Assimilating Remote Sensing Data Into Crop Growth Model By Using Ensemble Kalman Filter

Timely and accurate crop growth monitoring and yield prediction is significantly important to foodsecurity. Crop growth model and remote sensing as two means of crop yield estimation have theirrespective advantages and disadvantages. Combing crop growth model and remote sensing by dataassimilation has become one of the important development directions of quantitative remote andprecision farming.In order to improve the simulation accuracy of crop yield estimation in the area, this paperselected LAI as point and used ensemble Kalman filter to assimilate LAI simulated from crop growthmodel and LAI inverted form HJ satellite. In this paper, WOFOST model in potential production levelwas localized in the study area based on adjustment of crop and soil parameters; LAI were invertedfrom12images of HJ satellite by using ACRM; WOFOST-EnKF model on single point was developedto conduct the data assimilation study on single point; and on that basis, the regional study wasconducted through data regionalized. Main conclusion in this paper included：（1） This paper analysed the sensitivity of the crop parameters in WOFOST model usingobservation data in Raoyang to select14sensitive parameters, including initial dry weight, specific leafarea, maximum CO2assimilation rate, and then used FSEOPT optimization process to optimize thesensitive parameters; adjusted languish point humidity, field capacity and saturated soil water content byreferencing material; localized WOFOST model using the adjusted crop and soil parameters. The resultsshowed that the simulated LAI after the model localized was more close to the observation value, anddescribed growth process of winter wheat in Hengshui more accurately.（2） EFAST method was used to analyse the sensitivity of the parameters inACRM model toselect six sensitive parameters as free variables in the inversion process, including leaf area index,content of chlorophyll a/b, leaf structure parameter, and then used ACRM model to invert the LAI from12HJ satellite images of2008~2009winter wheat growth period. The inversion results describedaccurately the growth process of winter wheat in Hengshui area.（3） Establishment of WOFOST-EnKF model on the single point. WOFOST model was adjustedby inputting12images of HJ satellite data as external observations using ensemble Kalman filter, andconducted the data assimilation study in Raoyang, using observed growth period as validation. Theresults showed that the LAI curve after filter assimilated was more close to the observed LAI in thewhole growth period, and appeared a low value on March10,2009, which was meeting the situationthat LAI was decreasing because of the death of leaves during the winter.（4） Establishment of regional WOFOST-EnKF model. On the basis of the WOFOST-EnKFmodel on the single point, the regional WOFOST-EnKF model was established, and data assimilationstudy was conducted in the area through data regionalized, using the official statistics of yield in11countries and observed data from55field sample points as validation. The results show that theprediction precision of the yield after filter assimilated was significantly higher, the correlation coefficient raised from0.37to0.53; the relative error of the simulated growth periods after filterassimilated in were basically within5%, the average relative error of the flowering period was1.19%,the average relative error of the mature average was1.95%.