基于高分辨率遥感影像的森林地上生物量估算

以南京市紫金山林区为研究区，利用e-Cognition面向对象分类方法，基于光谱和空间信息融合后的IKONOS影像提取单木树冠阳性冠幅(PoCA, Positive crown area)信息，并结合野外实测的样方生物量数据，分别建立了针叶林和阔叶林地上生物量 (AGB, Aboveground Biomass)的遥感估算模型，并利用实测森林生物量数据对模型进行了验证。结果表明，基于遥感影像提取的PoCA与实测AGB存在较好的非线性相关关系，所建针叶林AGB估算模型的可靠性优于阔叶林模型。对建模样本而言，估算的针叶林和阔叶林AGB与观测数据比较的R2分别为0.62 (P<0.01，n=9) 和0.56(P<0.01，n=16)。验证表明，所建AGB估算模型的可靠性较好，估算的针叶林和阔叶林AGB与观测数据比较的R2分别为0.55(P<0.01，n=6) 和0.52(P<0.01，n=10)，但当AGB较低时，模型结果偏高，AGB较低时，模型结果偏低。研究说明通过高分辨率遥感数据的融合、提取树冠信息进行生物量估算是可行性的。

Estimation of forest aboveground biomass using high spatial resolution remote sensing imagery

Forests are important terrestrial ecosystems and play an important role in maintaining ecological environment. Globally, they are acting as a carbon sink and a crucial player in alleviating global climate change. Forest AGB (Aboveground Biomass) is an important indicator of forest functioning and a key component of the carbon cycle in forest ecosystems. The accurate estimation of AGB is of importance for study the budget of global terrestrial ecosystems. However, it is challengeable to map AGB at regional and global scales. In recent years, various studies have been conducted to retrieve AGB using remote sensing data at regional scales due to the more effectiveness and lower cost of remote sensing compared with the traditional inventory method. This study took Zijin Mountain, located in the urban area of Nanjing city, as the study area to explore the applicability of high resolution remote sensing data in retrieving AGB. This study area is dominantly covered by various species of trees. The spectral and spatial information in an IKONOS image were first fused using the Brovey transformation method to generate images of green, red, near infrared bands at a spatial resolution of 1m. Then, positive crown area (PoCA) of individual trees was delineated using the reflectance of green, red, near infrared bands and normalized difference index retrieved from the fused images and the object-oriented classification method implemented in the e-Cognition platform. The classification rules were determined using the See 5 software. Meanwhile, field campaigns were conducted to record the number of trees, the height and diameter at breast height (DBH) of individual trees, tree species, geological coordinates, and topography at 41 representative plots (16 plots for coniferous forests and 25 plots for broadleaf forests) with an area of 25 m×25m. For each plot, AGB was calculated using models developed in previous studies and field measured height and DBH of individual trees. These field samples were randomly selected for developing models (25 samples) and validating the developed modes (16 samples). The retrieved PoCA data were used in conjunction with plot-level AGB measured to develop empirical models for estimating AGB. The developed models were further validated using measured AGB. The results show that PoCA retrieved from remote sensing data shows distinguishable spatial patterns and is tightly correlated with AGB. The developed empirical model is more applicable for coniferous forests than for broadleaf forests in retrieving AGB. As to the samples used to develop the models, the R² values of estimated AGB against field measurements are 0.62 (P < 0.01,n=9) for coniferous forests and 0.56 (P < 0.01,n=16) for broadleaf forests, respectively. The validation indicates that the developed models are reliable. The R² values of estimated AGB against measurements are 0.55 (P < 0.01,n=6) for coniferous forests and 0.52 (P < 0.01,n=10) for broadleaf forests, respectively. However, the models tend to overestimate AGB under the condition of low AGB and to underestimate AGB when AGB is high. This study indicates that it is practically feasible to estimate forest AGB using the tree canopy area information retrieved from fused high resolution remote sensing image.