基于NARX模型的参考作物蒸散发预测

为了实现气象资料缺失情况下参考作物蒸散量(ET₀)精确计算和预测,以攀枝花站点为例,构建非线性自回归滤波器(NARX)模型预测ET₀.以Penman-Monteith模型计算的ET₀作为预测标准,将日最高温、日最低温、日照时数、风速和相对湿度作为模型的输入参数,建立11种不同气象因子组合的NARX模型,并与Hargreaves-Samani模型、Irmak-Allen模型、Makkink模型、Priestley-Taylor模型的计算结果进行比较.结果表明:不同气象因子输入下的NARX模型模拟精度表现出明显的差异.其中,基于全部气象因子的NARX-1模型的RMSE,MAE和MBE分别为0.425 mm/d,0.320 mm/d和0.069 mm/d,NSE为0.920,GPI排名第11,精度最差;而基于风速的NARX-9模型精度最高,其RMSE,MAE和MBE分别为0.285 mm/d,0.237 mm/d和0.019 mm/d,NSE为0.964,GPI排名第1.在输入相同气象参数情况下,基于温度和日照时数的NARX-4模型模拟精度优于Irmak-Allen,Makkink,Priestley-Taylor模型;基于温度的NARX-7模型模拟精度明显高于Hargreaves-Samani模型.因此,可将NARX模型作为四川西南山地缺失较多气象资料情况下计算ET₀的推荐模型,为农田精准灌溉提供科学依据.

Prediction of reference crop evapotranspiration based on NARX model

In order to obtain the accurate estimation and prediction of reference crop evapotranspiration(ET₀)in the absence of meteorological data, a nonlinear autoregressive exogenous inputs(NARX)model was constructed to predict ET₀ in Panzhihua station. Taken the daily ET₀ calculated by the Penman-Monteith model as the prediction standard values, the NARX models with 11 different combinations of meteorological factors was established by taking the daily maximum temperature, daily minimum temperature, sunshine hours, wind speed and relative humidity as the input parameters of the model. The calculation results of NARX models were compared with that of Hargreaves-Samani model, Irmak-Allen model, Makkink model and Priestley-Taylor model. The results show that the discrepancy of simulation performance is obvious for NARX model with different input meteorological factors. Among them, the RMSE, MAE, MBE of NARX-1 model is 0.425 mm/d, 0.320 mm/d, 0.069 mm/d, respectively, and the NSE is 0.920, GPI ranking the 11th in GPI with the worst simulation accuracy. NARX-9 model which is based on wind speed has the highest accuracy that RMSE, MAE, MBE is 0.285 mm/d, 0.237 mm/d, 0.019 mm/d, respectively, and the NSE is 0.964 and GPI ranks the 1. When inputting the same climate parameters of temperature and sunshine hours, the simulation accuracy of NARX-4 is better than that of Irmak-Allen, Makkink, Priestley-Taylor model. While, if inputting parameter is only temperature, the simulation accuracy of NARX-7 is better than that of Hargreaves-Samani model. Therefore, the NARX model can be used as a recommended model for the calculation of ET₀ with limited meteorological data in southwestern mountain areas of Sichuan, and can provide a scientific evidence for accurate irrigation of farmland.