Representing landscape heterogeneity through an integrated, spatially explicit model based on EO data

Landscapes are structurally diverse and heterogenic in various aspects. Still, the interplay of various structure-related aspects suggests that heterogenic appearance is not arbitrary, but a result of dissipative effects in a systemic sense. In other words, heterogeneity can be considered as an arrangement of homogeneous (sub-)units. In order to represent such ‘ordered heterogeneity’ we integrate different spatio-structural indicators derived from Earth observation (EO) data into a conceptually meaningful and statistically sound index. The use of EO data thereby offers advantages such as area-wide coverage, cost-efficiency, objectivity and transferability of results. Different data sources like VHR imagery and LiDAR data allow combining information on species composition and horizontal heterogeneity, with such on vertical structure of vegetation and terrain. For the integration of the multivariate datasets we use regionalization techniques following the geon concept. Thereby we derive units of uniform behavior (geons) through aggregating the multiple set of underlying indicators instead of using a-priori defined boundaries. The resulting homogenous units with minimized inner variance reveal gradients towards neighboring regions and thus explicitly represent the spatial distribution of the examined phenomenon. The approach is adaptive to the scale of interest. In addition, the resulting geons can be assessed by landscape metrics in terms of their spatial behavior. We tested the approach for assessing structure-related habitat quality of a riparian forest in Austria, under two aspects: (1) quantitatively by constructing a continuous habitat quality index and (2) qualitatively in form of categorical classes characterized by the share or the specific behavior of the underlying indicators. The index is based on EO-derive indicators of four dimensions of forest quality: (a) tree species composition, (b) horizontal forest structure, (c) vertical forest structure and (d) water regime. Findings of this study show that for patches with high habitat quality the occurrence of favourable tree species is mainly responsible for the good status; they are large patches that are well connected and spatially located primarily along water bodies. Instead, areas with low habitat quality are mainly smaller, separated forest plantations with non-characteristic tree species composition. To test the robustness of the habitat quality index, the set of indicators underwent a local sensitivity analysis. While all indicators have similar influence on the robustness during the calculation of the quality index, visual comparison of delineations excluding one of the indicators at a time reveal that tree species composition has the strongest influence during the delineation process of geons. Altogether we consider the presented approach suitable for the assessment and monitoring of habitat quality, in particular on landscape level. Advantages with respect to spatially explicit conservation measures are: the decomposability of the index in its underlying indicators, potentially adjusting the scale to a specific monitoring level, the integration of EO data and the provision of a statistically sound, repeatable method that can easily be transferred to other habitat types or monitoring sites.