Application of neural nets to rain rate retrieval from simulated multichannel passive microwave imagery

The retrieval of precipitation parameters from passive microwave imagery requires the use of a nonlinear statistical estimation technique operating on a multichannel data stream. The authors describe the application of a feed-forward neural net to the estimation of precipitation parameters from noisy, multispectral brightness temperature imagery. A comparison of the neural net retrieval technique vis-and-vis a nonlinear statistical technique shows that with proper training the neural net achieves slightly better performance. The neural net technique developed has been used to study the tradeoffs associated with using various microwave channel sets and sensor resolutions. Initial calculations using simulated nadir-viewing imagery for a MIMR-compatible set of six channels are presented