Introduction
The Global Precipitation Climatology Project (GPCP) was developed and implemented by the World Climate Research Program (WCRP 1986) to produce monthly rainfall estimates on global 2.5° lat × 2.5° long grids from January 1986 to 2000. The final monthly estimates are a blend of satellite and gauge data, and the version 1 record begins in July 1987, the start of the data collection from microwave sensors. Also, the individual components, that is, satellite rain estimates based on infrared and microwave observations and gauge data (also analyzed to a 2.5° × 2.5° grid), are available for study (Huffman et al. 1997).
A comparison of satellite estimates of rain with analyzed rain gauge data revealed an extraordinarily large discrepancy between the two fields in the rainy area of equatorial Africa. The satellite estimates are nearly twice as large as the gauge estimates. Another interesting feature is that this is the only tropical rainy area that exhibits such a large discrepancy between the gauge and satellite analyses. For example, tropical South America also exhibits large monthly mean rainfall amounts but the satellite and gauge analyses are comparable in magnitude. This is shown in Fig. 1, which displays results of the GPCP multisatellite (MS) rain (composed of microwave and IR estimates), the gauge-only analysis of the GPCP’s Global Precipitation Climatology Center (GPCC), and the difference between the two sets of estimates.
The daily mean MS rainfall (mm) from 1988–94 is displayed in the upper panel of Fig. 1 and the mean daily GPCC rainfall is shown in the middle panel. This 7-yr period was chosen from the full GPCP dataset to correspond to the other datasets used in this study that do not span the entire GPCP estimation period. The lower panel, which shows the difference between the estimates from both algorithms, illustrates the overestimation problem. The large discrepancy (greater than 4 mm day−1; an overestimation by a factor of 2), in equatorial central Africa is of concern because satellite estimates of rain are presumed to be most reliable in the Tropics where the bulk of rainfall is from deep convective clouds, which are readily detected by infrared and microwave scattering techniques.
The overestimation is present in each month as illustrated by time series of MS and GPCC rainfall (Fig. 2, upper panel) for the area average of the region of focus in Africa, from 10°S to 10°N and 20° to 30°E. For comparison, the time series from a region of South America of similar size and with similar magnitude of satellite-estimated rainfall (from 10°S to 10°N and 75° to 65°W) is displayed in the lower panel. It appears that there is no overestimation problem in this region of South America, because the GPCC estimates have similar magnitudes to the MS estimates with the exception of three peaks in the MS estimates. This region will be used for future comparisons to help provide an understanding of the overestimates observed in Africa. The GPCP Satellite–Gauge (SG) estimate, shown as a solid line in Fig. 2, is in better agreement with the gauge analysis, which is not surprising since the large-scale gauge analysis is utilized to adjust for possible biases in the satellite estimates (Huffman et al. 1997).
It should be noted that both the infrared estimates of the GOES Precipitation Index (GPI) and the microwave estimates from the Special Sensor Microwave/Imager (SSM/I) exhibit biases that resemble each other and the MS bias (Fig. 3). The similarity between SSM/I and MS biases is not unexpected, since the MS estimate is created by adjusting the GPI estimates to produce similar magnitudes as space–time-matched SSM/I rainfall (Huffman et al. 1995). The high rainfall estimates by both the GPI and SSM/I algorithms suggest that there is significant cold cloud (causing high GPI estimates) and ice scattering (causing high SSM/I estimates) in this region. Over land, high ice scattering implies deep convection associated with thunderstorm activity. This is supported by global maps of satellite-detected lightning observations (Christian et al. 1996). Interestingly, this region of equatorial central Africa had a higher lightning flash density (greater than 30 flashes km−2 yr−1) than any other region of the globe. For comparison, equatorial South America had between 15 and 30 flashes km−2 yr−1 for the same time period. This along with apparent higher biases in satellite estimates of rain over Africa in comparison with South America suggests that thunderstorms over equatorial Africa yield less rain at the surface than expected for a convectively active tropical region.
