Evaluation of Artificial Neural Networks for Estimating Reference Evapotranspiration in Western Himalayan Region

This paper aims to examine the utility of neural networks (NN) for estimating reference evapotranspiration (ET
₀
) in the western Himalayan region. The study considered meteorological data of 3 sub-regions (Jammu-Kashmir, Uttarakhand and Himachal Pradesh) comprising 47 stations for the period 1979-2011. Two NN models with different input data combinations were used in the study. Model A1 was comprised of 5 input variables (minimum temperature, relative humidity, maximum temperature, wind speed, and solar radiation) and model A2 was comprised of 2 input variables (maximum and minimum temperature). NN modelling was performed using MATLAB. NN predicted ET
₀
was compared with the ET
₀
computed using the FAO-56 Penman-Monteith method. The performance of the NN models was evaluated using mean absolute error, root mean squared error and R
²
values. Model A1 performed relatively better than A2 for all the sub-regions, however, model A2 performed better than A1 during the testing stage for most locations. The error statistics indicate that the performance of A2 is comparable to A1 and can be effectively utilized to estimate ET
₀
in absence of sufficient climatic data. A comparison was also conducted among the three sub-regions in the study area. The study validates the performing capability of NN in estimating ET
₀
with sufficient as well as a limited set of climatic data.