基于决策树与时序NDVI变化检测的耕地撂荒遥感监测——以四川省凉山州普格县为例

An Updated Method to Monitor the Changes in Spatial Distribution of Abandoned Land Based on Decision Tree and Time Series NDVI Change Detection:A Case Study of Puge County,Liangshan Prefecture,Sichuan Province,China

耕地撂荒遥感监测是推进撂荒地整治、保障粮食安全的常用手段。遥感影像解译易受到不同地物光谱信息的干扰而影响提取精度,并且难以区分轮休与撂荒。以四川省凉山彝族自治州普格县为研究区域,在分析该地区撂荒地光谱特征及农作物物候特征的基础上,采用基于决策树与时序NDVI变化检测耦合的方法,以Landsat系列数据作为数据源,对区域内2000—2019年的撂荒地进行提取和分析。结果表明:(1)从时间序列特征上看,研究区内撂荒地面积呈现“先增大后减小”的趋势,在2015年撂荒率达到最大值27.94%;空间上主要分布在河流两侧的耕地边缘,呈零散分布状态。(2)撂荒地的分布随高程和坡度的变化呈现一定的规律。在高程方面,撂荒地主要集中在1500~3000 m范围内,其面积占撂荒地总面积的85%左右;在坡度方面,10°~20°范围内撂荒地分布占比60%以上;另外,高程1000~3000 m范围及坡度0°~15°范围的撂荒面积、撂荒率均表现出“先增大后减小”的趋势,在2015年达到最大值,而到2019年则呈小幅下降,与研究区撂荒地总体变化趋势一致。经实地调查和无人机影像验证,本研究方法对撂荒地的提取精度分别达87.14%和90.00%,可靠性较高,可为我国西南山区撂荒地监测及乡村振兴战略实施等提供科技支撑。

Remote sensing monitoring of abandoned cultivated land is a common means to rectify abandoned land and ensure food security. Remote sensing image interpretation is easily interfered by spectral information of different ground objects by lowering extraction accuracy, and it is difficult to distinguish between cultivated land fallow and waster land. In this study, it took Puge county, Liangshan prefecture, Sichuan province, China as research case. Based on the analysis of spectral characteristics of abandoned land and crop phenology characteristics in this area, Landsat series images were used as data sources to extract and analyze the abandoned land from 2000 to 2019 by coupling decision tree and temporal NDVI change detection method. The results found:(1) From the time series characteristics, the area of abandoned land in the study area showed a trend of first increasing and then decreasing, and the increase rate of abandoned land reached the maximum in 2015, with a value of 27.94%;The abandoned land was mainly distributed at the edge of cultivated land on both sides of rivers, characterized by scattered pattern.(2) The distribution of abandoned land had certain regularity with the change of elevation and slope. In terms of elevation, abandoned land was mainly concentrated in the altitude range of 1500~3000 m, accounting for about 85% of the total abandoned land area. In terms of slope, abandoned land was the largest in the range of 10°~20°, accounting for more than 60% of the total abandoned land area. In addition, the abandoned land area and the abandonment rate increased first and then decreased in the elevation range of 1000~3000 m and in the slope range of 0°~15°, reaching the maximum in 2015 and presenting a slight downward trend in 2019. And it showed that the trend of the abandonment rate in these areas was consistent with that of the whole study area. According to our field investigation and UAV image verification, the extraction accuracy of abandoned land by our proposed method was 87.14% and 90.00%, respectively, with high reliability;Therefore, it can provide scientific and technological support for the monitoring of abandoned land in mountainous areas of southwest China and the implementation of rural revitalization strategy.