Modeling evapotranspiration: Some issues resolved

Evapotranspiration (ET) is an essential parameter for estimation of irrigation water requirements. Climatic variables (CV) along with soil and plant variables are known to influence ET exhibiting completely different patterns at some locations because of multi-collinearity (MC). These factors, coupled with the complex phenomenon of transpiration have been largely responsible for the development of a large number of ET models. However, none of the models has been found to be satisfactory for all locations. Researchers have analyzed the causes as the lack of understanding of (i) the physics of transpiration, (ii) the separation of the process of transpiration from pure evaporation. Researchers now look for Artificial Neural Networks (ANN). They have claimed that use of ANN is very promising and competing against the existing popular models. But the problem with ANN is two-fold: several trials are involved in training with different combinations of variables, and requirements of large number of reliable data. This background has provided motivations for the study in this paper. The goal is to develop a procedure for identifying the minimum number of variables which must be considered for use in the ANN approach, and serve to choose the particular model/models for the location. The experiments conducted in this study use climatic data (CD) of over fifteen locations, falling in different climatic regions of the world. It is shown that much of the labor and cost involved in selecting the best combination of variables in ANN can be drastically reduced. Also, it reduces the burden of data collection program, besides answering the question of why some models perform poorly and some others do well. Tools employed for development of the procedure are: singular value decomposition (SVD), and variance decomposition proportions (VDP).