星载LiDAR与HJ-1 A/HSI高光谱数据联合估测区域森林冠层高度

摘要

[Objective]In this study,ICESat-GLAS waveforms was combined with HJ-1A/HSI hyperspectral imageries to realize regional estimation of forest canopy height. [Method]We extracted parameters ( waveform length W and the terrain slope parameter TS) from ICESat-GLAS waveforms,and built the forest canopy height model with W and TS. The model was used to calculate forest canopy height within each GLAS footprint of the study area. For HSI imageries,the minimum noise fraction( MNF) method was applied to decrease noise and reduce the dimensionality of HSI imageries and the first three MNF components(MNF1,MNF2,MNF3)were selected for further research. Afterwards,the SVR method was applied to establish the relationship between GLAS estimated forest canopy height and the three MNF components,and accordingly the full covered regional forest canopy height map was produced.[Result]The results showed that there was a significant linear relationship between TS and terrain slope ( R2 =0. 78 ) . The R2 and RMSE value of the forest canopy height model built by W and TS were 0. 78 and 2. 51 m,respectively,and the validation results were R2 =0. 85 and RMSE=1. 67 m. The R2 and RMSE of SVR model were 0. 70 and 3. 62 m,respectively,with the validation results of R2 =0. 67,RMSE=4. 42 m. The estimation error of the forest canopy height map was calculated and analyzed by field sample data,and the maximum,minimum and mean value of estimation error were 7. 10 m,0. 07 m and 1. 78 m, respectively. The standard deviation was 1. 49 m,as well as Q1 and Q3 were 0. 75 m and 2. 31 m,respectively.[Conclusion]TS can perfectly reflect terrain slope. In addition,the linear relationship model built in the study overcomes the difficultly explaining problem of logarithm model in flat area. The study demonstrated that it holds great potential to estimate regional forest canopy height by combining ICESat-GLAS waveforms and HJ-1A/HSI hyperspectral imageries, which overcome the disadvantage of ICESat-GLAS in the aspect of regional estimation caused by its discrete distribution and improve the estimation accuracy.%【目的】将 ICESat-GLAS波形数据与 HJ-1A/HSI高光谱数据联合，借助 HSI高光谱数据提供的连续高分辨率光谱信息，实现区域森林冠层高度的估测，降低由于GLAS光斑呈离散条带状分布无法覆盖整个研究区造成的估测误差。【方法】首先，从平滑后的 ICESat-GLAS波形数据中提取波形参数(波形长度 W 和地形坡度参数 TS)，基于 W 和 TS建立 GLAS森林冠层高度估测模型，并利用此模型计算研究区所有 GLAS光斑内的森林冠层高度；然后，采用最小噪声分离法(MNF)对 HJ-1A/HSI高光谱数据进行降维，提取前3个 MNF分量(MNF1，MNF2，MNF3)；最后，基于支持向量回归机( SVR)算法，利用 GLAS估测的森林冠层高度和3个 MNF 分量建立区域森林冠层高度SVR估测模型，并估测研究区内无 GLAS 光斑覆盖区域的森林冠层高度，生成森林冠层高度分布图。【结果】从ICESat-GLAS波形数据中提取的地形坡度参数 TS与野外实测地形坡度具有显著线性关系( R2=0.78)；基于 W 和TS建立的 GLAS森林冠层高度估测模型的 R2=0.78，RMSE=2.51 m，模型验证的 R2=0.85，RMSE=1.67 m；基于支持向量回归机算法建立的 SVR模型建模的 R2=0.70，RMSE =3.62 m，模型验证的 R2=0.67，RMSE =4.42 m。采用野外数据对最终得到的森林冠层高度分布图的估测误差进行分析，结果估测误差最大值为7.10 m，最小值为0.07 m，平均值为1.78 m，估测误差的标准差为1.49 m，Q1为0.75 m，Q3为2.31 m。【结论】从ICESat-GLAS波形数据中提取的地形坡度参数 TS能够很好地反映地形坡度的变化，本研究建立的线性关系模型可克服对数关系模型在平坦地区解释困难的问题。基于支持向量回归机算法，将 ICESat-GLAS波形数据与 HJ-1A/HSI高光谱数据联合，可克服 ICESat-GLAS由于光斑呈离散条带状分布无法实现区域森林冠层高度估测的不足，实现对区域森林冠层高度的高精度估测。