Evaluation of Level-2 Precipitation Estimates from Satellite-based Passive Microwave Radiometers

NASA's satellite-based precipitation data records are produced in two levels: level 2 and 3. Level-2 products are derived from satellite sensors at their footprint scales, while level-3 products are typically gridded data using multiple level-2 products and other sensors. Satellite-based multi-sensor rainfall products (level-3), mainly generated from Passive Microwave (PMW) sensor rainfall retrievals (level-2), have been among the most widely used rainfall datasets in various hydrological applications. However, satellite-based multi-sensor rainfall estimates contain errors which need ground validation (GV). Errors associated with these products are from two sources: the upstream sensors used and the algorithms to merge the sensor retrievals. Many previous researches (e.g., Tian et. al., 2009) have proved that, several satellite-based multi-sensor rainfall products, generated from disparate merging algorithms, share remarkable similarities in error characteristics. This suggests these errors can be traced back to their upstream sensor inputs. This paper is focusing on evaluating the measurement errors of the upstream passive microwave radiometers (using the level-2 products). Five PMW radiometers on board nine satellites have been studied, including both imagers (TMI, AMSR-E, SSM/I) and sounders (AMSU-B and MHS). A high-resolution ground radar-based dataset, the next generation multi-sensor QPE (Q2) data over the continental US, has been used as the GV data. The high spatial and temporal resolution of the reference data, allows rigorous collocation (within 5 minutes) and relatively more precise comparison with satellite overpasses. From our results, PMW sensor retrievals exhibit fairly systematic bias varying be seasons and rain intensity, with overestimates in summer at intermediate to high-end rain rates and underestimates in winter at intermediate rain rates. This feature is also observed in the merged products, suggesting the dominant contribution of the sensor errors to merged products. Our study also reveals that rain retrievals from the imagers have less bias than those from the sounders, especially in summer.