In the present research, long-range prediction of average summer monsoon rainfall over India has been attempted through three layered artificial neural network models. The study is based on the summer monsoon data pertaining to the years 1871–1999. Nineteen neural network models have been developed with variable hidden layer size. Total rainfall amounts in the summer monsoon months of a given year have been used as input and the average summer monsoon rainfall of the following year has been used as the desired output to execute a supervised backpropagation learning procedure. After a thorough training and test procedure, a neural network with eleven nodes in the hidden layer is found to be the most proficient in forecasting the average summer monsoon rainfall of a given year with the said predictors. Finally, the performance of the eleven-hidden-nodes three-layered neural network has been compared with the performance of the asymptotic regression technique. Ultimately it has been established that the eleven-hidden-nodes three-layered neural network has more efficacy than asymptotic regression in the present forecasting task.
Identification of the best hidden layer size for three-layered neural net in predicting monsoon rainfall in India
Surajit Chattopadhyay, Goutami Chattopadhyay; Identification of the best hidden layer size for three-layered neural net in predicting monsoon rainfall in India. Journal of Hydroinformatics 1 March 2008; 10 (2): 181–188. doi: https://doi.org/10.2166/hydro.2008.017
Download citation file: