天水市近30年林地动态变化遥感监测研究

目的]以甘肃省天水市为例,基于遥感影像变化监测技术,探讨黄土高原丘陵沟壑与小陇山-西秦岭山地交接过渡区域近30年来森林(林地)资源空间分布规律、时间变化趋势及变化影响因素。方法]以1988—2015年5期夏季Landsat TM/OLI遥感影像为主要数据源,结合辅助数据和外业实地样本点,以光谱特征和指数特征为特征变量,分别利用随机森林(RF)和参数优化支持向量机(POSVM)分类器对土地覆盖类型进行分类,然后基于分类后比较法进行森林资源动态变化监测。结果]分类结果表明,两种分类器的分类效果均较好,且随机森林分类器在分类精度、效率和稳定性方面明显优于参数优化支持向量机分类器。变化监测结果表明,近30年来森林资源总体变化趋势为林地面积先减少后增加。1990—1996年,林地面积减少0.74%;1996—2002年,林地面积减少2.74%;2002—2008年,林地面积增加1.06%;2008—2015年,林地面积增加8.89%。结论]本研究采用的基于非参数分类器分类后比较法的变化监测技术是复杂地形地貌过渡区森林资源动态变化监测的一种有效途径,在分类结果分析统计的基础上,得出研究区森林资源变化的总体趋势:以2002年(2002年影像)为界,林地总体趋势为先减少后增加,2002年后林地面积增加显著。

Research on Remote Sensing Monitoring Technology of Forest Land Dynamic Change in Tianshui in Recent 30 Years

Objective] Tianshui of Gansu province as a case study, the spatial distribution law, time changing trends and influence factors of forest resource had been researched in the transition region of typical Loess Plateau and Xiaolongshan-western Qinling Mountains in the past 30 years.Method] The main data sources are Landsat TM/OLI remote sensing images with 5 series in summer from 1988 to 2015, combined with auxiliary data and field survey data. Image spectral features and indices characteristics were selected as the input characteristic variables. The land cover types were classified based on the random forest classifier and the optimal parameter SVM classifier. Subsequently, the forest resources dynamic change monitoring was implemented by the post-classification comparison method.Result] The results show that the classification performance based on two classifiers are good, and the random forest classifier is better than that of optimal parameter SVM classifier, especially in the classification accuracy, algorithm efficiency and stability. The change detection results show that over the past 30 years the overall change trend of forest area was first decreased and then increased. From 1990 to 1996, the forest land area decreased by 0.74%, and from 1996 to 2002, forest land area decreased by 2.74%. However, forest land area increased by 1.06% from 2002 to 2008, and more significantly, forest land area increased by 8.89% from 2008 to 2015.Conclusion] The forest change detection method based on post-classification comparison of non-parametric classifiers classification result proposed in this paper is an effective approach for monitoring of forest resources dynamic change and information extraction in complex terrain landform transition region, which could provide valuable reference for quantitative analysis of vegetation change and comprehensive evaluation, reasonable spatial allocation and optimization adjustment of forest resources, forest management and assistant decision making and dynamic monitoring of forestry major project and ecological environment evaluation.