It is important to understand the uncertainties of the GPCP rainfall products so that the users of these products can account for their limitations in accuracy. Thus, the possible causes of this overestimation problem need to be investigated so that these errors might be better predicted. In addition, important insights into satellite rainfall estimation may be gained, because, if it is the satellite estimates that are incorrect, the physical processes causing the overestimation may be important to consider in all satellite estimation algorithms.
The most straightforward explanation is an inadequate rain gauge network (few or no gauges), which can produce inaccurate estimates of rain in the gauge analysis. Since the satellite estimates are adjusted by the gauge analysis, this implies that the satellite estimates may be correct and that the blended gauge and satellite product would be incorrect. The other explanations may be due to different rain processes for different regions, indicating that a single calibration may not be adequate for use everywhere. In this study, the physical properties of air masses (aerosol and moisture content) will be considered for their effect on the rain processes. The airmass properties over central Africa may be influenced by the arid Saharan desert and the semiarid Sahel areas north of the rainfall maximum and the associated circulation, which is dominated by high pressure cells over the Sahara and South Africa and a monsoon circulation that brings moist air from the Atlantic Ocean (Fontan et al. 1992).
Because the satellite estimates of rain over tropical rainy South America do not exhibit the same type of bias, a comparative analysis between the two areas was performed to provide greater insight into the processes accounting for the biases observed over Africa. A brief review of the data sources and how they are blended to arrive at the final GPCP product will be presented below.
Brief description of data sources
The two classes of satellite algorithms used to produce GPCP rainfall products are infrared and microwave algorithms. Infrared algorithms are based on the relationship between rainfall rate and cloud-top temperature, as cold temperatures are usually associated with deep convective clouds, which account for much of the rainfall in the Tropics. The GPCP infrared algorithm, the GPI (Arkin and Meisner 1987), is based primarily on infrared data from geostationary satellites, which have the advantage of 3-h sampling. Data from the NOAA Advanced Very High Resolution Radiometer (AVHRR) are used if necessary to fill the gaps of missing geostationary satellite data.
The second class of satellite algorithms used by the GPCP is based on the data of the SSM/I on board Defense Meteorological Satellite Program (DMSP) satellites. The GPCP uses both an emission algorithm (Wilheit et al. 1991) and a scattering algorithm (Ferraro 1997). The emission algorithm is used over ocean only, because it is based on the microwave emission of microwave radiant energy by raindrops, which can be detected over the low-emissivity ocean surface. The scattering algorithm is based on the depletion of the upwelling radiation due to scattering by large ice particles. This algorithm is applicable over land and ocean, but the GPCP applies it only over land where the emission scheme is not useful because of the lack of emissivity contrast between the surface and rain clouds. The microwave algorithms have a disadvantage of less frequent temporal sampling than the infrared algorithm because the SSM/I data come from the polar-orbiting DMSP satellites, which observe a point on the earth at most two times per day. An important point to note is that the GPI and microwave estimates each use their own “calibration,” which has no regional dependence.
The rain gauge data used in the final GPCP product, the merged SG analysis product, are processed by the GPCC of the Deutscher Wetterdienst (the German Weather Service), Offenbach, Germany, using a system described by Rudolf (1993). The gauge analysis is produced by interpolating the data to a 0.5° mesh and averaging these points to a 2.5° grid (Rudolf et al. 1994). The gauge analysis enters into the blended GPCP product in two ways; the large-scale average of the gauge analysis is used to adjust the MS estimate at the same scale for possible biases, and then the gauges and the bias-adjusted MS estimates are merged using inverse error variance weighting. Details are presented in Huffman et al. (1997). Clearly the gauge analysis and the associated errors have a large influence on the final value of the merged analysis.
