基于遥感和作物生长模型的多尺度冬小麦估产研究

粮食安全是社会和谐、政治稳定和经济可持续发展的重要保障。准确预测区域乃至全球的农作物产量能够为各级政府、相关部门制定农业农村政策提供技术支持，保障粮食安全。目前关于农作物估产的研究大多具有地域性、经验性，过分依赖地面实测数据，一种基于多光谱卫星遥感数据和作物生长模型估算农作物产量的模型框架SCYM（Scalable Crop Yield Mapper）能够极大地减少模型对实测数据的依赖，快速应用于不同空间尺度、不同种类作物的估产，为多尺度农作物估产研究提供了一条有效的途径。以安徽省2012年-2018年冬小麦为研究对象，通过总结前人研究确定的敏感参数及其在研究区内的波动范围，结合大量实割实测数据优化WOFOST（WOrld FOod STudies）模型参数；将模拟产量、不同时段的模拟叶面积指数(LAI)同遴选出的天气变量训练随机森林模型，并以最佳观测日期组合下的MODIS-LAI代替对应时段的模拟LAI进行产量估算。结果表明：（1）模型产量估算值与站点实测值的总体相关性为0.758（R2为0.575），RMSE为790.92 kg·ha-1。精度较高的站点主要分布在淮北平原（<1%）而高误差区域集中于皖南丘陵地带（>40%）；（2）对2012年-2018年全省范围进行冬小麦估产，根据7年平均估产结果的空间分布，小麦单产由北向南逐渐减少，高值区出现在皖北的淮北平原，低值区主要分布于皖中、皖南地区；（3）2012年-2018年实测单产平均值为6 058.00 kg·ha-1，SCYM估算单产平均值为5 984.95 kg·ha-1，且估算产量与实测产量的年际时间序列的相关性为0.822，RMSE为189.96 kg·ha-1，每年估产的相对误差均不超过6%。研究表明SCYM估产框架对安徽省冬小麦产量估算具有一定的可行性，在产量预报方面效果良好。该方法能够在一定程度上改善以往估产模型存在的地域性、经验性问题，在区域尺度的应用方面具有极大的潜力，未来可为农业估产提供极其重要的理论依据和实用价值。

Research in Crop Yield Estimation Models on Different Scales Based on Remote Sensing and Crop Growth Model

Food security is a guarantee for social harmony, political stability and sustainable development of the economy. However, current research on crop yield estimation is mostly regional and empirical, relying too much on ground-measured data. Scalable Crop Yield Mapping (SCYM） is a satellite data based framework for estimating crop yield.It can be quickly applied to the estimated yield of different spatial scales and different types of crops without relying on measured data. This framework provides an important theoretical basis for multi-scale crop yield estimation research. We took the winter wheat of Anhui Province from 2012 to 2018 as the study object. Firstly, the sensitive parameters determined by the predecessors and their fluctuation ranges in the study area are summarized. Combined with a large amount of site data, the parameters optimization of the WOFOST model was completed. Secondly, random forest models were established based on the simulated yield, simulated leaf area index (LAI) at different periods, and selected meteorological indicators. Finally, the MODIS-LAI under the best observation date combination replaced the simulated LAI for the corresponding time periods to estimate the winter wheat yield in Anhui Province. The main outcomes in this study are as follows: (1) The overall correlation between the estimated outputs and the measured data of the stations is 0.758 (R2 is 0.575), and the RMSE is 790.92 kg·ha-1. The sites with higher production accuracy are mainly distributed in the Huaibei Plain (<1%), while the areas with high errors are concentrated in the hilly areas of southern Anhui (>40%). (2) The winter wheat yield in Anhui Province from 2012 to 2018 was estimated by SCYM. According to the spatial distribution of the 7-year average yield estimation, the yield is gradually decreasing from north to south. The high-value areas are located in the Huaibei Plain in northern Anhui, and the low-value areas are distributed in central Anhui and southern Anhui. (3) The average measured yield from 2012 to 2018 is 6 058.00 kg·ha-1, while the average yield of the SCYM is 5 984.95 kg·ha-1. The correlation between them in the interannual time series is 0.822, and the RMSE is 189.96 kg·ha-1. In seven years, the relative error each year does not exceed 6%. This study shows that the SCYM framework is feasible for estimating winter wheat yield in Anhui Province and has a good effect on yield forecast. This method can improve the regionality and empiricism of the previous crop yield estimation models to a certain extent. Meanwhile, it also solves the shortcomings of computationally intensive methods, which are costly and difficult to popularize. Thus, SCYM has great potential in applying of regional scales, and it will provide an extremely important theoretical basis and practical value for agricultural production in the future.