基于Landsat长时间序列的森林扰动参数提取与树高估算

【目的】研究基于遥感影像的森林扰动信息定量提取及其对树高估算的影响,为遥感反演森林参数(树高、生物量)提供参考和借鉴。【方法】选取黑龙江省凉水国家级自然保护区为研究区,以1984—2006年33期Landsat TM/ETM+多光谱遥感影像为数据源,对其进行缨帽变换提取缨帽角(TCA)和缨帽距离(TCD)2个扰动监测指数,采用时间轨迹分析方法(LandTrendr)对TCA与TCD指数进行时间序列重构,分别提取扰动发生的前一年(DBYEA)、扰动发生前的光谱值(DBVAL)、扰动持续时间(DDUR)、扰动量级(DMAG)、扰动后开始修复的时间(RBYEAR)、扰动后开始修复的光谱值(RBVAL)、修复量级(RMAG)和修复持续时间(RDUR)8个时间序列扰动参数。基于单时相Landsat影像光谱信息与单时相Landsat影像光谱信息+森林扰动参数2组变量分别采用随机森林(RF)算法估算树高。【结果】采用单时相Landsat影像光谱信息结合基于TCA和TCD提取的16个时间序列扰动参数建立的树高反演模型预估精度比采用单时相Landsat影像光谱信息建立的树高反演模型预估精度提高6.34%,均方根误差(RMSE)降低0.50 m。树高反演模型中基于TCA提取的时间序列扰动参数变量重要性高于基于TCD提取的时间序列扰动参数变量重要性。【结论】基于LandTrendr提取的森林时间序列扰动参数能够增强反射率与树高之间的相关性,提高遥感树高模型的反演精度,基于TCA提取的森林时间序列扰动参数对树高的解释能力高于基于TCD提取的森林时间序列扰动参数。

Extraction of Forest Disturbance Parameters and Estimation of Forest Height Based on Long Time-Series Landsat

【Objective】 Forest disturbance is the main factor that influences forest height. This study was aimed to extract forest disturbance parameter based on remote sensing image and to know its effect on forest height estimation.【Method】 Liangshui National Nature Reserve in Dailing District, Heilongjiang Province, China, was selected as study area, thirty-three periods of Landsat thematic mapper and enhanced thematic mapper plus(Landsat TM/ETM+) multispectral data from 1984 to 2006 in Liangshui National Nature Reserve were acquired as data sources. The tasseled cap transform was conducted to extracttwo disturbance monitoring indices of the tasseled cap angle(TCA)and tasseled cap distance(TCD). Time trajectory analysis method Landsat-based detection of trends in disturbance and recovery(LandTrendr) was applied to conduct time series reconstruction of the Landsat TCA(TCD) images and extract timeseries disturbance parameters of forest: previous year prior to disturbance onset(DBYEA), spectral value prior to disturbance onset(DBVAL), disturbance duration(DDUR), disturbance magnitude(DMAG), time to start recovery after disturbance(RBYEAR), spectral value for recovery start after disturbance(RBVAL), recovery magnitude(RMAG), recovery duration(RDUR).Two sets of spectral information variables of single-temporal Landsat image with or without time-series disturbance parameters were applied to estimate forest height by using random forest algorithm.【Result】 Compared withthe tree height inversion model based on the spectral information of single-temporal Landsat image, the prediction accuracy of the tree height inversion model according tocombination of the spectral information of single-temporal Landsat image with 16 time-series perturbation parameters based on TCA and TCD increased by 6.34%, and the RMSE decreased by 0.5 m. In the tree height inversion model, the time-series perturbation parameters based on TCA extraction were more important than those based on TCD extraction.【Conclusion】 Time-series disturbance information of forest extracted based on LandTrendr method could enhance the correlation between reflectance and tree height, and could also improve the prediction accuracy of the tree height model based on remote sensing. Time-series perturbation parameters based on TCA extraction is better than those based on TCD extraction to interpret forest height estimation. The method can be used as a reference for remote sensing inversion of forest parameters(e.g., tree height and biomass).