A Statistical Modeling Approach to Passive Microwave Rainfall Retrieval

Abstract

A new empirical algorithm for retrieving rainfall rates from passive microwave (particularly Special Sensor Microwave/Imager) data is presented. Errors caused by spatial and temporal variation of surface temperature, emissivity, and atmospheric effects are minimized by modeling the nonraining brightness temperatures within 0.5° latitude by 0.5° longitude regions based on the statistics of the satellite data from the whole of the month prior to the date of interest. Displacement of data away from the modeled relationship by more than a threshold value, calculated from the standard deviation of the nonraining data, is detected as rainfall. The algorithm is able to detect rainfall over land, sea, and coasts. An initial calibration and validation is performed using data from WetNet Precipitation Intercomparison Project 2 over the British Isles. To overcome collocation errors during calibration and validation, techniques are presented for matching radar rain pixels with appropriate satellite data. The algorithm detects light rainfall less than 0.5 mm h?1, which is common at midlatitudes, and exhibits a large dynamic range suitable for measuring heavy rainfall. The critical success index ranges from 50% over land to 61% over sea. The algorithm is not limited geographically, although it is likely that the rain rate relationship would benefit from recalibration for other regions or from the inclusion of a physical inversion technique designed to retrieve the vertical structure of the precipitation.