三北造林区人工林地上生物量动态的时序定量遥感估算方法研究

Dynamic change of artificial forest biomass is a key indicator assessing the ability of carbon sequestration of afforestation projects implemented in the ‘Three North’ afforestation area in China. However, ‘Three North’ region is an arid and semi-arid area, and the fractional coverage of trees in this region is relatively low. Currently, complicated background under trees and insufficient use of temporal information derived from remote sensing data have hindered widespread use of estimating dynamics of artificial forest aboveground biomass using remote sensing data. Therefore, in this study, Yulin region in the Shaanxi province is selected as the study area, and a new method of estimating artificial forest aboveground biomass dynamics is proposed based on remote sensing time-series data. Firstly, the surface reflectance products of all available Landsat time-series images since 1985 were derived through atmospheric correction, relative radiometric correction and terrain correction. Secondly, a detection algorithm was developed to acquire the background spectrum at pixel scale based on field measurements and spectrum analysis, and the subpixel-level artificial forest cover changing trajectories were reconstructed using relative linear spectral mixture analysis. Meanwhile, inter-annual and intra-annual variation characteristics for different typical land covers were explored, and the detection rules of change timing and direction of artificial forest were established, and tree age information with high accuracy was derived using remote sensing data and established detection rules. Finally, A remote sensing model of dynamic calculation for artificial forest aboveground biomass was developed, and artificial forest aboveground biomass time-series in the study area were accurately simulated, and temporal and spatial variation patterns of artificial forest aboveground biomass was also analyzed. This study about the estimation method is of great theoretical and practical significance for improving simulation accuracy of artificial forest carbon sequestration dynamics in the arid and semi-arid areas in China.

人工林生物量动态是评估三北造林区固碳能力的重要指标。然而，该区域属我国干旱/半干旱地区，乔木林覆盖率较低，复杂背景和遥感时序信息利用不足制约了区域人工林地上生物量动态遥感估算应用。为此，本项目以陕西榆林地区为例，开展人工林地上生物量动态遥感估算方法研究。首先通过大气校正、相对辐射校正以及地形校正，构建1985年至今所有可用的Landsat系列卫星地表反射率时序数据集。其次结合实地观测数据，探索像元尺度人工林背景光谱识别算法，利用相对线性光谱混合分析实现亚像元级人工林覆盖变化轨迹重建，同时，挖掘典型地物的年际时序变化特征和年内物候特征，构建人工林覆盖变化时间与变化方向时序探测规则，实现树龄信息的高精度动态提取；最后构建人工林地上生物量动态遥感模型，实现生物量的高精度动态模拟和时空变化格局分析。该方法研究对全面提高我国干旱/半干旱地区人工林碳储量动态估算精度具有重要的理论和现实意义。

干旱/半干旱地区植被时空变化高异质性为人工林地上生物量遥感估算带来极大挑战。为此，本项目测量了毛乌素沙地内126个人工林样方，基于无人机影像、Landsat卫星长时间序列数据，研发了稀疏乔灌木地上生物量估算方法。1）研发了林草/农田地表覆盖时变探测及植被类型自动分类算法。该方法可有效识别由降雨和人类活动对植被生长的无规则扰动造成的虚假地表覆盖变化。林草空间分类精度为88.9%，变化时间探测精度为±2年以内。2）研发了融合多尺度遥感观测信息的植被覆盖度反演方法。基于高分辨率无人机影像能够有效提取乔灌木冠层覆盖度（Kappa=0.97）。基于Landsat NDVI方法（RMSE=17.27%）相比，MESMA反演精度较高（RMSE=9.28%）。但对于乔灌草混合情况下的乔、灌、草冠层盖度准确提取仍需依赖高空间分辨率遥感技术。3）利用乔灌木所有树种数据建立的冠层盖度—生物量模型的精度最差（R2=0.23，RMSE=23.45t/ha），利用乔木和灌木不同植被功能型聚类策略建立的冠层盖度-生物量模型的精度有所提高（R2=0.6，RMSE=15.21t/ha），利用不同乔灌木树种聚类策略建立的冠层盖度-生物量模型反演精度最高（R2=0.77，RMSE=11.82 t/ha）。由于乔灌木不同树种具有不同生物物理参数特征，乔、灌木生物量对其冠层盖度的敏感性存在显著差异。随着类别划分的越精细，冠层盖度-生物量模型精度就会越高。4）由于木本植被包含枝干和叶子部分，而草本植被仅有叶子部分，因此木本和草本植被生物量对冠层LAI的敏感性存在显著差异，而光学遥感信号主要受LAI影响，因而光谱信息难以解释乔灌草混合情况下乔灌木地上生物量变异。利用不考虑草地背景的VIs-生物量模型精度普遍较低，反演精度排序为NDWI>NIRv>TCG> MSAVI> RVI>NDVI，其中NDWI精度最高（R2=0.52，RMSE=11.79t/ha）。考虑草地背景（0%，0%-15%，15%以上）聚类策略的VIs-生物量模型可增强光谱信息对乔灌木地上生物量变异解释能力，反演精度普遍显著提升，最高精度R2=0.86，RMSE=3.95t/ha。本研究对1986~2019年的稀疏乔灌木地上生物量进行了长时间序列反演。研究对全面提高我国干旱/半干旱地区人工林碳储量动态估算精度具有重要的理论和现实意义。