Developing a Combined Model of Multiple Linear Regression-Principal Componenet and Factor Analysis (MLR-PCA) for Estimation

Reference evapotranspiration (ETo) is an essential parameter required for proper management of agricultural crop irrigation. ETo is influenced by many different hydrological variables and as a result is a very complex procces. ETo is usually estimated by empirical or process-orinented models (mathematical relationships) from historical weather data. The need for accurate estimates of ETo and the complexity of developing models to describe such complex process magnifies the need for developing new data mining methods. In this paper, the possibility of using a combined method of multiple linear regressions with principal componenets analysis (MLR-PCA) for estimating reference evapotranaspiration was investigated. In this analysis, measured daily meteorological data of Kerman synoptic weather station recorded from 1996 to 2005 were used. Three principal componenets that explained 80% of the total variance of the data were recognized as the principle componenets and others as disorder. Using the extracted principle componenets, a multiple linear regression model was developed to estimate ETo. The statistic index of t for assessing the results of a fixed constant and each componenets of PC1 and PC2 were determined. According to the results, all coefficients were significant at the level of 95% and PC1 had more importance than the other component namely PC2. This revealed that the variables of radiation intensity, relative humidity, sunshine hours, minimum temperature and maximum temperature had more importance in estimating reference evapotranspiration than other climatological parameters. Comparison of MLR-PCA model with Penman-Monteith results showed that about 82% of the total amount of the ETo variance is defined by the three aformenstioned principle componenets.