基于更新和同化策略相结合的遥感信息与水稻生长模型耦合技术的研究

将遥感与作物模型耦合有利于提高作物模型在区域尺度应用时的精度。基于集合平方根滤波算法(Ensemble Square RootFilter,EnSRF)和粒子群优化算法(Particle Swarm Optimization,PSO),以叶面积指数(Leaf Area Index,LAI)和叶片氮积累量(Leaf Nitrogen Accumulation,LNA)共同作为同化耦合点和过程更新点,将同化与更新策略相结合,研究建立了基于遥感信息与水稻生长模型(RiceGrow)耦合的水稻生长与产量预测技术。结果表明,将更新和同化策略结合后,利用RiceGrow模型模拟的水稻生长指标和产量结果更接近于实测值。其中LAI、LNA和产量与实测值间的RMSE分别为0.94、0.47 g/m2和320.15 kg/hm2;RiceGrow模型直接模拟LAI、LNA和产量的RMSE为1.25、1.24 g/m2和516.83 kg/hm2;而单纯基于同化策略模拟LAI、LNA和产量的RMSE为1.01、0.59 g/m2和335.70 kg/hm2。此外,基于该技术的模型区域尺度预测结果能较好地描述水稻生长和产量的时空分布状况,生长指标及区域总产量的模拟相对误差均小于20%。显示基于更新和同化策略相结合的遥感与模型耦合技术具有较高的预测精度,从而为区域尺度作物生长和产量预测提供了技术支撑。

Coupling remotely sensed information with a rice growth model by combining updating and assimilation strategies

Coupling remote sensing (RS) with a crop growth model can improve the prediction accuracy of crop modeling at a regional scale. In this paper, a new coupling method was developed based on a combination of the updating and assimilation strategies. The optimized model was used to estimate rice grain yield at both the field and regional scales. Firstly, parameterization for regionalization of the integrated RiceGrow model was accomplished with the use of the Particle Swarm Optimization (PSO) optimization algorithm. Management parameters included sowing date, sowing rate and nitrogen rate. Then, analyzed values of model variables, leaf area index (LAI) and leaf nitrogen accumulation (LNA), which simultaneously served as the assimilation and updating parameters, were calculated based on the Ensemble Square Root Filter (EnSRF) and used to update the corresponding values simulated by the RiceGrow model. Finally, the growth status and final yield were simulated by the integrated model. This integrated technique was tested on independent datasets acquired from three rice field tests in different years for different rice varieties and at different treatments with regards to nitrogen rates and sowing densities. This was in addition to data obtained from study areas in Yizheng and Rugao counties in Jiangsu Province (in central eastern China), both of which are main production areas of high-quality rice in China. The test results showed that simulated values based on the integrated model were closer to the measured values than those simulated directly by the RiceGrow model, which produced RMSE values of 0.94 for LAI, 0.47 g/m2 for LNA and 320.15 kg/hm2 for grain yield. The compared to RMSE values of 1.25, 1.24 g/m2 and 516.83 kg/hm2 for these respective parameters based on the RiceGrow model alone, and 1.01, 0.59 g/m2 and 335.70 kg/hm2 for the RiceGrow model based on the assimilation strategy. The newly developed integrated technique also performed well at a regional scale and the predicted results were consistent with the temporal and spatial distribution of rice growth status and grain yield, with relative error (RE) values of <20% for both growth parameters and the grain yield. This error may have been due to the limited simulation ability of the RiceGrow model, or generated during the RS information extraction and the statistical RS estimating models, all of which need improvement. These results indicated that there are certain non-determinacy factors for the RiceGrow model when used at the regional scale, such as spatial variability in the soil and management parameters. However, the integrated technique based on combining RS and the RiceGrow model could reduce this problem. Therefore, this study provides an important step towards the more routine use of combined RS and crop modeling techniques to improve our ability to estimate regional rice grain yield predictions.