Simulation of Reference Crop Evapotranspiration Based on PSO-RF in the Hanjiang Basin
Sixteen daily scale ET0 calculation models were developed using the particle swarm optimization algorithm and random forest(PSO-RF)based on daily meteorological data collected from 24 meteorological stations in the Hanjiang dasin and it's vicinity from 1960 to 2017,and were compared with Hargreaves-Samani,Irmark-Allen and Makkink,which applicability and portability were evaluated.The results showed that:(1)the non-linear relation between daily ET0 and each input parameter can be well displayed by PSO-RF model.The PSO-RF2 established by Tmax,Tmin and Ra had sufficient accuracy(MAE=0.402 mm·d-1,RMSE=0.575mm·d-1,R2=0.863).With the increase of meteorological elements,the computational accuracy of PSO-RF model was improved(MAE decreased by 37.9%,RMSE decreased by 36.3%,R2 increased by 6.78%).(2)When the input factors and number of models are the same,the PSO-RF model had higher simulation accuracy than Hargreaves-Samani,Irmark-Allen,and Makkink models(MAE decreased by 43.6%,RMSE decreased by 34.5%,R2 increased by 4.12%).(3)The PSO-RF model had strong generalization ability and portability in the Hanjiang basin.High accuracy was shown in mutual portability of PSO-RF models established in different zone(MAE=0.159mm·d-1,RMSE=0.245mm·d-1,R2=0.974).Therefore,the model based on PSO-RF can be used as a recommended model for the calculation of daily ET0 in Hanjiang basin in the absence of meteorological data.
万方数据
维普
