Rainfall estimation from vertical profiles of reflectivity using neural networks

摘要

The neural network is a nonparametric method for representing the relationship between radar measurements and rainfall rate. Recent research had demonstrated that neural network techniques can be successfully used for ground rainfall estimation from radar measurements. An adaptive neural network has been developed to estimate rainfall rate from vertical profiles of reflectivity that gradually adapts itself over time, without retraining from the beginning. Such a network is also computationally stable. The performance of the neural network is evaluated by conducting tests on data sets using WSR-88D over Melbourne, FL and surface gage network data during 1998, 1999. The results show that the adaptive neural network can estimate rainfall fairly accurately and consistently.