Application of Back-Propagation Neural Network to Estimate Precipitation with Doppler Radar in Yishuhe Watershed of China

By means of the Doppler radar measurements and automatic precipitation station data collected during four precipitation processes of 2005 in the Yishuhe Watershed, the back-propagation neural network (BPNN) based on BFGS algorithm is used to train and estimate the rainfall. Reflectivity (Z) and rain intensity (R) relation are determined by an improved window probability matching method and used to verify and evaluate the precision of BPNN. The results suggested that the precision from BPNN is higher than from Z-R relation, especially in intensified rainfall process. The hourly rainfall and total accumulations of BPNN is in good consistence with rain gauge observation in intensified process and exists some extent overestimation in medium intensified process. Rainfall estimation of Z-R relation would yield underestimation of different degree with the change of rainfall intensity, the more underestimation, the more intensified rainfall process.