Estimation of Structural Diversity in Urban Forests Based on Spectral and Textural Properties Derived from Digital Aerial Images

Abstract

Urban forests generally have a heterogeneous structure consisting of small vegetation patches. High spatial resolution digital aerial images are still a primary data source for urban forest inventories. In the present study, the estimation possibilities of the structural diversity of urban forests were evaluated using image properties extracted from digital aerial images. Firstly, relationships between structural diversity indices and image properties were determined using the correlation analysis. It was found out that structural diversity indices were significantly correlated with spectral and textural properties. The strongest relationship was calculated between the normalized difference vegetation index and species-based Shannon–Wiener diversity index$\left( {H_{\text{s}}^{\prime } } \right)$ (r = 0.599, p < 0.01). The relationship between textural properties and structural diversity indices was slightly lower compared to spectral properties. The strongest relationship between textural properties and structural diversity indices was calculated between the Entropy values derived from DVI and $H_{\text{s}}^{\prime }$ (r = 0.478, p < 0.01). Afterward, each used diversity index was modeled as a function of the textural and spectral properties of digital aerial images. Univariate and multivariate linear regression models were used for this purpose. While the adjusted coefficient of determination $\left( {R_{\text{adj}}^{2} } \right)$ of univariate regression models varies between 0.07 and 0.37, the $R_{\text{adj}}^{2}$ values of a multivariate model vary between 0.13 and 0.57. Among the developed models, only the estimation models of tree size diversity $\left( {H_{\text{h}}^{\prime } } \right)$ and tree species diversity $\left( {H_{\text{s}}^{\prime } } \right)$ provided an estimation accuracy that could be used in practice.