Fusion of surface radar and satellite rainfall data using feature calibration and alignment

Discrepancies between estimates of rainfall from ground-based radar and sat ellite observing systems tan be attributed to either calibration difference s or to geolocation and sampling differences. These latter include differen ces due to radar or satellite misregistration, differences in observation t imes, or variations in instrument and retrieval algorithm sensitivities. A new methodology has been developed and tested for integrating radar- and sa tellite-based estimates of precipitation using a feature calibration and al ignment (FCA) technique. The parameters describing the calibration and alig nment are found using a variational approach, and are composed of displacem ent and amplitude adjustments to the satellite rainfall retrievals, which m inimize the differences with respect to the radar data and satisfy addition al smoothness and magnitude constraints. In this approach the amplitude com ponent represents a calibration of the satellite estimate to the radar, whe reas the displacement components correct temporal and/or geolocation differ ences between the radar and satellite data The method has bean tested on a number of cases of the NASA WetNet PIP-2 da taset. These data consist of coincident estimates of rainfall by ground-bas ed radar and the DMSP SSM/I. Sensitivity tests were conducted to tune the p arameters of the algorithm. Results indicate the effectiveness of the techn ique in minimizing the discrepancies between radar and satellite observatio ns of rainfall for a variety of rainfall events ranging from midlatitude fr ontal precipitation to heavy convection associated with a tropical cyclone (Hurricane Andrew). A remaining issue to be resolved is the incorporation o f knowledge about location dependencies in the errors of the radar and micr owave estimates. Once the satellite data have been adjusted to match the radar observations, the two independent estimates (radar and adjusted SSM/I rain rates) may be blended to improve the overall depiction of the rainfall event in a single analysis. The FCA technique also has potential applications in 1) the deve lopment of satellite rainfall retrieval algorithms that may be tuned to rad ar rain rates and 2) error assessment of rainfall predictions using radar o r satellite rain rates as verification.