Skip to main content
Climate & Radiation
Geodesy and Geophysics
Wallops Field Support
Code 614.3 LIS Science Meeting
Ali Behrangi, Center for Hydrometeorology and Remote Sensing (CHRS), The Henry Samueli School of Engineering, University of California, Irvine
Tuesday, August 5, 2008 - 23:00
Data from Geosynchronous Earth Orbiting (GEO) satellites equipped with visible and Infra-Red (IR) imagers are commonly used for rain retrieval due to the high spatial and temporal resolution of GEO observations, despite the lower quality compared to passive microwave observations from Low Earth Orbiting (LEO) satellites. Although GEO-based precipitation retrieval algorithms have begun to use data from spectral bands other than the longwave IR window, the potential benefit of these additional bands for precipitation retrievals has not yet been fully quantified. Here, we present a Neural Network-based framework to evaluate the utility of multi-spectral information in improving both rain/no-rain detection and rain rate estimation. The proposed algorithm consists of three stages. The first training stage involves feature selection and, if necessary, spectral dimension compression of multi-spectral images using the principal component analysis technique. The second training stage uses a feature classification scheme based on the self-organizing feature map and probability matching method against coincident LEO passive microwave precipitation estimates, resulting in rain probability and intensity estimates for each multi-spectral class. In the final stage, the trained neural network is applied to all of the GEO data to detect the occurrence of rain and estimate rain rate. Multi-spectral images from the current-generation GOES Imager and the Spinning Enhanced Visible and Infrared Imager (SEVIRI) instrument of Meteosat Second Generation satellites were used for this experiment. Since SEVIRI and future GOES-R Advanced Baseline Imager possess many common spectral channels, the capabilities demonstrated using SEVIRI image channels should be representative of expected GOES-R performance. Detailed examination of case studies in addition to overall statistics indicates that multi-spectral data are beneficial for screening out no-rain pixels associated with cold thin clouds, and identifying rain areas under warm, but rainy clouds. Both situations are problematic for IR-only algorithms. Overall, the improvement is significant for delineating the areal extent of precipitation (about 50%) and less so for estimating precipitation intensity (about 20%). Thus, multi-spectral data are found to promise improvements for precipitation retrievals from GEO platforms.