南方红壤侵蚀区林下水土流失区遥感判别研究

Understanding the spatial distribution of soil and water loss area under forest canopy could provide assistance information for targeted treatment of soil and water loss in the red soil erosion region of southern China. In the study, we firstly clarified the mechanism of "remote sensing estimation of stand net primary productivity → net primary productivity reflecting soil basal respiration → indicating understory coverage richness → indirect quantifying understory soil and water loss". Then, based on multi-angle measurement, better knowledge was obtained on the mechanism of "remote sensing canopy spectrum detection → forest quality difference → quantification of soil and water loss under forest". Finally, combined with topographic factors such as slope, from the perspective of "soil respiration" under the forest and "stand quality" on the ground, the mechanism of remote sensing quantification of soil and water loss under forest is expounded. Based on this finding, the discrimination of soil and water loss area under forest canopy in the red soil erosion region of southern China was achieved using remote sensing techniques.

受极端气候以及前期人为较为严重干扰的双重影响，当前的林下水土流现象仍十分常见！林下水土流失的基础数据及空间分布未能明确是影响这种现象存在的重要所在。为此，项目通过(1)量化、阐明遥感间接隐含体现林下覆盖物状况的机理，构建“遥感估测林分净初级生产力→净初级生产力反映土壤基础呼吸→指示林下覆盖物丰富度→间接量化林下水土流失” 的遥感间接估测机制；(2)基于无人机多角度遥感系统，研究多角度光谱信息特征与马尾松林分质量的内在表达机制，构建“遥感探测林分冠层光谱→体现林分质量差异→量化林下水土流失状况”的遥感量化机制；(3)结合坡度等地形因子，从“土壤呼吸”的林下、“林分质量”的地上两个角度，研究、阐明林下水土流失遥感量化机理，实现南方红壤区林下水土流失的量化，为水土流失靶向治理提供辅助，促进生态环境的“远看青山在，近听水淙淙”。

林下水土流失风险评估对于林下水土流失的灾害预警、治理以及灾后修复具有重要意义。建立林下水土流失风险评估体系和林下水土流失风险评估模型是林下水土流失风险评估的基础。然而，林下水土流失评估中未充分考虑森林垂直结构。本研究为进一步探讨森林垂直结构对林下水土流失风险的效应特点及机制，联合主被动遥感系统量化林下植被覆盖度的协同机制等，将林分参数加入到林下水土流失评价体系中，通过因子递归分析法，构建林下水土流失区风险评价体系；基于无人机近地面遥感系统反演评价因子，利用水利部现行的 SL190—2007《土壤侵蚀分类分级标准》将林下水土流失风险划分低风险、中风险与高风险三个等级，基于4种单一学习器与8种不同集成方式学习器分别建立林下水土流灾害风险评估模型，实现林下水土流失区的精准快速识别。结合实际调查的林下水土流失等级对各模型结果进行分类精度评价，实现对林下水土流失易发性风险程度的定量估算和动态监测。主要结果为：通过点云背向投影算法实现点云信息二维转变，将点云增强的空间信息赋予二维影像，完成主被动遥感系统联合，可以完整表征林下植被存续情况；林下水土流失风险的发生发展与林下植被覆盖度、植被覆盖度、坡度、郁闭度、地表起伏度、树高与坡位共7个指标密切相关 ；不同类型机器学习分类器中集成学习器的分类精度普遍比单一学习器高，其中以Bagging集成方法的分类器LightGBM的表现最佳，其总体精度为93.75%，Kappa系数为0.8832。本研究结果可为无人机近地面遥感对我国南方丘陵地区马尾松林林下水土流失风险评估提供技术支持。