Particle Swarm Optimization of Multi-Linear Regression for Evapotranspiration Estimation Model

Abstract

The data demanding Food and Agricultural Organization-56 Penman-Montieth model (FPM-56) is the most accurate model in estimating evapotranspiration (ET) but it is not applicable at data scarce region. This paper evaluates the performance of conventional MLR models and improved MLR by using PSO algorithms (MLR-PSO) in estimating potential evapotranspiration (ETp) by only using 2 significant parameters affecting ETp for tropical climate. In this study, 17 meteorological stations around Peninsular Malaysia were used in this study and obtained its both MLR and MLR-PSO models. These models were compared by using root mean square error (RMSE), coefficient of determination (R2) and its accuracy (Acc). The obtained results show MLR models itself has accuracy closed to 94% against FPM-56 models. Whereas optimized MLR-PSO models has improved up to 2.95% of accuracy. Out of 4 PSO algorithm, the standard c1=c2=2.0 and w=1.0 resulted better performance in 7 stations compared to others. The results proves that MLR and MLR-PSO models both useful for estimating ETp at data scarce region as it required only 2 main parameters affecting ETp.