综合叶面积指数和条件植被温度指数的作物长势监测方法研究

Crop leaf area index (LAI) and water stress status are closely associated with crop growth and development and yield formation, and are two most important parameters for indicating crop growth condition and estimating crop yield. Based on the progress and findings of our former NSFC’s supported projects, this project focuses on the retrieval and assimilation of integrated crop growth monitoring indices during the main growth and development period (from the reviving stage to the dough stage) of winter wheat, using remotely sensed data and simulation results of the CERES-Wheat model. The main remotely sensed data used in this study are Terra/Aqua MODIS data, as well as data acquired by field campaigns. The study area is the Guanzhong Plain of Shaanxi province where winter wheat is commonly covered from middle October to middle June. On the basis of quantitative remote sensing techniques, data assimilation approaches and dynamic modelling methods, the following research activities will be carried out: (1) remotely sensed retrievals of a drought indicator called vegetation temperature condition index (VTCI) and LAI at ten-day intervals, and dynamically simulation of soil and crop (wheat) water balances and LAI; (2) developing crop growth monitoring indices at both a given stage and the main growth and development period of winter wheat by using subjective and objective weighting methods, and time series data analysis methods; (3) developing suitable assimilation approaches for the crop growth monitoring indices using the Ensemble Kalman filter, particle filter and variational assimilation methods; (4) field campaigns in the Guanzhong Plain for validating and evaluating the developed crop growth monitoring indices. The characteristics and innovation of this project are to develop integrated crop growth monitoring indices during the main growth and development period of wheat using retrieval and assimilation approaches. The achievements of this study will be very important and useful for crop growth monitoring at crop growth and development periods.

作物叶面积指数（LAI）和水分胁迫状况是表征作物长势和进行作物产量估测的2个最重要的参数。以陕西省关中平原为研究区域，以CERES-Wheat模型为动态模型，基于旬尺度的干旱监测结果—条件植被温度指数（VTCI）和LAI，以及动态模拟的土壤水分和LAI的结果，以即将研发的冬小麦长势监测指数与其产量间的相关性为依据，采用主、客观赋权法和时间序列数据分析法，研发某一生育时期、主要生育期（返青期至乳熟期）冬小麦长势综合监测指数的遥感定量反演研究方法；采用集合卡尔曼滤波、粒子滤波和变分同化等方法，研发适用于冬小麦长势综合监测指数的数据同化方法；开展地面调查实验，对相关反演与同化方法进行验证和评价。项目的特色和创新是基于前期主要生育期冬小麦水分胁迫信息反演与同化的研究成果，开展基于VTCI和LAI的主要生育期冬小麦长势综合监测指数的反演与同化研究，以期研发服务于区域农业生产的冬小麦长势遥感监测方法。

以关中平原为研究区域，利用定量遥感技术和动态模拟技术，采用时间序列分析和机器学习的方法重点对综合叶面积指数（LAI）和条件植被温度指数（VTCI）的作物长势监测模型及其时空尺度转换方法进行了研究，并以LAI和VTCI对冬小麦生产的影响为基准，对所研发的主要生育期（返青期至乳熟期）冬小麦长势综合监测指数和产量估测模型进行了解释。..在构建冬小麦长势综合监测指数与产量估测模型方面取得了突破。基于主成分分析-Copula函数（PCA-Copula）和小波分析（Morlet、Mexican Hat和Paul(m=4)）等时间序列方法研发的冬小麦长势综合监测指数和单产估测模型不仅精度较高，而且可解释性强，即研发的模型能很好地解释不同生育时期的VTCI和LAI对冬小麦生产的影响。发现对于VTCI，在拔节期对产量的影响最大，其次为抽穗—灌浆期，最后为返青期和乳熟期；对于LAI，在抽穗—灌浆期和乳熟期对产量的影响要大于在返青期和拔节期的影响。..双参数同化与长势综合监测相结合的方式提高了模型的精度。分别采用四维变分、集合卡尔曼滤波和粒子滤波等算法对LAI和VTCI进行了同化，并应用同化参数构建了长势综合监测指数与产量估测模型，发现应用同化双参数构建的模型精度均优于未同化参数构建的模型，且基于粒子滤波同化参数构建的模型的精度最高。..基于空间尺度转换和时空数据融合的长势表征参数反演取得了进一步的成果。基于MODIS数据产品和Sentinel系列卫星遥感数据，将时空数据融合模型与地表温度降尺度技术相结合，提出了田块尺度VTCI和LAI等参数的时间序列重建架构和方法，并生成了相关数据产品，为深入开展田块尺度作物长势综合监测奠定了坚实的基础。..基于机器学习的长势综合监测指数的智能提取取得了初步成果。将机器学习，尤其是深度学习的方法引入作物长势综合监测指数的提取和产量估测，得出了所构建的长势综合监测指数和产量估测模型的精度一般要优于项目组已研发的基于客观赋权法的模型的精度，并在研究中强化了长势综合监测指数提取模型的可解释性，能够解释不同生育时期的VTCI和LAI对冬小麦生长的影响。这些初步研究成果有助于项目组继续开展基于多参数和深度学习的作物长势信息智能提取研究，促进我们系列研究成果的持续深化和迈向智能化应用，对促进深度学习和遥感大数据技术在农业领域的应用具有重要意义。