1.College of Forestry,Northeast Forestry University,Harbin 150040,China
2.Key Laboratory of Sustainable Forest Ecosystem Management (Northeast Forestry University),Ministry of Education,Harbin 150040,China
Show less
文章历史+
Received
Published
2025-11-10
2026-07-20
Issue Date
2026-09-24
PDF (3001K)
摘要
森林蓄积量作为衡量森林资源数量与质量的重要指标,其年度更新在森林资源管理中发挥着重要作用。然而,目前借助遥感数据完成森林蓄积量的年度更新存在较大的低值高估和高值低估问题。为此,以黑河市森林为研究对象,使用2010年森林资源一类清查固定样地数据与同期Landsat-7多光谱遥感影像数据,采用似乎不相关回归模型(seemingly unrelated regression,SUR)和最佳线性无偏预测器(best linear unbiased prediction,BLUP),构建“遥感-样地”协同优化的森林蓄积量估算体系。结果表明,“遥感-样地”协同优化后的森林蓄积量估算精度显著提高,同化校正后的模型调整决定系数(R2)从0.200提升至0.539,模型的拟合优度明显增加;均方根误差(RMSE)由23.207降低至17.541 m3/hm2,改进百分比达24.4%,表明蓄积量的预测值与真实值的绝对偏差程度显著降低;蓄积量估计值的标准差由17.367降低至12.146 m3/hm2,进一步证实优化后的模型参数估计的稳定性增强。研究表明所构建的数据同化模型显著提升了森林蓄积量遥感估算的精度和可靠性,为区域尺度森林碳汇评估和森林可持续经营决策提供可靠的技术支撑,同时支持了地级市尺度森林年度更新工作。
Abstract
Forest stock volume, a key indicator for measuring the quantity and quality of forest resources, plays a pivotal role in forest resource management through its annual updates. However, there remains a significant issue of overestimation for low values and underestimation for high values in the current annual update of forest stock volume using remote sensing data. Taking the forests in Heihe City as the research object, this study constructed a ‘remote sensing-plot’ synergistic optimization system for forest stock volume estimation by integrating seemingly unrelated regression (SUR) model and best linear unbiased prediction (BLUP), based on 2010 national forest inventory permanent sample plot data and concurrent Landsat-7 multispectral remote sensing image data. The results indicated that the estimation accuracy of forest stock volume after ‘remote sensing-plot’ synergistic optimization was significantly improved, the coefficient of determination (R2) of the assimilated and calibrated model increased from 0.200 to 0.539, manifesting a marked enhancement in model goodness-of-fit; the root mean square error (RMSE) decreased from 23.207 to 17.541m3/hm2, with an improvement percentage as high as 24.4%, which signified a substantial reduction in the absolute deviation between predicted and actual values. The standard variance for volume estimates dropped from 17.367 to 12.146m3/hm2, further confirming the enhanced stability of parameter estimation in the optimized model. It is evident that the data assimilation model constructed in this study significantly improves the accuracy and reliability of remote sensing-based forest stock volume estimation, providing robust technical support for regional-scale forest carbon sink assessment and sustainable forest management decision-making. It also supports the annual forest update work at the city-level scale.
森林资源清查数据选用了黑河市2010年森林资源一类清查样地数据,样地大小为0.06 hm2,共设置493个样地,覆盖研究区域内的主要森林类型和地形条件,分别测量样地经度、纬度、地貌、海拔、坡度、坡向、植被类型、优势树种、平均树高、平均胸径和郁闭度等信息。在每个样地内,对所有胸径(diameter at breast height,DBH,式中记为DBH)大于或等于5 cm的数据进行每木检尺。根据样地每木检尺数据,通过材积方程计算不同树种的材积,对于常见树种,材积计算公式如下。
本研究选用的Landsat-7多光谱数据是由美国地质调查局(united states geological survey,USGS)和美国国家航空航天局(national aeronautics and space administration,NASA)联合运营,于1999年4月15日发射,其搭载的增强型专题制图仪(enhanced thematic mapper,ETM+)是对早期Landsat-5 TM传感器的改进,具有优越的数据质量和良好的数据连续性。但需要注意的是由于Landsat-7卫星在2003年5月31日发生了故障,扫描线校正器(scan line corrector,SLC)永久失效,出现了一景影像中约有四分之一数据丢失的现象[19]。所以在进行Landsat-7多光谱数据预处理时需要进一步进行条带修复处理以确保影像的一致性、完整性和数据质量。本研究下载了2010年7月份研究区域Landsat-7多光谱数据,并对其进行辐射定标、大气校正、几何校正、条带修复以及镶嵌裁剪等预处理。
式中:是由子模型特定误差方差和交叉子模型误差协方差组成的正定对称矩阵; In 为n阶单位矩阵;n为样本容量;为方程间误差协方差矩阵的元素。本研究采用广义最小二乘法(feasible generalized least squares,FGLS)法估计参数向量,首先对每个子模型使用普通最小二乘法(ordinary least squares,OLS)拟合一个初始模型,以获取残差信息,并根据残差信息,估计残差项的方差-协方差矩阵。然后使用此矩阵对原始模型进行加权最小二乘估计。
。
。
式中:为模型参数估计向量; X 为块对角形式的设计矩阵;为 X 的转置矩阵;为的逆矩阵; y 为堆叠的因变量观测值向量; In 为n阶单位矩阵;n为样本容量;为方程间误差协方差矩阵元素的估计值;为两方程残差向量的内积。
