欢迎访问林业科学,今天是

林业科学 ›› 2020, Vol. 56 ›› Issue (5): 80-88.doi: 10.11707/j.1001-7488.20200509

• 论文与研究报告 • 上一篇    下一篇

基于Landsat时序数据的森林干扰监测

钟莉,陈芸芝*,汪小钦   

  1. 福州大学空间数据挖掘与信息共享教育部重点实验室 卫星空间信息技术综合应用国家地方联合工程研究中心 数字中国研究院(福建) 福州 350108
  • 收稿日期:2017-12-07 出版日期:2020-05-25 发布日期:2020-06-13
  • 通讯作者: 陈芸芝
  • 基金资助:
    国家重点研发计划课题(2017YFB0504203);国家自然科学基金项目(41401488);中央引导地方科技发展专项(2017L3012)

Forest Disturbance Monitoring Based on Time Series of Landsat Data

Li Zhong,Yunzhi Chen*,Xiaoqin Wang   

  1. Key Laboratory of Spatial Data Mining & Information Sharing of Ministry of Education National & Local Joint Engineering Research Center of Satellite Geospatial Information Technology The Academy of Digital China(Fujian) Fuzhou University Fuzhou 350108
  • Received:2017-12-07 Online:2020-05-25 Published:2020-06-13
  • Contact: Yunzhi Chen

摘要:

目的: 针对我国森林干扰频繁、干扰类型复杂多样的特点,采用时间序列轨迹分析方法进行森林干扰监测,为陆地森林生态系统碳循环和碳蓄积以及气候变化研究提供参考。方法: 以福建省长汀县为研究区,基于2000—2016年15期Landsat时序数据,采用时间序列轨迹分析(LandTrendr)方法,对时序轨迹进行适当分段和线性拟合以识别干扰变化,获取长汀县森林干扰信息。结合Google影像和全球30 m分辨率GDEM数字高程产品等辅助信息分析监测结果,并基于实地调查和像元2种方法进行精度评估和验证。结果: 2000—2016年长汀县森林干扰总面积192.49 km2,平均每年受扰动森林12.83 km2;2001年干扰量最小,不足1 km2;2004、2008和2009年森林受干扰较为严重,均在30 km2以上,约占当年森林面积的1.3%,共占干扰总面积的50%,其中2004年干扰面积高达32.85 km2;2003、2006、2007和2010—2011年干扰面积略大于10 km2,但均小于当年森林面积的0.6%,其余各年干扰面积远小于10 km2。森林干扰面积在个别年份波动较大,总体上随时间呈下降趋势。森林干扰持续时间主要为1~3年,发生在1年的干扰面积比例最大,达82%;森林干扰主要集中在长汀县东部非森林区域附近,随海拔升高干扰呈明显下降趋势,超过60%的干扰发生在中低海拔地区。结合Google影像目视解译,长汀县森林干扰主要是由森林火灾和人工砍伐造成的急剧干扰事件,且主要发生在非森林区域附近的低海拔地区。结论: 基于目视解译和实地调查结果与研究得到的干扰监测结果一致,干扰斑块可被完整提出,且边界准确清晰,细小干扰也能逐一识别。基于像元尺度精度验证的总体精度达96.26%,Kappa系数为0.92,各年份用户精度均在80%以上,除个别年份外,生产者精度均在75%以上,具有较高的监测精度。

关键词: 森林干扰, Landsat数据, 时间序列, LandTrendr

Abstract:

Objective: Considering the character of forest disturbance in China is frequent and complicated, we use the series trajectory analysis method to monitor forest disturbance which can provide references for carbon cycle and carbon accumulation of terrestrial forest ecosystems and climate change research. It is of important significance to the terrestrial carbon cycle and climate change research. Method: We conduct a study in Changting county base on 15 Landsat time-series data from 2000 to 2016 by adopting a time-series trajectory analysis technique(LandTrendr)to identify forest disturbance which apply segment properly and fit linearly a time sequence on the time-series trajectory. The results are verified based on field survey and pixel of interference by using the Google image and global 30 m resolution of GDEM digital elevation products. Result: The total area of forest disturbance is 192.49 km2 and the mean area is 12.83 km2 within the area from 2000 to 2016, of which that in 2001 is the minimum and less than l km2.The most serious disturbance occurred in 2004, 2008 and 2009. The area all over 30 km2 and occupy about 1.3% of forest area in those years. In the three years, the total area represent 50% of the forest disturbed area and the biggest disturbance area is 32.85 km2 in 2004. In 2003, 2006, 2007 and 2010-2011, the disturbance area was slightly larger than 10 km2, but less 0.6% of forest area in this year. It's much less than 10 km2 for remaining years. The area of forest disturbance fluctuates greatly in individual years, but it is decreasing over time. The duration of forest disturbance is mainly about 1 to 3 years and the largest disturbance area occurred in 1 year which up to 82%. Disturbance is mainly concentrated in the east of Changting where is near the non-forest area. It's obvious that disturbance area is declined with the elevation and more than 60% of disturbance occurred at low-middle altitudes. Combined with visual interpretation of Google images, it is shown that the forest disturbance in Changting is mainly caused by forest fires and artificial deforestation, which mainly occurs in low altitude areas near the non-forest areas. Conclusion: The results of field survey and pixel of interference are consistent with the results of monitoring. Disturbance patch can be extract completely and the boundary is accurate and distinct, so is the fine disturbance. Base on pixel scale accuracy verification to monitor forest disturbance, it is showed that the overall accuracy reaches 96.26%, with a Kappa coefficient of 0.92. In all years, the accuracy of users is over 80%., and the accuracy of producers is more than 75% except several years. The results require a high monitoring accuracy, indicating a significant potential of the technique for forest disturbance monitoring.

Key words: forest disturbance, Landsat data, time series, LandTrendr

中图分类号: