Pine wood nematode (PWN) disease is one of the major disasters in forests of southern China, causing substantial forest resources and ecological and economic losses. Based on field surveys and WFV image data from the GF-1 satellite, we constructed a spatial identification model of PWN disease with the random forest model to explore the relative influences of topography, human activities and stand factors on the occurrence of diseases and predict their spatial distribution. We then used the spatial autocorrelation analysis to assess the distribution characteristics of PWN disease at the regional scale. The results showed that the random forest model constructed in this study was effective in identifying pine nematode diseases (AUC value=0.99, overall accuracy=0.96). The norma-lized difference greenness index (NDGI), the distance to the highway, and normalized vegetation index (NDVI) were important factors in explaining the spatial variations of PWN disease occurrence. There was a positive spatial correlation in the occurrence of PWN disease (not randomly distributed but with obvious spatial aggregation characteristics). The high occurrence areas of pine wood nematode disease concentrated in Chitu Township, Zhufang Township and Shibatang Township, low occurrence areas concentrated in the vicinity of Rongjiang Street. The areas far away from the highway, low in elevation, and close to county roads were suffered to PWN disease. The results could serve the regional monitoring of pine nematode disease occurrence and provide practical guidance for PWN disease management.
松材线虫病是我国南方森林面临的主要灾害之一。本文基于野外调查和高分一号(GF-1)卫星WFV影像数据,采用随机森林模型构建松材线虫病空间识别模型,探究地形、人类活动和林分因子等对病害发生的影响,监测病害空间分布,并采用空间自相关性分析评估江西省赣州市南康区松材线虫病发生特征。结果表明: 构建模型对松材线虫病的识别效果良好(AUC值=0.99,总体精度=0.96),可以实现对区域松材线虫病空间分布情况的有效监测;归一化差异绿度指数(NDGI)、距高速公路的距离、归一化植被指数(NDVI)是重要的建模因子;空间自相关性分析表明,松材线虫病的发生存在明显的空间正相关性即空间聚集性特征;南康区松材线虫病高发生区集中于赤土乡、朱坊镇和十八塘乡,低发生区集中于蓉江街道附近;分析变量的边际效应发现,离高速公路远、离县道近的低海拔地段是松材线虫病易发区域。研究结果可服务于区域松材线虫病分布的快速监测,对该病害防治和管理具有一定的指导意义。.
Keywords: GF-1 WFV; monitoring model; multispectral data; pine wood nematode disease; random forest.