The Forecasting of ET based on Artificial Neural Network-A Case Study in Tongzhou District of Beijing

摘要

The estimation of spatial variation of evapotranspiration (ET) in a catchment is fundamental to many applications in water resources and climate modeling. In order to enhance the water use efficiency, especially that of the agricultural sector, the crop water consumption by ET should be decreased. Thus, the calculating and forecasting of ET is getting more and more important. Based on the analysis of meteorological indices and the water use actuality affecting ET in Tongzhou District of Beijing, this paper investigated the prediction of ET using the artificial neutral network (ANN). Using the meteorological indices and agricultural water use as the inputs and ET as output, three ANN models were established in this study, e.g., six-factor input vector model, five-factor input vector model and four-factor input vector model. By using the monthly meteorological data, agricultural water use and remote sensing ET of Tongzhou during 2002-2004, the models were trained and used to forecast the ET in 2005. The results showed that the BP neural computing technique could be employed successfully in modeling ET. The precision of six-factor input vector model including the agricultural water use was the highest among the three. This conclusion showed that the agricultural water use is an important factor affecting ET in the study area. The four-factor input vector model using the air temperature, sunshine hours, precipitation and the agricultural water use had a considerable result, which can be used as an convenient and effective forecasting method for ET where data availability is limited.