本研究基于Henderson(1975)建立的线性混合效应模型理论体系,采用频率学派框架下的最佳线性无偏预测器(best linear unbiased prediction,BLUP)这一具有严格理论保证的统计方法,构建了面向森林蓄积量估计的多源数据同化系统。从概率论的角度看,BLUP方法本质上是多元正态分布条件下条件期望的显式解,对于一个L维的随机向量 y 被分为两部分,其数学基础可表述为
。
式中: y1为待预测的变量; y2为观测的变量或已知的辅助变量,在联合分布的前提下,通过求解条件实现最优预测。具体而言,设随机向量 y 具有如下分布特性。
式中:是第b个重采样的Bootstrap估计,b=1,2,…,B,且B=1 000。校准后BLUP的Bootstrap方差可能向下偏移,这将产生过窄且偏离中心的置信区间;对于小样本量这种影响更加明显。为了克服这个问题,研究使用了拓展参数Bootstrap程序[22]。具体而言,首先将SUR模型拟合到初始样本数据 y 上,以获得协方差矩阵和回归系数的估计值。随后,生成一个与维度相同的误差向量,的元素独立地从分布中抽取。基于构建Bootstrap数据,并进一步构建完整的Bootstrap样本。在Bootstrap样本上,再次拟合SUR模型,以获得新的估计值和。基于这些估计值,计算自助法最佳线性无偏预测(Bootstrap BLUP),得到和,为第b次Bootstrap的固定效应参数估计,其生成依赖于原始协方差。这一过程重复B次,每次重复都计算和,其中,b表示第b次Bootstrap重复, I 是n×1的单位矢量。对模型参数的不确定性进行量化,基于Bootstrap的方差估计及其标准误表示如下。
ZHAOS.Effects of remote sensing-based dynamic monitoring of forest resources on the ecological environment and resources[J].China Forestry Industry,2023(11):89-91.
ZENGW S, XIAR.Discussion on methodology for generating annual estimates in national forest inventory[J].Forest Resources Management,2021(2):29-35.
[9]
李海月.基于遥感纹理特征的北京市森林蓄积量估测模型研究[D].北京:北京林业大学,2020.
[10]
LIH Y.Research on Beijing forest volume estimation model based on remote sensing texture features[D].Beijing:Beijing Forestry University,2020.
[11]
ROESCHF A, SCHROEDERT A, VOGTJ T.Effects of cycle length and plot density on estimators for a national-scale forest monitoring sample design[J].Forests,2017,8(9):325.
WUH, LYUD W, WANGS Y,et al.Research progress in multi-scale annual monitoring of forest carbon storage under the background of “dual carbon” goal[J].World Forestry Research,2024,37(1):59-64.
[14]
FRANKLINS E, AHMEDO S, WULDERM A,et al.Large area mapping of annual land cover dynamics using multitemporal change detection and classification of Landsat time series data[J].Canadian Journal of Remote Sensing,2015,41(4):293-314.
[15]
SMITHW B.Forest inventory and analysis:A national inventory and monitoring program[J].Environmental Pollution,2002,116:S233-S242.
[16]
NIW J, YUT Y, PANGY,et al.Seasonal effects on aboveground biomass estimation in mountainous deciduous forests using ZY-3 stereoscopic imagery[J].Remote Sensing of Environment,2023,289:113520.
DINGX Y, CHENE X, LIZ Y,et al.A review of remote sensing application in national forest inventory[J].Journal of Nanjing Forestry University (Natural Science Edition),2023,47(1):1-12.
QINS X.Development and experiment of intelligent algorithm in land surface data assimilation[D].Beijing:University of Chinese Academy of Sciences,2013.
WANGZ R, SUNH H, LEIL L,et al.The importance of data assimilation components for initial conditions and subsequent error growth[J].Science China Earth Sciences,2024,54(1):110-121.
ZENGW S, SUNX N, WANGL R,et al.Developing stand volume,biomass and carbon stock models for ten major forest types in forest region of Northeastern China[J].Journal of Beijing Forestry University,2021,43(3):1-8.
Liaoyang Municipal Forestry and Grassland Bureau. One-way and two-way tree volume tables of main tree species in Liaoyang region:DB2110/T 0004—2020 [S].Liaoyang:Liaoyang Municipal Bureau of Market Supervision and Administration,2020.
MAY P.Study on Karst rocky desertification in Karst area based on Landsat-7 SLC-OFF ETM image[J].Journal of Hubei Minzu University (Natural Sciences Edition),2020,38(1):13-18.
WUY S, ZHANGX L.Object-oriented tree species classification with multi-scale texture features based on airborne hyperspectral images[J].Journal of Beijing Forestry University,2020,42(6):91-101.
[36]
李靖.黑龙江穆棱市林业局森林蓄积量估测模型研究[D].北京:北京林业大学,2012.
[37]
LIJ.Study on quantitative estimation of forest volume in Muling Forestry Bureau[D].Beijing:Beijing Forestry University,2012.
[38]
HOUZ Y, MEHTÄTALOL, MCROBERTSR E,et al.Remote sensing-assisted data assimilation and simultaneous inference for forest inventory[J].Remote Sensing of Environment,2019,234:111431.