A study of optimal model lag and spatial inputs to artificial neural network for rainfall forecasting

Artificial neural networks (ANNs), which emulate the parallel distributed p rocessing of the human nervous system, have proven to he very successful in dealing with complicated problems, such as function approximation and patt ern recognition. Due to their powerful capability and functionality, ANNs p rovide an alternative approach for many engineering problems that are diffi cult to solve by conventional approaches. Rainfall forecasting has been a d ifficult subject in hydrology due to the complexity of the physical process es involved and the variability of rainfall in space and time. In this stud y, ANNs were adopted to forecast short-term rainfall for an urban catchment . The ANNs were trained to recognise historical rainfall patterns as record ed from a number of gauges in the study catchment for reproduction of relev ant patterns for new rainstorm events. The primary objective of this paper is to investigate the effect of temporal and spatial information on short-t erm rainfall forecasting. To achieve this aim, a comparison test on the for ecast accuracy was made among the ANNs configured with different orders of lag and different numbers of spatial inputs. In developing the ANNs with al ternative configurations, the ANNs were trained to an optimal level to achi eve good generalisation of data. It was found in this study that the ANNs p rovided the most accurate predictions when an optimum number of spatial inp uts was included into the network, and that the network with lower lag cons istently produced better performance. (C) 2000 Elsevier Science B.V. All ri ghts reserved.