Remote sensing and GIS based ecological modelling of potential red deer habitats in the test site region DEMMIN (TERENO)

Introduction: The destruction of habitats has not only reduced biological diversity but also affected essential ecosystem services of the Central European cultural landscape. Therefore, in the further development of the cultural landscape and in the management of natural resources, special importance must be attached to the habitat demands of species and the preservation of ecosystem services. The study of ecosystem services has extended its influence into spatial planning and landscape ecology, the integration of which can offer an opportunity to enhance the saliency, credibility, and legitimacy of landscape ecology in spatial planning issues.

Objective: This paper proposes a methodology to detect red deer habitats for e.g. huntable game. The model is established on remote sensing based value-added information products, the derived landscape structure information and the use of spatially and temporally imprecise in-situ data (e.g. available hunting statistics). In order to realize this, four statistical model approaches were developed and their predictive performance assessed.

Methods: Altogether, our results indicate that based on the data mentioned above, modeling of habitats is possible using a coherent statistical model approach. All four models showed an overall classification of > 60% and in the best case 71,4%. The models based on logistic regression using preference data derived from 5-year hunting statistics, which has been interpreted as habitat suitability. The landscape metrics (LSM) will be calculated on the basis of the Global Forest Change dataset (HANSEN et al. 2013b ). The interpolation of landcover data into landscape-level was made with the software FRAGSTAT and the moving window approach.  Correlation analysis is used to identify relevant LSM serving as inputs; logistic regression was used to derive a final binary classifier for habitat suitability values. Three model variations with different sets of LSM are tested using the unstandardized regression coefficient. Results lead to an insight of the effect of each LSM but not on the strength of the effect. Furthermore, the predicted outcome is rather difficult to interpret as different units and scales for each LSM are used. Hence, we calculated the fourth model using the standardized regression coefficient. It harmonized the measurement units of the LSM and thus allowed a better comparison, interpretation, and evaluation.

Conclusion:  Our research reveals that applying a statistical model using coarse data is effective to identify potential red deer habitats in a significant qualitative manner. The presented approach can be analogously applied to other mammals if the relevant structural requirements and empirical habitat suitability data (e.g. home range, biotopes, and food resources) are known. The habitat preferences of red deer are best described by LSM concerning area-relation and wildlife-edge relations. Most important are edges between meadows, pastures or agricultural field and forest, as well as short paths between those elements for food resources. A large proportion of forest is important for species survival and positively influences the occurrence of red deer. Outcomes help to understand species –habitat relation and on which scale wildlife perceives the landscape. In addition, they support the practical habitat management and thus the overall species diversity.