TR(BR)-20/99-2000 : Investigation on the capability of artificial neural network for estimating evapotranspiration from minimum weather data

Abstract

Most of the current hydrologic, water management, and crop growth model require an accurate estimate of evapotranspiration (ET), for reliable applications. A large number of methods for calculation of ET from weather data have been developed and tested for varying geographical and climatological conditions. However, most of these methods require weather data that are not widely available. A recent series of technical papers have discussed the capabilities of ANN to approximate any continuous input-output mapping to an arbitrary degree of accuracy. Accordingly, a research study was conducted to estimate ET from most widely available weather data. Three combinations of input data were considered and three different ANN models were developed. One of the models developed requires only average temperature as input and was estimating daily values of ET with 99% efficiency. The performance of ANN models was evaluated against that of popular ET estimating methods, and was found performing superior to others. The study demonstrated the applicability of ANN technique in accurately estimating ET from minimum weather data.