星载大光斑LiDAR与HJ-1A高光谱数据联合估测区域森林地上生物量

以吉林省汪清林业局经营区为研究区,利用HJ-1A/HSI高光谱数据和ICESat-GLAS波形数据,估测区域森林地上生物量。从平滑后的GLAS波形数据中提取波形长度W和地形坡度参数TS,建立GLAS森林最大树高估测模型;从GLAS波形数据中提取能量参数I(植被回波能量Ev和回波总能量E之比),建立GLAS森林郁闭度估测模型;利用GLAS估测的森林最大树高和森林郁闭度联合建立森林地上生物量模型。由于GLAS呈离散条带状分布,无法实现区域估测,因此研究将GLAS波形数据与HJ-1A/HSI高光谱数据联合,基于支持向量回归机算法实现森林地上生物量区域估测,得到研究区森林地上生物量分布图。研究结果显示,基于W和TS建立的GLAS森林最大树高估测模型的adj.R~2=0.78,RMSE=2.51m,模型验证的adj.R~2=0.85,RMSE=1.67m。地形坡度参数TS能够有效的降低地形坡度的影响;当林下植被高度为2m时,得到的基于参数I建立的GLAS森林郁闭度估测模型效果最好,模型的adj.R~2=0.64,RMSE=0.13,模型验证的adj.R~2=0.65,RMSE=0.12。利用森林最大树高和森林郁闭度建立的森林地上生物量模型的adj.R~2=0.62,RMSE=10.88 t/hm~2,模型验证的adj.R~2=0.60,RMSE=11.52 t/hm~2。基于支持向量回归机算法,利用HJ-1A/HSI和GLAS数据建立的森林地上生物量SVR模型,生成了森林地上生物量分布图,利用野外数据对得到的分布图进行验证,验证结果显示森林地上生物量估测值与实测值存在很强的线性关系(adj.R~2=0.62,RMSE=11.11 t/hm~2),能够满足林业应用的需要。因此联合ICESat-GLAS波形数据与HJ-1A高光谱数据,能够提高区域森林地上生物量的估测精度。

Estimation of regional forest aboveground biomass combining spaceborne large footprint LiDAR and HJ-1A Hyperspectral Images

HJ-1A/HSI Hyperspectral images and ICESat-GLAS waveform were used to estimate regional forest aboveground biomass (AGB) in the Wangqing forestry area of Jilin Province, China. Waveform parameters (e.g., waveform length W and the terrain slope parameter TS) extracted from GLAS waveform, were used to build the maximum forest height model. In addition, the energy parameter I (the ratio of vegetation energy and total energy) extracted from GLAS waveform was used to build the forest canopy density model. The final forest AGB model was built using both the maximum forest height and forest canopy density models. However, since GLAS footprints are geographically discrete, the AGB model was unable to produce the full regional coverage of forest AGB. To overcome the discontinuity limitations, HJ-1A/HSI Hyperspectral images were combined with GLAS waveforms to predict the regional forest AGB based on the support vector regression (SVR) method, to fully map the distribution of forest AGB. Results showed that the adj.R² and RMSE of the maximum forest height model were 0.78 and 2.51 m, respectively, with adj.R² of 0.85 and RMSE of 1.67 m as validation results. In the model, TS effectively reduced the impact of terrain slope. When the below vegetation height was set at 2 m, the forest canopy density model with I as dependent variable produced the best fit, with adj.R² and RMSE of 0.64 and 0.13, respectively, and adj.R² of 0.65 and RMSE of 0.12 as validation results. Overall, the adj.R²and RMSE of the forest AGB model were 0.62 and 10.88 t/hm², respectively, with validation results of adj.R²=0.60 and RMSE=11.52 t/hm². Estimated AGB had a strong linear relationship with field inventory AGB (adj.R²=0.62, RMSE=11.11 t/hm²). This study demonstrates that combining GLAS waveform and HJ-1A/HSI hyperspectral images has significant potential to map the full coverage regional forest AGB distribution with a high degree of accuracy.