Estimation of Daily Reference Evapotranspiration by Neuro Computing Techniques Using Limited Data in a Semi-Arid Environment

Abstract

In this paper, the daily reference evapotranspiration (ET<inf>0</inf>) for Bulawayo Goetz was estimated from climatic data using neuro computing techniques. The region lacks reliable weather data and experiences inconsistencies in the measuring process due to inadequate and obsolete measuring equipment. This paper aims to propose neuro computing techniques as an alternative methodology to estimating evapotranspiration. Firstly, ET<inf>0</inf> was calculated using FAO-56 Penman-Monteith (PM) equation from available climatic data. Data was divided into training, testing and validation for neuro computing purposes. The study also investigated the effect of different normalisation techniques on neuro computing ET<inf>0</inf> estimation accuracy. In another application, neuro-computing ET<inf>0</inf> estimates were compared against those obtained using empirical methods and their calibrated versions. The Z-score normalisation technique for all data sets gave best results with a Multi-layer perceptron (5–5-1) model having RMSE, MAE and R2 values in the range 0.12–0.25 mm day−1, 0.08–0.15 mm day−1 and 0.94–0.99 respectively. There were no significant differences in ET<inf>0</inf> estimation accuracy by neuro computing techniques due to normalisation technique. The Neuro computing techniques were superior to empirical methods in ET<inf>0</inf> estimation for Bulawayo Goetz. The Neuro computing techniques are recommended for use in cases of limited climatic data at Bulawayo Goetz. © 2017 Informa UK Limited, trading as Taylor & Francis Group.