摘要
掌握森林病虫害的发生范围和危害程度,对于森林管理部门制定及时、有效的防治决策至关重要。本研究以2014年湖北省神农架林区华山松大小蠹(Dendroctonus armandi)灾害为背景,以野外调查数据、多光谱陆地资源卫星影像(Landsat)和数字高程模型(DEM)为基础数据源,结合最大熵(Max Ent)模型和迭代阈值分割算法,提出了适用于复杂林区的森林病虫害遥感监测方法(Max Ent-Segmentation),实现了神农架林区华山松大小蠹灾害空间分布范围和灾害程度的专题制图与精度评价。同时,为衡量所提出方法对于灾害程度评估的可靠性与准确度,本文还与传统光谱指数分析法进行了对比研究。结果表明:结合遥感光谱指数、海拔、坡度及有效太阳辐射等环境因子构建的Max Ent模型能够较为准确地监测华山松大小蠹灾害发生范围,受试者工作特征曲线下面积(AUC)值为0.938;当分类类型包括健康、轻度和重度时,Max Ent-Segmentation法分类精度最高达73.68%,明显高于传统光谱指数分析法(64.47%),表明该算法能够提高森林虫灾监测精度,适合用于植被类型多样、地形复杂林区的病虫害遥感监测。
Accurate spatial information on the location and extent of the forest pest infestation is important for the manager to take prompt and effective preventative measures. In 2014, Pinus armandii in the Shennongjia Forestry District was largely attacked by Dendroctonus armandi, a typical tree trunk-boring pest. In this study, we mapped the pest infestation based on the forest inventory data, Landsat images and DEM products. We proposed a novel method that employed a MaxEnt model and iteration threshold segmentation algorithm （MaxEnt-Segmentation） for this purpose. In order to evaluate reliability and accuracy of the proposed method, the traditional spectrum index analysis algorithm was also carried out and its performance was compared. The re- suits showed that the MaxEnt model was capable of accurately mapping the infested area using spectral indices, elevation, slope, potential solar radiation, with the AUC as high as 0.938. Max- Ent-Segmentation algorithm had higher overall classification accuracy （73.68%） compared witil the traditional spectral index algorithm （64.47%） when three classification classes （health, low- severity infestation, and high-severity infestation ） were included. The results suggest that this proposed algorithm can improve the accuracy of pest detection and is suitable for mapping forest pest infestation in areas with mixed forest stands and variable terrains.
出处
《生态学杂志》
CAS
CSCD
北大核心
2016年第8期2122-2131,共10页
Chinese Journal of Ecology
基金
国家自然科学基金优秀青年科学基金项目(41222004)资助
