同化遥感信息与WheatSM模型的冬小麦估产

Estimation of winter wheat yield based on coupling remote sensing information and WheatSM model

小麦生长模型WheatSM是针对中国不同冬小麦品种类型研发的,已在科研和业务中得到应用。将遥感信息与作物生长模型耦合,进行区域大范围作物长势监测与估产具有重要的应用价值。以位于中国冬小麦主产区的河南省鹤壁市为研究区,分别利用SCE-UA优化同化法和EnKF同化法,将经优化重建的2013—2017年MODIS LAI时间序列数据和小麦生长模型WheatSM进行耦合,在站点和区域两种尺度开展冬小麦估产研究。结果表明:单站点运行时,在小麦模型参数严格标定的前提下,引入本身带有不确定性的遥感数据,并不能提高模型模拟结果的精度;遥感观测数据的质量对EnKF算法同化结果的影响大于对SCE-UA优化算法同化结果的影响;在区域尺度运行时,SCE-UA优化同化和EnKF同化模拟单产精度较同化前均有明显提高,分县模拟单产与统计单产相比,RMSE由2036.0 kg·hm^-2分别降低到1641.6 kg·hm^-2和1587.7 kg·hm^-2,分别降低19.4%和22.0%,且EnKF同化法的运行效率显著优于SCE-UA优化同化法。研究结果可为WheatSM小麦生长模型与遥感信息同化策略的选择提供依据。

WheatSM, a wheat growth model developed for different types of winter wheat in China, is applied in scientific research and public service. The coupling of remote sensing information with crop growth model has important application value in crop growth monitoring and yield estimation in large area. Hebi City in Henan Province is a main winter wheat producing area of China. With Hebi as the study area, the optimized reconstructed time series MODIS LAI data from 2013 to 2017 was coupled with WheatSM model with both SCE-UA optimal assimilation and EnKF assimilation methods to estimate the yield of winter wheat at both site and regional scales. The results showed that the introduction of remote sensing data with uncertainties did not improve the simulation accuracy of the crop model on the premise of strictly calibrating the parameters of WheatSM at site scale. The quality of remote sensing observation data had greater effects on the results of EnKF assimilation method than that of SCE-UA optimal assimilation method. At regional scale, the accuracy of data assimilation results with both SCE-UA and EnKF methods were higher than that without data assimilation. RMSE between simulated yield and statistical yield decreased from 2036.0 kg·hm^-2 to 1641 kg·hm^-2 with SCE-UA method and to 1587.7 kg·hm^-2 with EnKF method, with a reduction of 19.4% and 22.0%, respectively. The efficiency of EnKF assimilation method was higher than that of SCE-UA optimal assimilation method. Our results could provide a basis for the selection of data assimilation strategies coupling WheatSM with remote sensing data.