Modelling forest habitat of zoological communities - Which sensors do we actually need?

Kurzfassung

"Modelling forest habitat types is of crucial importance for resource and conservation management on landscape scales. During the past decades substantial progress has been made in the development of remote sensing based forest type mapping, for which a multitude of different sensor systems have been employed. Yet, researchers frequently face the difficult decision which remote sensing data to acquire for a particular modelling task. Apart from specific sensor to sensor comparisons there is, however, no general understanding on how fundamental sensor characteristics such as spectral and spatial resolution interplay in determining mapping success. Based on airborne hyperspectral data we developed a general modelling framework for the identification of both sufficient and optimal sensor characteristics for a given modelling task with respect to spectral and spatial resolution. To this end, we simulate a spectral information gradient by means of a novel spectral clustering algorithm and perform spatial upscaling using general point-spread functions. The focus of our study lays on mapping habitat for different zoological communities in montane mixed forest stands in the Bavarian Forest national park, Germany. Habitats are defined for continuous community gradients obtained by means of non-metric multidimensional scaling of zoological ground survey data and are subsequently modelled using random forest regression with remote sensing data serving as predictors. A major result of direct practical relevance is the observed complementarity between spatial and spectral resolution, e.g. cases in which higher spectral resolution can make up for a less suitable spatial resolution. Moreover, this approach allows us to estimate the expected increase in model accuracy if transitioning from less suitable data to more suitable data. Besides providing recommendations on suitable sensor choice and sensor set-ups the proposed framework is a valuable tool to support the definition of sensor requirements for future missions in an objective, data-driven way."