Modeling forest aboveground biomass by combining spectrum, textures and topographic features

Abstract

Many textural measures have been developed and used for improving land cover classification accuracy, but they rarely examined the role of textures in improving the performance of forest aboveground biomass estimations. The relationship between texture and biomass is poorly understood. In this paper, SPOT5 HRG datasets were ortho-rectified and atmospherically calibrated. Then the transform of spectral features is introduced, and the extraction of textural measures based on the Gray Level Co-occurrence Matrix is also implemented in accordance with four different directions (0°, 45°, 90° and 135°) and various moving window sizes, ranging from 3 × 3 to 51 × 51. Thus, a variety of textures were generated. Combined with derived topographic features, the forest aboveground biomass estimation models for five predominant forest types in the scenic spot of the Mausoleum of Sun Yat-Sen, Nanjing, are identified and constructed, and the estimation accuracies exhibited by these models are also validated and evaluated respectively. The results indicate that: 1) Most textures are weakly correlated with forest biomass, but minority textural measures such as ME, CR and VA play a significantly effective and critical role in estimating forest biomass; 2) The textures of coniferous forest appear preferable to those of broad-leaved forest and mixed forest in representing the spatial configurations of forests; and 3) Among the topographic features including slope, aspect and elevation, aspect has the lowest correlation with the biomass of a forest in this study.