基于遥感与作物模型同化的区域棉花产量对灌溉量的响应机制研究

The shortage of water resources in the arid zones of northwest China affects the growth process of cotton while the rational irrigation scheme can ensure cotton production. Currently, the response mechanism of the cotton growth process and yield formation on the regional irrigation amount is not clear, which affects the implementation of large-scale precision agriculture and the allocation of regional water resources. Therefore, this project aims to study the mechanism of irrigation amount on cotton growth process and yield by employing assimilation of remote sensing data and crop growth models at the regional scale. Based on multi-year field observations, the crop growth model was calibrated to accurately simulate the cotton growth process and yield at the field scale. Regional leaf area index (LAI) derived from Landsat-8 and Sentinel-2 optical remote sensing data reveals spatial and temporal features of cotton growth process. During the vegetative and reproductive growth period of cotton, the assimilation models coupling multi-source remote sensing data and crop growth model were respectively employed to estimate regional cotton yield by adjusting sowing density and irrigation amount in the crop model. Both sowing density and irrigation amount are critical factors affecting LAI dynamics of cotton. Then the response mechanism of regional cotton yield to irrigation amount was elaborated based on the assimilation model. Moreover, the spatial relationship between them was also revealed based on spatial analysis methods. Research findings can propose a suitable irrigation scheme for regional cotton growth management and also provide scientific evidence for the regional water resources allocation.

西北干旱区水资源短缺，影响到棉花的生长过程，合理灌溉则是保证棉花产量的关键措施之一。目前区域尺度棉花生长过程及产量对灌溉量的响应机制尚不明确，影响到精准农业的实施及区域水资源的配置。因此，本项目利用遥感数据与作物生长模型同化的方法，研究区域尺度灌溉量对棉花生长过程及产量的影响机制。基于多年的田间观测数据，校正作物生长模型，准确模拟田块尺度棉花生长过程及产量；利用Landsat-8与Sentinel-2光学遥感数据反演区域尺度棉花冠层叶面积指数（LAI），揭示棉花冠层生长过程的时空特征；在棉花营养生长及生殖生长阶段内，通过调整影响LAI变化的两个关键农艺参数：播种密度与灌溉量，分阶段同化多源遥感数据与作物模型，估算区域尺度棉花产量；通过同化模型与空间分析方法，阐述区域尺度棉花产量对灌溉量的响应机制，并揭示其空间关系，探索区域尺度适宜的灌溉模式，为精准农业发展及区域水资源分配提供科学依据。

在区域尺度上厘清灌溉等农艺管理影响棉花产量的时空差异，有利于干旱区棉花精准管理。同化遥感数据与作物生长模型，是促进区域作物估产及农艺管理模拟的关键技术。本项目以区域棉花估产及灌溉模拟为研究目标，选择新疆阿拉尔棉花垦区为试验区,以作物生长模型来模拟棉花叶面积指数LAI动态及估算产量；同时，利用覆盖棉花完整物候期的时序遥感数据反演LAI，采用粒子群优化等同化算法将遥感反演的LAI数据在营养生长及生殖生长两个阶段内来优化作物模型参数及更新状态变量，实现区域棉花产量估算及农艺管理。主要研究成果如下：（1）实现作物模型参数等对棉花产量及LAI的敏感性分析，并选择关键参数，并基于连续多年实测数据，实现作物模型参数本地化来适用于研究区棉花产量估算。（2）基于2014–2018年遥感数据生成时序EVI数据并通过对其进行平滑，构建棉花等作物EVI时间序列曲线，基于时序EVI及作物物候信息选择最优特征波段，利用随机森林分类算法获取逐年棉花等作物类型图。（3）基于棉花完整生长季内的时序遥感数据获取植被指数，与同步观测的覆盖研究区多个棉田的LAI实测数据，建立线性回归模型，并验证该遥感反演模型的准确度，基于该模型及时序遥感影像并耦合遥感识别的棉花种植格局数据，来获取区域棉田的生长季内LAI时序数据。（4）利用粒子群优化等同化算法同化遥感数据与AquaCrop模型模拟2014–2018年棉花产量，发现同化模型模拟产量与实测产量接近（RMSE=0.08 ton/ha）。另外，同化遥感与APSIM模型产量模拟值也接近于实测值，表明不同作物模型与遥感同化均能模拟精准的产量。分别在棉花的营养生长、生殖生长两个阶段内，根据 LAI 发展及产量形成的机理差异，构建棉花生育期内的作物模型与光学遥感的同化模型，实现区域尺度棉花产量估算，并发现不同灌溉量及尿素施肥量的组合对产量的影响有所差异。