Journal of Applied Meteorology and Climatology | Hirose et al. 
Sixteen-year observations of the Tropical Rainfall Measuring Mission Precipitation Radar (TRMM PR) yielded hundreds of large precipitation-systems (≥100 km) for each 0.1° grid over major rainy regions. More than 90% of the rainfall was attributed to large systems over certain mid-latitude regions such as the La Plata basin and the East China Sea. The accumulation of high-impact snapshots reduced the significant spatial fluctuation of the rain fraction arising from large systems and allowed obtaining sharp images of the geographic rainfall pattern. Widespread systems were undetected over low-rainfall areas such as regions off Peru. Conversely, infrequent large systems brought significant percentage of rainfall over semi-arid tropics such as the Sahel. This demonstrated an increased need for regional sampling of extreme phenomena.
Differences in data collected over a period of 16-years was used to examine sampling adequacy. The results indicated that more than 10% of the 0.1°-scale sampling error accounted for half of the TRMM domain even for a 10-year data accumulation period. Rainfall at the 0.1°-scale was negatively biased in the first few years for over more than half of the areas due to a lack of high-impact samples. The areal fraction of the 0.1°-scale climatology with a 50% accuracy exceeded 95% in the ninth year, and those areas with rainfall >2 mm day−1 in the fifth year. A monotonic increase in the degree of similarity of fine-scale rainfall to the best estimate with an accuracy of 10% illustrated the need for further sampling.
Full text can be found here.