Worldview-2不同波段纹理特征对森林蓄积量估算精度影响

为检测高分辨率遥感影像不同波段纹理特征对于森林蓄积量估算精度的影响，以湖北省荆门市京山县太子山林场马尾松纯林为对象，基于灰度共生矩阵的方法分别提取高分辨率遥感影像Worldview-2 红光、绿光、蓝光、近红外波段和全色波段的纹理特征，利用随机森林算法，分别建立野外样地蓄积量与纹理参数的模型。结果表明，全色波段对马尾松森林的精度最高（R²=0.86，RMSE=47.37 m³·hm^-2），其次是绿色波段（R²=0.85，RMSE=50.82 m³·hm^-2）和近红外波段（R²=0.85，RMSE=46.85 m³·hm^-2），蓝色波段（R²=0.68，RMSE=60.72 m³·hm^-2）和红色波段（R²=0.69，RMSE=56.27 m³·hm^-2）的精度最低；窗口大小对模型精度影响较小，全色波段的R²取值在0.82~0.86，RMSE取值在47.66~51.99 m³·hm^-2，多光谱波段的R²取值在0.88~0.89；蓝色和红色波段的非相似度（DIS）的估算模型精度相对较高，绿色波段的对比度（CON）（R²=0.87，RMSE=46.21 m³·hm^-2）估算精度最高，红色波段的非相似度（R²=0.68，RMSE=58.30 m³·hm^-2）估算精度较高，近红外波段的角二阶矩阵（ASM）（R²=0.68，RMSE=60.30 m³·hm^-2）精度最高，全色波段的对比度、相关性、熵、变化量模型精度较高，R2为0.85。利用高分辨率遥感影像纹理特征估算森林参数时需综合考虑不同波段的纹理特征对模型的贡献。

Effects of Texture Parameters of Different Bands of Worldview-2 Images on the Estimation of Forest Volume

To test the influences of texture parameters of different bands of high resolution images on the estimation accuracy of forest volume,pure Pinus massoniana plantations occurring in Taizishan Forest Farm located in Jingshan County,Jingmen City of Hubei Province were used as the research objects.The method of gray-level co-occurrence matrix was adopted to extract texture parameters of red,green,blue,near-infrared and panchromatic bands of Worldview-2 images.Random forest algorithm was applied to establish the models of forest volume based on field investigation and texture parameters.The results showed that panchromatic band had the highest accuracy for the volume estimation of P.massoniana plantations (R²=0.86,RMSE=47.37 m³·hm^-2),followed by the green band (R²=0.85,RMSE=50.82 m³·hm^-2) and near-infrared band (R²=0.85,RMSE=46.85 m³·hm^-2),while the blue band (R²=0.68,RMSE=60.72 m³·hm^-2) and red band (R²=0.69,RMSE=56.27 m³·hm^-2) had the lowest accuracy.Window size presented little effect on the accuracy of the model.R² value of the panchromatic band was between 0.82 and 0.86 and RMSE value was between 47.66 and 51.99 m³·hm^-2.R² value of the multispectral band was between 0.88 and 0.89.The accuracy of the dissimilarity (DIS) of the blue and red bands was relatively high,and the contrast (CON) of the green band (R²=0.87,RMSE=46.21 m³·hm^-2) had the highest estimation accuracy,and the DIS of the red band (R²=0.68,RMSE=58.30 m³·hm^-2) had higher estimation accuracy.Angular Second Moment (ASM) in the near infrared band (R²=0.68,RMSE=60.30 m³·hm^-2) had the highest accuracy.The accuracy of contrast,correlation,and entropy of the panchromatic band had higher accuracy,R² was 0.85.The results indicated that when texture features of high-resolution remote sensing image were used to estimate forest parameters,it was necessary to consider the contribution of texture features of different bands to the model building comprehensively.