ARTIFICIAL NEURAL NETWORK MODEL FOR PREDICTION OF RAINFALL AND EVAPORATION

Abstract

In the present study, artificial neural network technique has been employed to predict daily rainfall and evaporation for Pattambi, Kerala. The total 6 years (2014-2019) meteorological data of months June, July, August and September was taken for rainfall prediction. Similarly for evaporation model 6 years (2014-2019) meteorological data was taken. Climatic variables namely, maximum temperature, minimum temperature, wind speed, relative humidity, sunshine hours were taken as input parameters and rainfall and evaporation were taken as output parameters for prediction. For both rainfall and evaporation prediction models 75% of the data were used for training and 25% of the data were used for testing. Statistical data analysis was carried out to identity important climatic variables for developing ANN model. In this study the MATLAB software was used for ANN model development having neural network tool with different network architectures, transfer function, learning functions etc. The networks were developed by using feed forward hack propagation with log sigmoid and tan sigmoid transfer functions. The performance of the models were evaluated quantitatively by using different performance indices viz. root mean square error, correlation coefficient and coefficient of determination. The model with highest value of correlation coefficient (R) and lowest value of root mean square error (RMSE) is considered as the best fit model. It was observed that log sigmoid transfer function is capable of predicting the evaporation and tan sigmoid transfer function is capable of predicting the rainfall with almost equal prediction efficiency. For evaporation simulation, model 4 is found to be the best fit model having R and RMSE as 0.5954 and 2.060 respectively. And with normalized data for evaporation simulation, model 3 is found to be the best fit model having R and RMSE as 0.553 and 2.077 respectively. For rainfall simulation, model 1 is found to be the best fit model having R and RMSE as 0.544 and 11.132 respectively. These models may be used to predict daily rainfall and evaporation of Pattambi, Kerela.