Neural network based prediction models for evaporation

Abstract

From statistical perspective, artificial neural networks (ANNs) are interesting because of their potential use in prediction. In this study, ANNs based approach has been used to assess the prediction of evaporation with meteorological variables, viz., maximum temperature (MaxT), minimum temperature (MinT), relative humidity in the morning (RHI), relative humidity in evening  (RHII), bright sunshine hours (BSH) and wind speed (WS) for different locations (Una, Karnal, Pantnagar, Raipur, Anantpur, Bangalore and Pattambi) in India. ANNs models were developed using Multilayer perceptron (MLP) architecture with two-phase algorithm of Backpropagation (BP) and Conjugate gradient descent (CGD) for prediction of evaporation as output and different combination of meteorological variables as input in different locations. Weekly predictions of evaporation have been obtained for subsequent years not included in model development. The performances of the developed models with different combination of weather variables compared based on mean absolute percentage error (MAPE). The sensitivity analysis indicated that the mean temperature and mean relative humidity are more sensitive to evaporation.