Modelling habitat suitability for alpine rock ptarmigan (Lagopus muta helvetica) combining object-based classification of IKONOS imagery and Habitat Suitability Index modelling

Abstract

The maintenance and restoration of high-quality habitats of wildlife species in alpine ecosystems are key issues in conservation biology. The rock ptarmigan (Lagopus muta helvetica), which prefers open habitats above the treeline, is listed in Annex II of the EU Bird Directive. Large areas identified as potentially important for conservation and restricted financial resources for the implementation of conservation activities necessitate the development of tools supporting habitat monitoring and management. We developed a knowledge-based Habitat Suitability Index (HSI) model for rock ptarmigan and combined it with results of an object-based image analysis of very high resolution (VHR) satellite images (IKONOS) to create a rock ptarmigan habitat suitability map in the eastern Alps of Austria. The mechanistic habitat model contained 9 habitat variables, decisive for summer habitat use of rock ptarmigan (i.e., slope, patchiness, and land cover classes for: rocks, rocks intermixed with vegetation, scree, dwarf shrubs, dwarf pine, alpine/subnival grassland, and forest). We used Definiens Professional 5.0© software for the object-based image analyses, applying multi-resolution segmentation methods. We generated a classification hierarchy comprising the same variables used in the HSI model, each augmented by areas “in shadow”, and took into account clouds, water bodies and human infrastructures. We assessed classification accuracies, applying an Error Matrix based on TTA Mask. We reached an overall classification accuracy of 0.75 and a kappa statistic value of 0.70, the latter indicating good to very good agreement. The classification results indicated that the object-oriented image classification approach using VHR data was appropriately used to create an adequate thematic map for further habitat modelling. We calculated the habitat suitability maps using MapModels. Model output was validated with ptarmigan presence–absence data, using signs (droppings) as indicators of presence. We compared presence–absence data and results of habitat suitability classification employing contingency tables and non-parametric correlations. Frequencies of sample plots with rock ptarmigan signs significantly differed between the habitat suitability classes and significantly correlated with the HSI level. Combining the mechanistic HSI model with an object-base image analysis of VHR satellite images was an effective tool for the spatially explicit assessment of habitat suitability and could be useful in regional monitoring, planning and management activities for ptarmigan.