ARTIFICIAL NEURAL NETWORK MODELING FOR ESTIMATION OF REFERENCE EVAPOTRANSPIRATION AT NAGPUR

M. M. Deshmukh

Abstract: Precise estimation of reference crop evapotranspiration (ETo) is of paramount importance in water resources planning. ETo depends on several interacting climatological factors. Artificial Neural Networks (ANN) is effective tool and universal approximators to model complex and nonlinear process like ETo. The multilayer back propagation feed forward neural networks were developed for estimating ETo at Nagpur having hot dry sub-humid climate, using different ANN model strategies with different input combinations, as ANN1 (Tmax, Tmin), ANN2 (Tmax, Tmin, SH), ANN3 (Tmax, Tmin, RHmax, RHmin), ANN4 (Tmax, Tmin, RHmax, RHmin, SH), and ANN5 (Tmax, Tmin, RHmax, RHmin, SH, WS). In ANN model validation, using independent evaluation data set of the same location, mean absolute error (MAE), mean absolute relative error (MARE), root mean square error (RMSE), correlation coefficient (r), determination coefficient (R2 ), index of agreement (D) and model efficiency (E) were found to be good enough for all network models and showed that ANN1 to ANN5 models are suitable and can be used for estimation of ETo, according to the availability of data. However, Nagpur ANN5 model emerged as best model and ranked first followed by ANN2, ANN4, ANN1 and ANN3 models. Nagpur ANN5 model provided accurate estimation of ETo with 0.2200 mm day-1 RMSE and 0.9777 model efficiency and followed by Nagpur ANN2 model with 0.4147 mm day-1 RMSE and 0.9207 model efficiency.

Keywords: Reference evapotranspiration, Artificial neural networks, Climatological factors

DOI: https://doi.org/10.15623/ijret.2016.0530005