结合农学物候信息的玉米产量遥感估计方法研究

Real-time, accurate and large-scale crop yield estimation based on remote sensing technology is of great significance to the effective allocation of agricultural resources, grain market regulation and food security. The accumulation of crop grain yield is a complex physical process. The key for accurate estimation of crop yield is to clarify the growth stage (agronomic phenology) and yield accumulation process of specific crops. However, currently few researches focused on remote sensing based estimation of crop agronomic phenology and crop yield estimation that incorporated crop agronomic phenology. Based on the previous research on remote sensing based corn agronomic phenology estimation, this research aims to establish two corn yield estimation models: semi-physical semi-empirical model and physical processes based model, by combining crop agronomic phenological information and introducing crop water stress factors. Shijin Irrigation District in Hebei Province is selected as the study area. By combining the observations and statistical data, the estimation accuracy of corn yields of the two models will be verified and evaluated. Robustness of the two models will be analyzed with the uncertainty of remote sensing and meteorological data. In addition, the relationship between agronomic phenological information and crop yield, and its value in establishing yield estimation model will be analyzed, from both remote sensing macroscopic observation and mechanism model based simulation. This research is expected to provide theoretical and practical basis for modern agricultural management.

基于遥感技术实现实时、准确、大范围的作物产量估计对于农业资源配置、粮食市场调控乃至粮食安全具有重要意义。作物产量的形成与积累是一个复杂的物理过程，准确估产的关键在于明确具体作物的生长阶段（农学物候期）和产量积累过程，而目前针对作物农学物候期的遥感估计以及结合农学物候期的产量估计研究缺乏，也一直是一个难点。本课题拟基于申请人前期对玉米农学物候期遥感监测的研究基础上，以位于河北省的石津灌区为研究区域，结合作物农学物候信息，引入作物水分胁迫因子，建立半物理半经验的玉米估产模型和基于全物理过程数据同化的玉米估产模型；并结合观测和统计数据，验证两种模型玉米产量的估计精度，分析两种模型在遥感及气象数据具有不确定性条件下的鲁棒性；同时通过多源遥感宏观观测与机理模型微观模拟相结合的方式，分析农学物候期与产量的相互关系及其在建立产量估计模型中的价值，为现代农业管理提供理论和实践依据。

及时准确地掌握作物产量信息，对加强农业资源有效配置、农作物生产管理、粮食市场调控乃至保障我国粮食安全具有重要意义。作物产量的形成与积累是一个复杂的过程，涉及多个生理生化过程，受诸多因素的影响。理解产量积累过程的关键在于明确作物的生育阶段、不同生育阶段作物参数与产量的关系。本项目结合地面实验和区域统计数据，在构建空气温度遥感估计和高精度高时间分辨率时序数据重构模型的基础上，融合非完整生长周期遥感时序数据与物候响应函数，提出了一种生长季内玉米农学物候期遥感估计方法；结合作物的生长过程与机理，采用线性回归和深度学习等建模方法，提出了半物理半经验的玉米产量估计和预测模型，揭示并量化了遥感观测信息与作物产量的非线性关系，提高了作物产量的遥感估计精度；结合作物模型，构建了基于作物生长物理过程的区域数据同化估产模型，揭示了通过遥感估计的物候信息在作物模型模拟中的价值。研究成果可直接用于区域尺度玉米产量的实时监测与评估，对于精准农业发展具有重要的理论意义和应用价值。