This study presented the application of partial least squares regression （PLSR） in estimating daily pan evaporation by utilizing the unique feature of PLSR in eliminating collinearity issues in predictor variables. ...This study presented the application of partial least squares regression （PLSR） in estimating daily pan evaporation by utilizing the unique feature of PLSR in eliminating collinearity issues in predictor variables. The climate variables and daily pan evaporation data measured at two weather stations located near Elephant Butte Reservoir, New Mexico, USA and a weather station located in Shanshan County, Xinjiang, China were used in the study. The nonlinear relationship between climate variables and daily pan evaporation was successfully modeled using PLSR approach by solving collinearity that exists in the climate variables. The modeling results were compared to artificial neural networks （ANN） models with the same input variables. The resuits showed that the nonlinear equations developed using PLSR has similar performance with complex ANN approach for the study sites. The modeling process was straightforward and the equations were simpler and more explicit than the ANN black-box models.展开更多
基金supported in part by the National Natural Science Founda-tion of China (Grant Nos.51069017,41071026)their sincere appreciation of the reviewers’ valuable suggestions and comments in improving the quality of this paper
文摘This study presented the application of partial least squares regression （PLSR） in estimating daily pan evaporation by utilizing the unique feature of PLSR in eliminating collinearity issues in predictor variables. The climate variables and daily pan evaporation data measured at two weather stations located near Elephant Butte Reservoir, New Mexico, USA and a weather station located in Shanshan County, Xinjiang, China were used in the study. The nonlinear relationship between climate variables and daily pan evaporation was successfully modeled using PLSR approach by solving collinearity that exists in the climate variables. The modeling results were compared to artificial neural networks （ANN） models with the same input variables. The resuits showed that the nonlinear equations developed using PLSR has similar performance with complex ANN approach for the study sites. The modeling process was straightforward and the equations were simpler and more explicit than the ANN black-box models.