Possible causes of African rainfall overestimation
Number of rain gauges
Errors in the rain gauge estimates were mentioned previously as a possible explanation for the discrepancies. Rain gauges measure rainfall fairly accurately at an isolated point in space (Groisman and Legates 1994), but area-averaging of rain gauge data introduces spatial sampling errors. The spatial sampling error depends primarily on the number and distribution of gauges within the 2.5° lat × 2.5° long grid box and regional rainfall variability (Rudolf et al. 1994; Morrissey et al. 1995; Huffman et al. 1995; Huffman 1997).
The mean number of GPCC gauges in each 2.5° lat × 2.5° long grid box over Africa is illustrated in the upper panel of Fig. 4. The mean is calculated because the number of gauges used in the analysis can change from month to month. The spatial sampling error will be high over Africa because gauges are sparse there (Rudolf et al. 1994; Morrissey et al. 1995; Huffman 1997). Because of this we cannot be confident that the gauge analysis is more accurate than the satellite estimates, and there are no other coincident independent data that can be used to establish which is more correct. However, we are able to examine the long-term average rainfall over Africa using a subset (1950–93) of Nicholson’s (1993) African rainfall archive. A climatology based on these data should be more accurate than the limited period used by the GPCP because it contains data from more gauges (Fig. 4b) and is based on a period greater than 30 yr. This climatology (Fig. 5b) shows great similarity to the GPCP gauge climatology (Fig. 5a) and lends credence to the notion that the discrepancy between satellite and rain gauge estimates is probably due to explanations other than the scarcity of rain gauges in central Africa.
Aerosols and drop size distribution
A second possible explanation for satellite versus gauge biases is regional differences in the distribution of aerosols. Precipitation processes are very sensitive to the cloud drop size distribution (DSD). This is due to the physics of accretion and coalescence processes (Kessler 1967). If the liquid water droplets in a cloud are small, it is difficult for raindrops to grow to sufficient size to achieve a velocity high enough to overcome the updrafts, fall out of the cloud, and reach the ground without evaporating. DSD depends primarily on the concentration of CCN in the atmosphere. Large numbers of CCN result in the formation of many small droplets. This, in turn, will further limit the growth of the droplets to larger size. Also, coalescence is less efficient within clouds having large concentrations of small droplets (Cotton and Anthes 1989). It also has been shown (e.g., Johnson 1980) that the temperature of the cloud base influences the activation of CCN, because cold cloud bases (i.e., with lifted condensation levels at higher elevations due to a drier atmosphere) have higher peak supersaturation levels than warm cloud bases, resulting in more activated CCN.
CCN also affect ice formation in the cloud. In convective clouds, the ice particles form in the mixed phase zone. The depth of this zone depends on the liquid water content there. Convective clouds with a large concentration of small droplets (the time required of droplets to freeze is longer for small droplets) and/or with strong updrafts at the 0° isotherm, will likely produce a deep mixed phase zone. Water vapor pressure at the surface of a supercooled droplet is larger than water vapor pressure at the surface of an ice particle at a given temperature. Thus, an ice particle placed among a population of supercooled droplets will grow at their expense. The droplets will evaporate to supply water vapor that will feed the growth of the ice particle. This mechanism is known as the Findeisen–Bergeron process of ice particle growth. Therefore, a mixed-phase zone with many supercooled droplets will enable ice particles to grow to larger size.
Strong updrafts are conducive to both a deep mixed-phase zone and lightning formation. Ba et al. (1998) have shown that clouds with deep mixed-phase zones produce more lightning than clouds with shallow mixed-phase zones. Also, Ba et al. (1998) show positive correlation between lightning and the amount of passive microwave scattering. These findings are consistent with the satellite observations over central Africa; the region with maximum intense lightning activity corresponds to the region of large SSM/I scattering (i.e., high SSM/I rainfall estimates). In comparison with its Amazon counterpart, it is hypothesized that clouds forming in central Africa develop mainly within air masses richer in CCN than the Amazonian air masses.
