Training artificial neural networks to perform rainfall disaggregation

Hydrologists and engineers need methods to disaggregate hourly rainfall dat a into subhourly increments for many hydrologic and hydraulic engineering a pplications. In the present engineering environment where time efficiency a nd cost effectiveness are paramount characteristics of engineering tools, d isaggregation techniques must be practical and accurate. One particularly a ttractive technique for disaggregating long-term hourly rainfall records in to subhourly increments involves the use of artificial neural networks (ANN s). A past investigation of ANN rainfall disaggregation models indicated th at although ANNs can be applied effectively there are several consideration s concerning the characteristics of the ANN model and the training methods employed. The research presented in this paper evaluated the influence on p erformance of several ANN model characteristics and training issues includi ng data standardization, geographic location of training data, quantity of training data, number of training iterations, and the number of hidden neur ons in the ANN. Results from this study suggest that data from rainfall-gau ging stations within several hundred kilometers of the station to be disagg regated are adequate for training the ANN rainfall disaggregation model. Fu rther, we found the number of training iterations, the limits of data stand ardization, the number of training data sets, and the number of hidden neur ons in the ANN to exhibit varying degrees of influence over the ANN model p erformance.