Research into the sensitivity of fragmentation metrics
Final Report Abstract
Forests are receiving much attention in various functions, among them their role as carbon sinks and sources and as home to the greatest terrestrial biodiversity. They are in the core of international policy processes where reliable data and information about forest resources become crucial for policy formulation and decision making. Development of monitoring techniques to provide this kind of information to decision makers is a key task for the successful implementation of the policies. Landscape metrics describe the fragmentation status of a landscape by quantifying the number, size, shape and position of forest patches. They are fundamental for many landscape monitoring and ecology applications. Thus, they may contribute to providing the required information about forests given that they are sensitive to changes in the fragmentation status but robust against changes of the assessment method. Overall objective of this study was to analyze the sensitivity of landscape metrics in the framework of remote sensing based landscape monitoring in order to increase the interpretability of landscape metrics. Therefore we studied the influence of factors which are inherent in forest cover monitoring that is: i) forest and forest edge definitions, ii) types of landscape models (2D vs. 3D models) and iii) scale of the used landscape models (horizontal /vertical resolution). By analyzing the impact of those factors using geometrical and spatial models we could show that most of the uncertainties when determining forest fragmentation metrics arise from the uncertainties in the underling forest definitions. To give but one example: The Food and Agriculture Organization of the United Nations (FAO) uses a minimum crown cover criterion (e.g. 10%) without defining a reference area where the crown cover is measured. Our simulation showed large differences in forest cover (>50%) for a fixed crown cover threshold when reference area is varied. This also affected the number and mean size of the patches but had little influence on the patch shape. From this findings we concluded that a unified forest definition which includes exact specifications of how the different criteria need to be measured is a prerequisite for forest fragmentation monitoring. An intensive literature review revealed that most of the remote sensing based forest cover assessments lack the implementation of specific forest definitions. Usually they are based on a supervised classification where the train data for classification is selected by an expert without measuring quantitative criteria. Therefore we developed a classification system for remote sensing data implementing a minimum area, a crown cover threshold and a fixed reference area. We tested this method for two study sites in Costa Rica using RapidEye images which resulted in forest / nonforest maps in line with the FAO forest definition. This study gave new insights into the relation of factors inherent in remote sensing based forest cover monitoring and the derived landscape metrics. The findings show how the interpretability of landscape metrics can be improved and which factors hinder a meaningful comparison of metrics from different sites or points in times. Some of the findings are relevant not only for research but have potential for official and commercial use which is reflected by the cooperation with partners from the industry.
Publications
-
(2010). Forest cover monitoring using RapidEye: a case study in Costa Rica, Monographie 3. RESA Nutzerworkshop, Neustrelitz
Magdon, P., Fuchs, H., Fischer, C. & Kleinn, C.
-
(2011). Einsatz von 3D Landschaftsstrukturmetriken zur Erfassung der Waldfragmentierung - Simulationsstudie zur Analyse kritischer Faktoren. IALE-D Jahrestagung 12-14.Oktober 2011, Berlin
Magdon, P. & Kleinn, C.
-
(2011). Uncertainties of Forest Mapping Caused by the Minimum Crown Cover Criterion - A critical scale issue. Conference on Spatial Statistics, 2011, Enschede, The Netherlands
Magdon, P., Kleinn, C., Beckschäfer, P. & Schlather, M.