CCN can result from human activities such as biomass burning and by natural actions such as wind-blown dust. The Sahara Desert and the semiarid Sahelian region represent a large potential source region for aerosols and CCN. The circulation in the area is complex and dominated by a large seasonal shift in the position of the intertropical convergence zone (ITCZ), which is determined by the northeast and southeast trade winds, and the monsoon circulation from the Atlantic, which provides southwesterly circulation (Fontan et al. 1992). The NE trade winds can bring aerosol-laden air from North Africa into the ITCZ providing many cloud condensation nuclei, resulting in relatively small drops and inefficient rain processes. Désalmand et al. (1982) noted that Saharan dust is a significant source of CCN in West Africa, as is decomposing vegetation in equatorial forests of Africa. It is likely that these sources supply CCN to central Africa as well. In addition, Andreae et al. (1992) found an enriched layer of aerosols over equatorial Africa that they attributed to biomass burning in sub-Saharan Africa. While these sources of CCN reside primarily at lower altitudes, it appears that strong convective activity pumps the aerosols up into the high clouds (Kaufman and Nakajima 1993) so that drop size is also affected in convective storms that may result in high rainfall estimates from satellite algorithms. As an example of the effect of aerosols on the cloud drop size, Kaufman and Nakajima (1993) found that the presence of dense smoke can reduce the remotely sensed drop size of continental clouds from 15 to 9 μm for a study area in tropical South America.
To compare the possible effect of aerosols for both Africa and Amazonia, the distribution of the effective radii of cloud droplets at the cloud top was estimated for a test period of September 1989 over Africa and South America from the 3.7-μm channel of AVHRR using a lookup table computed by Rosenfeld and Gutman (1994). The lookup table was obtained through a radiative transfer model (Nakajima and King 1990). The basic principle of this model is that cloud reflectance in spectral regions (i.e., 3.7-μm wavelengths), where liquid water and ice are absorbing, is strongly dependent upon hydrometeor size.
The results, shown in Fig. 6, indicate that the modal value of effective cloud droplet radii (the larger radii from ice particles are not shown) over Africa is about 13–14 μm, in comparison with about 23–24 μm over South America. Although this was for a limited sample, dictated by the availability of the AVHRR data, these results support the idea that aerosols may be playing an important role in the rain processes over Africa. The high density of lightning flashes observed over Africa and the large amount of scattering observed by the SSM/I may also indicate the presence of many small droplets in convective clouds there, since small droplets are likely to be carried aloft to the freezing level, resulting in increased ice formation.
Moisture environment
The moisture availability in a region affects the rainfall processes with relation to the satellite estimation problem. Convective clouds forming under dry conditions generally have cloud bases considerably higher than close of clouds forming in moist environments. This leads to an increase in the evaporation rate of the falling rain, resulting in less precipitation reaching the ground. For example, Rosenfeld and Mintz (1988) analyzed radar rainfall data from the semiarid region of central south Africa and found that even at rain intensities as high as 80 mm h−1, about 15% of the rain evaporated by 1 km and 30% by 1.6 km below the cloud base level. Global datasets containing estimates of water vapor amounts and fluxes will be presented in this section to describe the moisture distribution and fluxes of equatorial central Africa and to compare this to the region of South America where the gauge rainfall is in better agreement with the satellite estimates.
Water vapor
Information about the global distribution of water vapor is provided by the National Aeronautics and Space Administration Water Vapor Project (NVAP) dataset (Randel et al. 1996). This dataset contains global, 7-yr (1988–94), 1° × 1° spatial resolution, atmospheric water vapor products for three layers (1000–700, 700–500, and 500–300 mb). The analyses combine water vapor retrievals from the Television and Infrared Operational Satellite (TIROS) Operational Vertical Sounder (TOVS), the SSM/I, and radiosonde observations. The layered products were constructed by partitioning the total column water vapor using information from TOVS and radiosondes.
