同化叶面积指数和蒸散发双变量的冬小麦产量估测方法

Research on Winter Wheat Yield Estimation Based on Assimilation of Leaf Area Index and Evapotranspiration Data

同化遥感信息到作物生长过程模拟模型,是估测区域作物产量的重要方法之一。同化变量的选取对同化结果精度至关重要。本文在标定WOFOST作物模型参数的基础上,优化了WOFOST模型的默认灌溉参数。利用ET和LAI作为同化变量,分别构建了时间序列趋势信息的代价函数和四维变分代价函数;采用SCE-UA算法最小化代价函数,重新初始化WOFOST模型初始参数——作物初始干物质重、作物35℃生命期和灌溉量。最后利用MODIS LAI产品(MCD15A3)、MODIS ET产品(MOD16A2),同化到作物模型估测产量,并对比分析了水分胁迫模式下同化单变量(ET或LAI)和同化双变量(ET和LAI)的估产精度。结果表明:同化双变量ET和LAI的策略,优于同化单变量LAI或ET,双变量策略的冬小麦产量估测精度为R2=0.432,RMSE=721 kg/hm2;单独同化高精度LAI对提高估产精度具有重要作用,其冬小麦产量估测精度为R2=0.408,RMSE=925 kg/hm2;单独同化ET的趋势信息改善了WOFOST模型模拟水分平衡的参数,但是,产量估测精度(R2=0.013,RMSE=1134 kg/hm2)与模型模拟估测产量精度(R2=0.006,RMSE=1210 kg/hm2)相比改善效果有限。本研究为其他区域的遥感数据与作物模型的双变量数据同化的作物产量估测研究提供了参考价值。

Assimilating remote sensing information into crop growth model is an important approachto estimate regional crop yield. The assimilation algorithm and corresponding assimilation variables are the keys of the as- similation system, which greatly impact the accuracy of assimilation results. In thepaper, the default irrigation pa- rameters of WOFOST were optimized firstly with the help of calibrating WOFOST crop model parameters. Then, ET data was chosen as the assimilation variable to build the cost function of time series trends using MO- DIS ET products （MOD16A2） and WOFOST simulation. And LAI data was assimilatedwith the cost function of four dimensional variational data assimilation method using MODIS LAI （MCD15A3） products and WOFOST simulation. Furthermore, parameters including the crop initial dry matter（TDWI）, the lifetime of crop in 35℃ （SPAN） and the irrigation（RIRR） were optimizedcontinuously by SCE--UA algorithm, which would stop run- ning the program when the cost function isoptimal. The estimatedcrop yield results were obtained usingfour methods comparatively under water limited mode, including the method thm2t does not assimilate and methods thm2t assimilateET, assimilateLAI and assimilate both ET and LAI.We address the assimilation of double vari- ables to be the methodthm2t ET and LAI arebothm2ssimilated. Finally, the accuracies of yield estimation by as-similating double variables and a single variable under water limited mode were compared and analyzed. The re- suits indicated thm2t the method of assimilating double variables was better than assimilating a single variable, which got the highest accuracy （R2=0.432, RMSE=721 kg/hm2）. The method of assimilating high precision LAI- significantly improved the accuracy of yield estimation （R2=0.408, RMSE=925 kg/hm2）. The method of assimilat- ing ET demonstratedbetter performance when the WOFOST model simulates the water balance during crop growing period, but had a limited impacton improving the accuracy of yield estimation （R2=0.013, RMSE=1134 kg/hm2） compared with model simulation （R2=0.006, RMSE=1210 kg/hm2）. This research provided a reference for studies in other areas on predicting crop production at regional scale thm2t based on assimilating double vari- ables.