Prediction of lake evaporation is very much essential for effective water resources planning, operation and management. In India, usually, the lake evaporation is estimated from the pan evaporation and the average water spread area. Accurate prediction of lake evaporation by conventional method is a cumbersome process, since it is in non-linear relationship with the storage and other meteorological parameters. The recently evolved soft computing techniques are proved to be efficient to model these non-linear hydrological processes. Thus in the present study, two artificial neural network algorithms (ANN) namely, multi-layer perceptron (MLP) and time lagged recurrent neural network (TLRN) are compared to predict the lake evaporation. The daily Shivajisagar lake evaporation data collected from the Koyna dam circle for a period of 49 years has been used in the modelling. About 70% of the dataset is used for training the ANN models and the remaining 30% is used for testing. It is found that both the ANN algorithms predicted the lake evaporation very well with a correlation coefficient around 0.99. This shows that, if the input data series exhibits good pattern with less noise, the soft computing techniques results in better performances.

Key words: lake evaporation, artificial neural network, multi-layer perceptron, time-lagged recurrent neural network, Koyna dam Shivajisagar Lake