The layered NVAP product (Fig. 7) provides some insight into the possible contribution of water vapor to the overestimation problem. The upper (500–300 mb) and middle (700–500 mb) levels show relatively constant water vapor amounts along any latitude band, with more water vapor as the latitude bands become closer to the equator. However, for the lower-level (surface–700 mb) moisture, there is a sharp decline in water vapor over central Africa. While the lower-level precipitable water typically has values between 25 and 40 mm along the equator (including the land areas of equatorial South America), the values drop to below 25 mm over central Africa. The range of total column water vapor over equatorial South America is between 40 and 50 mm, while the values over equatorial Africa range between 30 and 40 mm. This lower level dryness in Africa may result in higher cloud bases and significant evaporation of raindrops, resulting in less rain reaching the ground. Analysis of mean soundings for July and January from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis data (Kalnay et al. 1996) support this in that the lifting condensation levels over Africa are 20 mb (40 mb) lower in pressure, that is, higher in elevation, than over South America in July (January). Since satellite estimates relate rainfall to cloud height (low temperature for IR, scattering for microwave), a drier atmosphere, especially in the lower levels, will result in overestimation of rainfall.
Another way of assessing the magnitude of the satellite estimates is through precipitation recycling. A simple but effective definition of precipitation recycling is given as the daily average precipitation divided by the average precipitable water. This is also known as precipitation efficiency as defined by Sellers (1965) and Kessler (1967). This ratio can be thought of as the fraction of average moisture overhead that falls as precipitation. Using the NVAP water vapor and two precipitation estimates (MS and SG), we calculated the precipitation efficiency for the entire latitude belt between 20°N and 20°S. Looking at the MS data (Fig. 8a) we find that the precipitation efficiency over Africa is somewhat greater than 25% and, perhaps more important, greater than anywhere else in the Tropics. When using the blended data analysis, which over land has the large-scale values adjusted by the gauges, the precipitation efficiency over Africa is comparable to other tropical rainy areas. This offers further evidence that the satellite estimates of rain over Africa may be too high.
Moisture flux
To study reasons for the low water vapor amounts in equatorial central Africa in comparison with other equatorial regions, NCEP–NCAR reanalysis data were used. The reanalysis dataset was created using a frozen state-of-the-art weather analysis/forecast system and assimilating past data to create a long-term record of global atmospheric fields. This dataset has been used recently (Higgins et al. 1996; Mo and Higgins 1996) for moisture studies similar to the one done here for equatorial Africa and South America.
The NCEP–NCAR reanalysis data support the results from the NVAP data showing lower moisture contents over the region of Africa than over South America. Figure 9 shows the time series of monthly mean precipitable water calculated from NCEP data. The precipitable water values over South America are in general 10 mm (∼30%) higher than those over Africa for the 7-yr period of study.
The NCEP annual average vertically integrated moisture fluxes [in units g (cm s)−1] for 1988–94 for Africa and South America are displayed in Figs. 10 and 11, respectively. The vertically integrated moisture flux indicates that moisture flux from the Indian Ocean appears to provide the bulk of moisture to central Africa. However, when looking at the lowest layer (1000–865 mb), where the flux of moisture is greatest, we see that the flux is blocked by the mountains in eastern Africa and that a small but significant amount of low-level moisture is transported from the Atlantic Ocean. The middle tropospheric layer of 865–500 mb shows the most consistent and larger westward flux of moisture over central equatorial Africa. In contrast, the moisture flux pattern of equatorial South America (Fig. 11) shows significant flux from the east, where there is no barrier to moistening the low levels. These moisture flux results may explain the low values of moisture in the surface–700-mb layer noted in the NVAP data over Africa and account for the differences in total column water vapor between the two regions.
