Estimating Stand Volume Of Coniferous And Broadleaf Forest Based On Texture Of ALOS Imagery In Huairou District

In this study, ALOS imagery data and Landsat TM5imagery data were chosen as data material. Object-oriented classification method was employed to extract coniferous forest information in Huairou District. Broadleaf forest information was extracted on the basis of texture indices derivated from ALOS imagery in Huairou District. Stepwise multiple regression models were developed to describe the relationship between textures (including texture parameters and derivative texture indices of fusion and multispectral image) and field measurements of coniferous and broadleaf forest stand volume. Finally, the distribution map of coniferous and broadleaf forest stand volume was drawn. The main conclusions are as follows:1. In the stage of image pre-processing, an uneven distribution of ground control points in mountainous area results in the phenomenon of ground objects dislocation between high-resolution orthophotos’edges in the overlap area. A wedge-type edge matching method was proposed to accomplish high resolution orthophotos edge matching fast and accurately.2. Based on Landsat TM5image of winter, object-oriented classification method was employed to extract coniferous forest information in the study area under the guidance of aspect subarea classification and multiple thresholds extraction without the participation of ground training plots. The result showed that the classification accuracy was up to94.44%.3. Based on ALOS image texture features and derivative texture indices, special band by which arbor can be separated from shrub and grass vegetation was selected. Object-oriented classification method was employed to extract arbor forest information accurately. The map of coniferous forest information was superimposed on the map of arbor forest information to obtain the distribution of broadleaf forest. Accuracy test showed that the final classification accuracy of broadleaf was up to89.62%. 4. Based on ALOS multispectral image and fusion image, texture parameters and derivative texture indices were calculated at7different window sizes. Stepwise multiple regression models were developed to describe the relationship between textures (including texture parameters and derivative texture indices) and field measurements of coniferous and broadleaf stand volume. The result showed that:(1) The accuracy of retrieval model of forest stand volume was influenced by window sizes at which texture parameters and derivative texture indices were calculated, moreover, the influence of laws can be also followed.(2) The accuracy of retrieval model of forest stand volume can be improved significantly by derivative texture indices, which were calculated on the basis of texture parameters of optical remote sensing data. The value of adjusted R2of fitting models established by derivative texture indices were better than those of texture parameters at the same window size.(3) The accuracy of retrieval models of forest stand volume established by texture parameters and derivative texture indices of fusion image was better than those of multispectral image, namely, the value of adjusted R2of fitting models based on texture can be improved significantly as spatial resolution being improved.(4) The accuracy of retrieval model of forest stand volume can be improved significantly by combination of texture parameters and derivative texture indices at the same window size, the optimal estimation model of forest stand volume was obtained when all of the texture parameters and derivative texture indices of all window sizes were introduced into stepwise multiple regression.