| It is very vital for national food security to obtain in timely the agricultural information, such ascrop growing and yield. But the traditional approaches to survey and monitor agricultural informationon regional scales encounter some problems in term of timeliness, economic cost and accuracy. It isdifficult to show the change of space-time information of regional crop growing and can not meet theneeds for agriculture production management. It is feasible and available scheme to handle these issuesthrough assimilation of remote sensing information into crop growth model. But it is much later andscattered for the studies on data assimilation based on crop growth models. There are some crucialproblems, especially inversion of regional crop leaf area index, integration of crop growth model withdata assimilation algorithms, uncertainty and feasibility of data assimilation system in agriculturaloperations. Therefore, it is a much significant and valuable research work which will contribute toimprove the level of agricultural monitoring operation based on remote sensing information. In this study, the sensitivity and uncertainty analysis were conducted firstly in order to calibratecrop growth model CERES-Wheat model which was used to integrate with four algorithms includingensemble Kalman filter (EnKF), particle filter (PF), four dimension variation based on the properorthogonal decomposition technology, shuffled complex evolution (SCE). Remote sensing images wereemployed to invert regional winter wheat leaf area index which was assimilated into CERES-Wheat.The data assimilation schemes and accuracy were evaluated finally. The main contents and results asfollows. (1) Sensitivity and uncertainty analysis is an effective method to localize the model and select theoptimized variables with much uncertainty on regional scale. Some most sensitive parameters wereselected in order to localize CERES-Wheat model. Some regional parameters, such as the date andvolume of fertilization and irrigation, planting population, planting date and depth, are considered as theoptimized variables in data assimilation scheme. (2) An inversion scheme of canopy reflectance model ACRM was improved. The HJ CCD imageswere successfully used to invert winter wheat LAI in Hengshui. An acceptable result was obtainedsuccessfully using reflectance of winter wheat canopy measured on crop filed sites, with R2of0.88,relative error (RE) of8.45%, RMSE of0.31. On regional scale, the relationship between inverted LAIand measuered LAI were R2of0.77, RMSE of0.43, RE of17.43%. Comparing with current LAIproductions, the results were reasonable, and can be used to data assimilation studies. (3) Four schemes of data assimilation for CERES-Wheat model based on EnKF, PF, POD-4DVarand SCE-UA were designed to improve the simulating process of winter wheat growing, and outputrealistic yield successfully. For precision of yield estimation, the best is EnKF scheme, the worst isSCU-UA scheme. For efficiency of computation, the highest is POD-4DVar scheme, the lowest isSCE-UA scheme. A seriers of LAI of winter wheat measured field sites in Hengshui were assimilatedinto EnKF, PF, POD-4DVar and SCE-UA schemes. The results of yield estimation show that acceptable yields were obtained, with R2of0.57,0.51,0.52,0.22, RMSE of454kg hm-2,483kg hm-2,476kg hm-2,1281kg hm-2, RE of5.30%,5.60%,5.49%,15.24%correspondingly. It is feasible for fourschemes to be used on regional scale. (4) Four schemes were used successfully to estimate regional yield of winter wheat in Hengshui.The better results of yield estimation were obtained for EnKF and POD-4DVar schemes, the next is PFscheme secondly, SCE-UA scheme is the worst. The statistic results were RMSE of561kg·hm-2?753kg·hm-2,1123kg hm-2,2483kg hm-2, and RE of7.29%,10.37%,17.40%,41.16%correspondingly.Compromised between accuracy and efficiency, POD-4DVar scheme is the best choice on regional orglobal scale. (5) It is not significantly improved for regional yield estimation to increase resolution of remotesensing information assimilated. There has no significant difference of estimated yield error when fiveLAI data sets for resolutions1km,2km,4km,8km and16km, were assimilated respectively into fourschemes. The changing range of error is on average of0.36%~0.85%. The similar accuracy of winterwheat yield estimation was obtained when HJ LAI and MOD09A1LAI were assimilated into fourschemes. For operational application, LAI images need to be selected reasonablely for resolutions andsources, considering accuracy and compute price. In this study, potential application of three kinds of data assimilation methods were analyzed anddiscussed for regional crop yield estimation. Some essential problems include improvement of inversionapproach of LAI, integration of crop growth model with data assimilaton algorithms, analysis of yieldestimation and uncertainty. An optimal scheme for operation of data assimilation of crop growth modelis proposed. The results and conclusion will be reference for agricultural operation of data assimilationsystem in the future. |
PDF Full Text Request