However, these flux computations should be taken with caution, because the reanalysis dataset is new and its uncertainties are not fully understood. For example, Higgins et al. (1996) show that the change in the overall moisture budget due to the assimilation of observations with the model predictions is as large as the predicted moisture divergence itself, which indicates rather large uncertainty in the flux terms from the prediction model. Trenberth and Guillemot (1995) found that biases in the reanalysis can be related to steep orography. The study regions are bounded by mountain ranges (to the east for Africa and to the west for South America), so the estimated fluxes in these vicinities may be biased. Nevertheless, some degree of confidence is given to these calculations because all the results presented here are consistent with each other. The region of low-level moisture deficit indicated by the NVAP dataset is consistent with the area of satellite overestimation, and the moisture flux for these regions is consistent with the estimated moisture contents and the orography.
Summary and discussion
The problem of satellite rainfall overestimation over equatorial central Africa is an intriguing problem with several possible solutions.
The analysis of a long-term climatology of rain gauges has led to the conclusion that the sample of rain gauges used in the GPCP is not the reason for the difference between the satellites and the gauge analysis or the blended satellite–gauge analysis. One can also conclude that the analysis methods are operating properly by adjusting biased satellite estimates to a more reasonable value while still allowing the satellites to influence the distribution of rainfall. Accepting the reasonableness of the gauge analysis leads toward the other possible causes, that is, cloud microphysical and rain processes and the moisture distribution of the environment. In the case of the former it was argued that the North African deserts and semiarid areas of the Sahel provide a rich source of CCN. Many CCN can result in smaller cloud drops for the same amount of moisture, resulting in less-efficient coalescence and accretion processes and thus less-efficient rain processes. A comparison of effective cloud droplet radii derived from a sample of AVHRR 3.7-μm radiances suggests a higher concentration of CCN over central Africa than that over South America.
An analysis of the moisture distribution and the moisture flux was also consistent with less rainfall over Africa than over South America. The vertical distribution of moisture shows that over Africa the total column water vapor was about 1 cm less than over South America, and the surface–700-mb layer is relatively dry in comparison with South America and other tropical locations. This implies that the cloud bases may be high and that there may be significant evaporation of rain before reaching the surface.
The distribution of moisture fluxes as calculated from the NCEP–NCAR reanalysis indicates that the moisture flux over both Africa and South America has a large westward component in the low levels from the Indian Ocean and the Atlantic Ocean, respectively. However, over Africa this flux is blocked from the interior by high terrain in east Africa, whereas over South America there is no impediment to the flux of moisture from the Atlantic. The blockage of maritime moisture in the low levels suggests that equatorial central Africa may have drier continental air with greater aerosol amounts and CCN than the more-maritime air over South America.
Our analysis has suggested that both the moisture distribution and aerosol content are acting in concert in reducing the efficiency of rain processes over central Africa. There may well be other factors such as the strength of the vertical motion and the structure of individual storms that can contribute to the overall distribution of rain over Africa. Also, the physical processes presented here are based mostly on indirect observations and need to be confirmed with specific case studies and/or field programs. In that regard, National Aeronautics and Space Administration Tropical Rainfall Measuring Mission observations from the precipitation radar and collocated microwave, visible, and infrared data can help in analyzing the rain rate and providing some information on the microphysical properties of raining convective clouds as was done by Ba et al. (1998), who used AVHRR and SSM/I data to deduce some microphysical properties of convective clouds.
It appears that, over central equatorial Africa, the cloud systems are less maritime in nature and exist over a drier atmosphere than most tropical cloud systems. As a result, the infrared and microwave remote sensing techniques, which use a single model everywhere, significantly overestimate the amount of rainfall in central Africa that actually reaches the surface. This suggests separate models, at least for this region, may be required if remote sensing procedures are to yield reasonable magnitudes. As an alternative, these results suggest the possibility of using other remotely sensed data such as effective drop radii distribution as additional parameters in rainfall estimates. This is an area of future research.
