schumacher C. & R. A. Houze Jr. 2000 Comparison of radar data from the TRMM satellite and Kwajalein oceanic validation site J. Appl. Meteor. 39 2151 2164

    • Search Google Scholar
    • Export Citation
  • Shin D. & G. R. North 2000 Errors incurred in sampling a cyclostationary field J. Atmos. Oceanic Technol. 17 656 664

  • Shin D., G. R. North & K. P. Bowman 2000 A summary of reflectivity profiles from the first year of TRMM radar data J. Climate 13 4072 4086

  • Shin D., L. S. Chiu & M. Kafatos 2001 Comparison of monthly precipitation derived from the TRMM satellite Geophys. Res. Lett. 28 795 798

  • Short D. A. & G. R. North 1990 The beam filling error in ESMR-5 observations of GATE rainfall J. Geophys. Res. 95 2187 2194

  • Simpson J., C. Kummerow, W-K. Tao & R. F. Adler 1996 On the Tropical Rainfall Measuring Mission (TRMM) Meteor. Atmos. Phys. 60 19 36

  • Wilheit T. T., A. T. C. Chang & L. S. Chiu 1991 Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions J. Atmos. Oceanic Technol. 8 118 136

    • Search Google Scholar
    • Export Citation
  • Xie P. & P. A. Arkin 1995 An intercomparison of gauge observations and satellite estimates of monthly precipitation J. Appl. Meteor. 34 1143 1160

  • Xie P. & P. A. Arkin 1996 Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions J. Climate 9 840 858

    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

GPCC rain gauge distribution over Africa (Jan 2000). The dots represent grid cells in which at least one rain gauge is present in a 1.0° lat × 1.0° lon box.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 2.
Fig. 2.

Main climate zones of Africa used in this study. Their geographical coordinates are specified in the text.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 3.
Fig. 3.

Seasonal variations of the zonal mean precipitation for the three algorithms (Dec 1997–Nov 2000).

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 4.
Fig. 4.

Zonal mean %MBE for the four seasons relative to 3B43 (Dec 1997–Nov 2000).

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 5.
Fig. 5.

Scattergrams of mean DJF precipitation of rain gauge vs (left) TRMM PR, (middle) 3B43, and (right) TMPI for the arid, semiarid, savanna, and tropical-wet regions

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 6.
Fig. 6.

Same as in Fig. 5, but for JJA.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 7.
Fig. 7.

Comparisons of the mean DJF and JJA precipitation over the southern Atlantic Ocean: (left) 3B43 with TRMM-PR and (right) 3B43 with TMPI.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 8.
Fig. 8.

Comparisons of the overall correlation coefficient obtained between the various algorithms and GPCC rain gauge data and also 3B43 for the different seasons and regions.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 9.
Fig. 9.

Seasonal variations of %MBE for the seven regions in Africa.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Fig. 10.
Fig. 10.

Same as in Fig. 9, but for %rmse.

Citation: Journal of Applied Meteorology 42, 2; 10.1175/1520-0450(2003)042<0331:VOTRRD>2.0.CO;2

Table 1.

The 3-yr mean seasonal rainfall (mm month–1) for the different algorithms and regions of Africa

Table 1.
Table 2.

The valid seasons () for which the standard error of the PR, TMPI, and 3B43 is less than 50% in each climatic region. The minimum seasonal mean (safe threshold) rainfall is also indicated

Table 2.
Save