基于无人机遥感数据与作物模型同化的冬油菜生长监测与估产方法研究

The threat of rapeseed production in ‘north drought with south flood’ and inefficiency growth monitoring in China is the important factors restricting the high yield harvest. Accurately and rapidly obtains winter rapeseed growth states, grain yield is great significance for field management, food and oil storage, and national economic development. Assimilating remote sensing data and crop model can complement and eliminate the advantages and drawbacks of both methods and eliminate drawbacks, which can accurately describe crop growth and improve crop yield estimation in a regional scale. Based on the different sowing dates and seeding density during three consecutive growing seasons of winter rapeseed (Brassica napus), our study will comprehensively analyze crop growth mechanism and process, using various equipment for rapeseed growth states monitoring at different scales (leaf, plant, and canopy) and time series(continuous growing season). By combining field observation with crop model numerical simulation, this study mainly focuses on the effect of detection model for sowing dates and growth stages in rapeseed and retrieving multiple states variables using UAV remote sensing spectral data; development and evaluation of multiple famous rapeseed growth mechanism model with atmosphere, crop and soil for simulating crop growth and production; establishing assimilation system by assimilating multiple state variables, parameters, and algorithm, analyzing the effect of different assimilation schemes on the accuracy of crop yield estimation, and proposing the optimum method for improving yield estimation accuracy. This project will provide an important theoretical basis for promoting the development of agricultural information application.

我国油菜生长常年所面临的“北旱南涝”威胁是制约油菜丰产保收的重要因素。准确、快速的获取油菜生长信息与产量预报对农田生产管理、保证粮油安全以及国家经济健康发展具有重要的意义。基于数据同化算法耦合的遥感与作物模型系统能够实现两者的优势互补，在准确描述作物生长状态的同时又能够提高作物区域估产精度。本课题拟以冬油菜为研究对象，油菜生长机理和过程机制模拟为基础，采用田间观测和模型数值模拟相结合的方法，开展不同尺度(叶片、单株和冠层)与时间序列的作物生长状态变量监测。从油菜播期、关键生育期以及多状态变量无人机遥感数据识别与反演、冬油菜生长过程机理模型研究和优化、以及多变量多参数多同化算法协同应用等方面，对同化系统的构建与实现进行系统研究。建立冬油菜产量估测同化框架，明确有效提高单产估测精度的同化方案，为促进农业信息监测研究的发展提供重要理论基础。

我国是世界油菜种植与生产大国，准确、快速的获取油菜长势信息，进行产量预报对农田生产管理、保证粮油安全和食用油供应具有重要的意义。本项目在陕西省西北农林科技大学节水灌溉试验站和江苏省扬州大学农水与水文生态试验场两种不同气候区进行田间试验，对不同播种时间、播种密度、灌溉策略下的冬油菜田间生长进行了监测，获取了3年的冬油菜生长指标、遥感影像、土壤和环境动态数据，研究了基于无人机遥感数据和作物模型的多变量油菜估产同化系统框架构建。系统研究了油菜播种日期和关键生育期的遥感识别与提取能力，提出了一种基于非对称高斯函数的改进形状模型方法(SMM)，其可以有效地估算油菜物候天数，RMSE最小为3.7天。系统开展了各种光谱分析方法对作物生长变量的估算与反演能力研究，结果表明人工神经网络的反演油菜FPAR的精度最优，其在苗期的估算效果最好，花期最差。研究了AquaCrop模型对不同播种和灌溉策略下冬油菜物候天数、冠层覆盖度、以及生物量等指标的解释能力，发现在中密度和高密度条件下油菜生物量和产量模拟达到了较高的模拟精度(RMSE: 0.8-2.1 t∙ha-1)。以上研究成果为进一步研究多尺度农田信息监测奠定了基础，具有十分重要的科学价值和生产实际意义。