基于机器学习和遥感数据的蒸散发成分分割模型研究

The dual effects of climate change and human activities have intensified the frequency, duration and intensity of extreme disasters such as drought, which poses more severe challenges to the optimal allocation and management of water resources. Evapotranspiration is an important process and key link of water cycle and energy cycle. Improving the evapotranspiration and component estimation accuracy has become one of the important contents of hydrometeorology. This project takes evapotranspiration as the research object, and uses machine learning method as the main research method to establish various models of meteorological variables and vegetation characteristic variables (from remote sensing data) and vegetation transpiration, focusing on the combination of different variables and modeling methods. The advantages and disadvantages, according to the specific circumstances to choose the optimal combination, combined with multi-source evapotranspiration and canopy interception data sets, and finally develop an empirical model of evapotranspiration based on machine learning and remote sensing information, to achieve high experimental requirements, single vegetation, the scale is small enough to use remote sensing data, suitable for multiple vegetation types and large-scale transformations. The research results of the project will provide valuable reference for water resources management, agricultural irrigation design and optimization, increase crop yield and drought monitoring and forecasting, and are expected to achieve certain innovations in international frontier research.

气候变化和人类活动的双重影响加剧了干旱等极端灾害发生频率、历时和强度，从而对水资源优化配置与管理提出了更为严峻的挑战。蒸散发是水循环和能量循环的重要过程和关键环节，提高蒸散发及成分估算精度已成为目前水文气象学的重要内容之一。本项目以蒸散发为研究对象，以机器学习方法为主要研究手段，建立植被特征变量（来源于遥感数据）与植被蒸腾的多种模型，重点研究不同变量组合和建模方式之间模拟植被蒸腾的准确性，并依照具体情况选择最优组合，耦合多源蒸散发和冠层截留蒸发数据集，最终开发基于机器学习和遥感信息的蒸散发分割经验模型，实现模型从实验要求高、适用植被单一、尺度较小到运用遥感数据、适用于多种植被类型和大尺度的转变。项目的研究成果将为水资源管理、农业灌溉设计与优化、提高作物产量和干旱监测与预报等提供有价值的参考，有望在国际前沿研究领域取得一定的创新。

蒸散发是水循环和能量循环的重要过程和关键环节，提高蒸散发及成分估算精度已成为目前水文气象学的重要内容之一。本项目以澜沧江流域为主要研究区域，金华江流域为辅助研究区域，开展基于机器学习和遥感数据的蒸散发成分分割研究，内容包括：（1）基于流域植被变化对蒸散发，特别是植被蒸腾具有非常重要的影响，分析澜沧江流域植被指数NDVI的时空变化特征；（2）评估多种蒸散发遥感产品在中国的适用性评估，为后续蒸散发成分分割奠定基础，比较多种蒸散发遥感产品改善分布式水文模型模拟效果；（4）构建三种基于机器学习方法的植被蒸腾模型，评估了不同输入变量组合的模拟情况，选用最优模型和最优变量组合，完成蒸散发成分的分割。研究成果丰富了蒸散发成分分割方法，对于区域水资源高效利用和科学管理具有重要的参考意义。