影像纹理窗口大小对山地阔叶林不同群落有效叶面积指数估测的影响

森林叶面积指数是陆地表面过程和地球系统气候模型的基本参数,更是森林结构的关键参数之一,已广泛应用于辐射、植物光合作用和降雨截流估测等方面。论文以川西南山地阔叶林5种不同群落类型为研究对象,基于地面调查的112个20 m×20 m样地和SPOT 5数据,运用5种图像处理技术,包括光谱反射率、植被指数、影像单波段纹理、简单波段比纹理和主成分纹理,提取相应影像信息,建立多元回归模型估算有效叶面积指数（LAIe）。结果表明：光谱反射率、单波段纹理参数和植被指数对LAIe估测能力相对较低,利用植被指数仅获得实测LAIe约65%的精度（R²=0.65,RMSE=0.28 m²/m²）;更为有效的是运用所有比值处理的纹理特征参数值来估测LAIe,可获得实测LAIe约74%的变异（R² =0.74,RMSE=0.20 m²/m²）;改进最理想的是利用主成分处理建立的回归模型（R²=0.85,RMSE=0.10 m²/m²）。不同群落的LAIe估测,整体上相应地优于研究区结果,其中栲群落决定系数R²更是高达0.89（RMSE=0.07 m²/m²）。对于研究区阔叶林以窗口7×7、9×9比较成功,而各群落以窗口9×9较好。因此比值处理、主成分处理的纹理特征参数引入及高空间分辨率数据的使用,能显著提高LAIe估测精度。

Effects of Image Texture Window Sizes on LAIe Estimation of Different Communities in Montane Broad-leaved Forest

Forest canopy leaf area index (LAI), a critical forest structural parameter, has been proven to be representative of canopy foliage content and crown structure and has been widely used for the estimation of radiation attenuation, plant photosynthesis, and precipitation interception among others. It is further a fundamental parameter in land surface processes and earth system climate models. Remote sensing methods offer an opportunity to improve in each of these requirements but are typically limited by the necessity for labor intensive validation and sparsely collected in situ measurements. This research investigates the potential of high resolution optical data from the SPOT 5 VGR sensor for LAIe estimation in five communities of montane broad-leaved forest in southwest Sichuan, using five different types of image processing techniques including 1) spectral reflectance, 2) commonly used vegetation indices, 3) texture parameters, 4) texture parameters of band ratio and 5) texture parameters of principal component（PC）. Simple linear and stepwise multiple regression models were developed with LAIe data from 112 field plots and image parameters derived from these techniques. Results indicated that spectral reflectance, texture parameters of spectral bands and commonly used vegetation indices had relatively low potential for LAIe estimation, as only about 65% of the variability in the field data was explained by the model (R²=0.65,RMSE=0.28 m²/m²) when using vegetation indices. However, the simple ratio of texture parameters were found to be more effective for LAIe estimation with explained variability of 74% (R²=0.74,RMSE=0.20 m²/m²). The result was further improved to R²=0.85 (RMSE=0.10 m²/m²) when using the texture parameters of PCs. With regard to five communities, LAIe estimation was found to be more effective than in the whole study area. Castanopsis fargesii community was proven to have the best model (R² =0.89,RMSE=0.07 m²/m²). Generally, window sizes of 7 × 7 and 9 × 9 were more successful for the whole study area, and window size of 9 × 9 performed well for the five communities. The results suggest that the performance of LAIe estimation can be improved significantly by using the texture parameters of high resolution optical data, and further improvement can be obtained by using the texture parameters of PCs as this method combines the advantages of both the texture and the PCs.