A simple approach for improving spatial interpolation of rainfall using ANN

Abstract

Interpolation of hydrological variables such as rainfall, ground water level, etc. is necessary to arrive at many engineering decisions. This study suggests an innovative and yet simple idea to effectively interpolate the rainfall at non-sampling locations using artificial neural network (ANN). The method has been demonstrated in the Tamirabarani basin of Tamil Nadu State (India), where rain gauge information for 18 rain gauge stations is available. With the help of land use map of the basin and also the proximity of rain gauge stations to each other in the neighbourhood, the most appropriate input variables for ANN are designed and used for training ANN to estimate rainfall in the unknown stations. It is also interesting to note that with appropriate selection of input variables for training ANN, one of the major lacunae of ANN (to have sufficiently lengthy training records) can be suitably addressed. ANN results are compared with Kriging method. Further, the proposed method is applied to improve the prediction of inflow to Ramanadhi reservoir in Tamirabarani basin, and the results seem to be very encouraging.