【摘要】 提出了一个基于多源遥感数据和随机森林算法的塑料大棚空间分布特征提取方法，并以黑龙江省哈尔滨市中心城区为研究区，进行方法的应用验证。基于Google Earth Engine平台上的Sentinel-1/2影像和SRTM数字高程模型，结合简单非迭代聚类(SNIC)算法来识别空间集群，构建随机森林分类器，对哈尔滨市9个城区2024年的塑料大棚进行了遥感提取和空间格局分析，探讨光谱指数可分性以及不同特征组合对塑料大棚提取精度的影响。结果表明：通过多特征融合的方法，选取6种光谱特征参数可以实现塑料大棚与其他土地覆盖类型的精确区分。其中：光谱、极化、地形、纹理特征组合的塑料大棚提取效果最优，生产者精度和用户精度分别为96.4%、87.1%。基于此特征，随机森林分类总体精度为95.3%, Kappa系数为93.6%,满足实际工作需求。2024年哈尔滨市9区塑料大棚面积为175.03 km~2,占研究区总面积的1.7%,双城区和呼兰区的塑料大棚面积显著高于其他区域，2区合计占比达研究区塑料大棚总面积的70%以上。核密度分析、最邻近分析发现研究区塑料大棚呈现显著集聚分布(最邻近指数R=0.294<1)。塑料大棚在空间上形成“多核”聚集格局，具体表现为以双城区西部、研究区中部(松北区东南部、呼兰区南部)为中心的高密度核心区，呈现“核心—边缘”的空间结构，并针对塑料大棚的空间分布特征对其管理措施提出建议。更多还原

【Abstract】 This study proposes a method for extracting the spatial distribution characteristics of plastic greenhouses based on multi-source remote sensing data and a random forest algorithm. The method was applied and validated in the central urban area of Harbin City, Heilongjiang Province. Using Sentinel-1/2 imagery and the SRTM Digital Elevation Model available on the Google Earth Engine platform, and combining them with the Simple Non-Iterative Clustering algorithm to identify spatial clusters, a random forest classifier was constructed to extract and analyze the spatial pattern of plastic greenhouses in nine districts of Harbin in 2024. The study also explored the spectral separability and the effects of different feature combinations on extraction accuracy. The results show that through multi-feature fusion, the use of six spectral feature parameters enables accurate differentiation of plastic greenhouses from other land cover types. Among these, the combination of spectral, polarization, topographic, and texture features yielded the best extraction performance, with producer and user accuracies of 96.4% and 87.1%, respectively. Based on these features, the overall classification accuracy of the random forest model reached 95.3%, with a Kappa coefficient of 93.6%, meeting the requirements of practical applications.In 2024, the total area of plastic greenhouses in the nine districts of Harbin was 175.03 km~2, accounting for 1.7% of the total study area. The areas of plastic greenhouses in Shuangcheng and Hulan districts were significantly higher than in other districts, together comprising more than 70% of the total greenhouse area in the study region. Kernel density and nearest neighbor analyses revealed that plastic greenhouses in the study area exhibit a significant clustering distribution pattern(R=0.294 < 1). Spatially, plastic greenhouses form a “multi-core” clustered pattern, with high-density core areas centered in the western part of Shuangcheng District and the central area of the study region(southeastern Songbei District and southern Hulan District), displaying a “core-periphery” spatial structure. Based on the spatial distribution characteristics of plastic greenhouses, the study also proposes management recommendations.更多还原