Acknowledgments
We would like to thank Wayne Higgins, Sharon Nicholson, and Daniel Rosenfeld for fruitful discussions concerning our study and the data we used. Also, thanks to Ivan Csiszar, Kingtse Mo, and Julian Wang for their help in providing the data. Finally, special thanks to R. Ferraro and the anonymous reviewers for their thorough comments that greatly improved the manuscript. This work was supported in part by the NOAA Office of Global Programs Grant 8R1DA1AG, and the first author was partially supported by the National Research Council Research Associateship Program.
REFERENCES
Andreae, M. O., and Coauthors, 1992: Ozone and Aitken nuclei over equatorial Africa: Airborne observations during DECAFE 88. J. Geophys. Res.,97, 6137–6148.
Arkin, P. A., and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–84. Mon. Wea. Rev.,115, 51–74.
Ba, M. B., D. Rosenfeld, and A. Gruber, 1998: AVHRR multispectral derived cloud parameters: Relationships to microwave scattering signature and to cloud-to-ground lightning. Preprints, Ninth Conf. on Satellite Meteorology and Oceanography, Paris, France, Amer. Meteor. Soc., 408–411.
Christian, H. J., K. T. Driscoll, S. J. Goodman, R. J. Blakeslee, D. A. Mach, and D. E. Buechler, 1996: The Optical Transient Detector (OTD). Proc. 10th Int. Conf. on Atmospheric Electricity, Osaka, Japan, International Commission of Atmospheric Electricity and Society of Atmospheric Electricity of Japan, 368–371.
Cotton, W., and R. A. Anthes, 1989: Storm and Cloud Dynamics. Academic Press, 883 pp.
Désalmand, F., J. Baudet, and R. Serpolay, 1982: Influence of rainfall on the seasonal variations of cloud condensation nuclei concentrations in a sub-equatorial climate. J. Atmos. Sci.,39, 2076–2082.
Ferraro, R. R., 1997: Special Sensor Microwave Imager derived global rainfall estimates for climatological applications. J. Geophys. Res.,102, 16 715–16 735.
Fontan, J., A. Druilhet, B. Benech, and R. Lyra, 1992: The DECAFE experiments: Overview and meteorology. J. Geophys. Res.,97, 6123–6136.
Groisman, P. Y., and D. R. Legates, 1994: The accuracy of United States precipitation data. Bull. Amer. Meteor. Soc.,75, 215–227.
Higgins, R. W., K. C. Mo, and S. D. Schubert, 1996: The moisture budget of the central United States in spring as evaluated in the NCEP/NCAR and the NASA/DAO reanalyses. Mon. Wea. Rev.,124, 939–963.
Huffman, G. J., 1997: Estimation 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, and P. R. Keehn, 1995: Global precipitation estimates based on a technique for combining satellite-based estimates, rain gauge analysis, and NWP model precipitation information. J. Climate,8, 1284–1295.
Huffman, G. J., and Coauthors, 1997: The global precipitation climatology product (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc.,78, 5–20.
Johnson, D. B., 1980: The influence of cloud-base temperature and pressure on droplet concentration. J. Atmos. Sci.,39, 2076–2082.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc.,77, 437–471.
Kaufman, Y. J., and T. Nakajima, 1993: Effect of Amazon smoke on cloud microphysics and albedo—Analysis from satellite imagery. J. Appl. Meteor.,32, 729–744.
Kessler, E., 1967: On the continuity of water substance. Tech. Memo. IERTM-NSSL 33, U.S. Department of Commerce Environmental Science Services Administration, 125 pp.
Mo, K. C., and R. W. Higgins, 1996: Large-scale atmospheric moisture transport as evaluated in the NCEP/NCAR and the NASA/DAO reanalyses. J. Climate,9, 1531–1545.
Morrissey, M. L., J. A. Maliekal, J. S. Greene, and J. Wang, 1995: The uncertainty in simple spatial averages using rain gauge networks. Water Resour. Res.,31, 2011–2017.
