We studied two recent high‐resolution precipitation datasets derived from satellite remote‐sensing, TRMM 3B42 and CMORPH, and compared them with ground‐based radar and gauge measurements in the southeastern U. S. We found there are systematic differences in rainfall estimates between inland water‐body pixels and land pixels. On average, there are about twice as many raining days over water bodies than over land pixels in the satellite products, causing much higher false alarm rates over water bodies. The increased false alarms occur mostly in the form of light rain (<2 mm/day), and lead to significantly different rain rate distributions between water‐body and land pixels. We speculate that this inconsistency is caused by deficiencies in emissivity characterization for the passive microwave‐based rainfall retrievals that serve as input to these merged products.