基于MODIS/NDVI时间序列的森林灾害快速评估方法——以贵州省为例

A rapid assessment method for forest disaster based on MODIS/NDVI time series:a case study from Guizhou Province

以2005—2008年中分辨率成像光谱仪(MODIS)的归一化植被指数(NDVI)数据集为主要数据源构建研究区时间序列,结合统计学分析方法设计一套快速有效的森林灾害评估体系,用以探测森林资源损失的空间分布范围与灾害等级,并利用森林资源雪凝灾害损失的实际调查数据对评估结果进行一致性检验。结果表明:通过计算2005—2007年研究区所有森林像元的平均值R珔2005-2007和平均绝对偏差δall,确定森林灾害阈值DT为0.048;利用该阈值,获取2008年受灾较重的森林空间分布,主要密集分布在贵州省东南部和东北部,包括了黔南自治州、黔东南自治州和铜仁市等地区;受灾森林像元数占总森林像元数的28.6%,高于凝冻灾害森林资源损失实际调查结果(17.7%);在县域尺度上,根据MODIS/NDVI关键期影像获取的森林资源损失区域和灾害等级,确定德江、沿河和都匀等11个重度受灾县(市、区)和湄潭、榕江和桐梓等10个中度受灾县(市、区),与小班调查结果具有较高的一致性,其Kappa系数为0.86。方法为大区域尺度的森林灾害快速评估提供了一种新思路。

Snow disasters are one of the major natural disturbances to forest ecosystems in China. The increased frequency and severity of forest disturbances in recent years requires rapid and accurate regional forest damage assessment to support post-disturbance forest management, hazardous fuel management, post-hazard relief activities, and government compensation claims. Interpretation of MODIS/NDVIs to construct Guizhou Province＇s remote sensing images time series between 2005 and 2008 was involved in this analysis. Specifically, the mean value of usefulness index was applied to identify key MODIS/NDVI images and the Savitzky-Golay filter was used to reconstruct key images first. Then, the ratio of forest pixels NDVI value （R2005-2007） was computed before and after the corresponding period of snow disaster in 2008, the respective mean and mean absolute of R2005-2007 were derived to determine undisturbed forest pixels and forest damage threshold （DT）. Ultimately, with the support of remote sensing and geographical information system, disaster areas and damage ranks were identified by using this method. Light, moderate and severe damage were classified in the light of the R2007-2008 of forest damage pixels at county level. Further, in conjunction with field surveys of forest resources implemented by the Department of Forestry of Guizhou Province, the consistency between our derivations and field surveys was assessed. The preliminary results were as follows： （1） With the aid of usefulness index, MODIS/NDVI products captured on September 30（September 30 2005, September 30 2006 and September 30 2007） and May 9（May 9 2006, May 9 2007 and May 9 2008） were respectively chosen as key images corresponding to the period of the snow disasters, which had the lower mean value of usefulness index （2.06 and 2.41, respectively） form 2005 to 2008. After reconstructing the chosen MODIS/NDVI products using the Savitzky-Golay filter at the specified parameters （No.of the envelope iterations：3, adaption strength：2 and the window size：5）,the mean NDVI value of six key images increased by about 0.04.（2） Based on statistical analysis, the 2005-2007 and δall（damage threshold）were estimated at 0.044 and 0.048 respectively without large-scale natural disturbance observed during September 30 and May 9, but there was distinct shift amplitude（0.041） between μ2005-2007 and μ2007-2008 after snow disasters. According to forest damage threshold, damaged pixels accounted for 28.6% of the total forest pixels, which was above the result based on subcompartment investigation （17.7%）.（3） The snow disasters in the southeast and northeast Guizhou Province were the worst, containing Autonomous Prefectures of South Guizhou Province, Autonomous Prefectures of Southeast Guizhou Province, Tongren City, etc. And then, in terms of different damage levels, 11 Severe damage counties （Dejiang, Yanhe, Duyun, etc） and 10 moderate damage counties （Meitan, Rongjiang, Tongzi, etc） were determined at county level. The derived forest damage maps due to disasters from the current methods were in consistency with field surveys in some extent, with a kappa coefficient of 0.86（above 0.75） derived from a statistical test. The adopted analytical flow for assessing forest losses due to snow disasters in Guizhou Province was based on remote sensing and geographical information systems operations. This method also provides a new idea for quick assessment on forest disasters at a regional scale, without relying on ground inventory or sampling.