基于优化k-NN模型的高山松地上生物量遥感估测

针对传统k-最近邻法（k-nearest neighbor，k-NN）在搜索最近邻单元时赋予特征变量相等的权重，缺少对特征变量加权优化等不足问题，在云南省香格里拉市，以高山松Pinus densata为研究对象，基于49块实测标准地，116株高山松样木和Landsat 8/OLI影像，在前期进行基于遗传算法（genetic algorithm，GA）优化的k-NN模型实现的基础上，对k-NN的3个参数（k，t和d）进行反复测试优化组合，在像元尺度上对研究区高山松地上生物量进行遥感估算。结果表明:基于遗传算法优化的k-NN模型精度优于传统的k-NN模型，优化前均方根误差为30.0 t·hm-2，偏差为-0.418 t·hm-2，相对标准误差百分比（R_MSE）为54.8%；优化后均方根误差为24.0 t·hm-2，偏差为-0.123 t·hm-2，R_MSE为43.7%。基于优化k-NN模型的研究区高山松地上生物量总储量估测结果为0.89×107 t。

Optimizing the k-nearest neighbors technique for estimating Pinus densata aboveground biomass based on remote sensing

For the traditional k-nearest neighbor (k-NN), there are insufficient problems that give the weight of the feature variables equally when searching the nearest neighbor population units and a lack of weight vectors for the feature variables. In this study, Shangri-la City, Yunnan Province, was selected as the research area, and Pinus densata was taken as the research object. Based on 49 field data plots, 116 P. densata data samples, and Landsat 8/Operational Land Imager (OLI) imaging, a genetic algorithm was used to optimize the k-nearest neighbor model in the early stages, and the aboveground biomass of P. densata in the study area was estimated at the pixel scale after the k-NN three parameters (k, t, and d) were repeatedly tested and optimized. Results showed that accuracy of the k-NN model optimized by a genetic algorithm was better than the traditional k-NN model. Before optimization, the root mean square error was 30.0 t·hm-2, deviation was -0.418 t·hm-2, and R_MSE was 54.8%; after optimization, the root mean square error was 24.0 t·hm-2, deviation was -0.123 t·hm-2, and R_MSE was 43.7%. Finally, the estimated total aboveground biomass of P. densata in the study area was 0.89×107 t based on the optimized k-NN model.