Nakajima, T., and M. D. King, 1990: Determination of the optical thickness and effective particle radius of clouds from reflected solar radiation measurements. Part I: Theory. J. Atmos. Sci.,47, 1878–1893.
Nicholson, S. E, 1993: An overview of African rainfall fluctuations of the last decade. J. Climate,6, 1463–1466.
Randel, D. L., T. H. Vonder Haar, M. A. Ringerud, G. L. Stephens, T. J. Greenwald, and C. L. Combs, 1996: A new global water vapor dataset. Bull. Amer. Meteor. Soc.,77, 1233–1246.
Rosenfeld, D., and Y. Mintz, 1988: Evaporation of rain falling from convective clouds as derived from radar measurements. J. Appl. Meteor.,27, 209–215.
Rosenfeld, D., and G. Gutman, 1994: Retrieving microphysical properties near the tops of potential rain clouds by multispectral analysis of AVHRR data. Atmos. Res.,34, 259–283.
Rudolf, B., 1993: Management and analysis of precipitation data on a routine basis. Proc. Int. WMO/IAHS/ETH Symp. on Precipitation and Evaporation, Bratislava, Slovakia, Slovak Hydrometeor. Inst., 69–76.
Rudolf, B., H. Hauschild, W. Rueth, and U. Schneider, 1994: Terrestrial precipitation analysis: Operational method and required density of point measurements. NATO ASII/26, Global Precipitations and Climate Change, M. Desbois and F. Désalmand, Eds., Springer Verlag, 173–186.
Sellers, W. D., 1965: Physical Climatology. University of Chicago Press, 272 pp.
Trenberth, K. E., and C. J. Guillemot, 1995: Evaluation of the global atmospheric moisture budget as seen from analyses. J. Climate,8, 2255–2272.
Wilheit, T., A. Chang, and L. Chiu, 1991: Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution function. J. Atmos. Oceanic Technol.,8, 118–136.
WCRP, 1986: Report on the workshop on global large-scale precipitation datasets for the World Climate Research Program. WCRP-111, WMO/TD-No. 94, 45 pp. [Available from the World Meteorological Organization, P. O. Box 2300, CH-1211 Geneva 2, Switzerland.].
Mean daily estimated rainfall (mm) from 1988–94 for the (a) merged satellite product (MS) and the (b) rain gauge product (GPCC), and (c) the difference between the two products (MS − GPCC).
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Monthly mean time series for SG, GPCC, and MS daily precipitation (mm) over (a) equatorial central Africa, 10°S–10°N and 20°–30°E; and (b) equatorial South America, 10°S–10°N and 75°–65°W.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Differences in mean daily estimated rainfall (mm) from 1988–94 for (a) GPI and rain gauge estimates (GPI − GPCC) and (b) scattering algorithm estimates and rain gauge estimates (SSM/I − GPCC).
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Mean number of rain gauges in each 2.5° × 2.5° grid box from (a) the GPCC dataset and (b) the long-term dataset.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Mean daily rainfall (mm) from the quantities of rain gauges shown in Fig. 4.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Effective liquid droplet radius for Sep 1989 based on AVHRR 3.7-μm data.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Liquid equivalent water vapor for different vertical levels from the NVAP dataset.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Precipitation efficiency calculated from mean rainfall and mean water vapor from 1988–94.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Time series of monthly mean precipitable water from the NCEP–NCAR reanalysis dataset for the study areas in Africa and South America.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Multilevel NCEP–NCAR vertically integrated moisture flux [units: g (cm s)−1] for Jan 1988–94, for the African continent. Topography is indicated by the shading at 400-m intervals (starting at 800 m).
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
Same as Fig. 10 for the study region of South America.
Citation: Journal of Applied Meteorology 39, 5; 10.1175/1520-0450-39.5.666
