Combining remotely sensed spectral data and digital surface models for fine-scale modelling of mire ecosystems

Küchler, M. and Ecker, K. and Feldmeyer-Christe, E. and Graf, U. and Küchler, H. and Waser, L.T. (2004) Combining remotely sensed spectral data and digital surface models for fine-scale modelling of mire ecosystems. Community Ecology, 5 (1). pp. 55-68. ISSN 1585-8553

Preview

Text comec.5.2004.1.6.pdf Download (775kB) | Preview

Abstract

The detection and evaluation of changes in vegetation patterns is a prerequisite for monitoring programs. The Swiss mire monitoring program aims to assess the changes in mire vegetation in order to examine the efficiency of the management measures. A promising way to explore and detect vegetation structure and vegetation change is the application of predictive vegetation mapping that combines image classification and predictive habitat distribution models. These models deal with predictor variables derived from remotely sensed spectral data and from environmental variables such as a digital surface model (DSM). Low accuracy of environmental data to predict vegetation at the local scale is due to the difficulties to capture dominant fine-scale enironmental gradients. Using high resolution spectral and topographical data sets of 50 cm pixel size and below, the study presented here aims to improve the simulation of local-scale vegetation properties. The spectral data for fine-scale modelling are based on CIR orthoimages with a ground resolution of 32 cm. Various spectral variables and spectral-textural variables were derived for the modelling process. A new method to reduce the number of predictor variables, the composite modelling is presented in this paper. In comparison to existing methods, composite modelling has the advantage of being independent of the scale of the predictor variables, and at the same time being transferable among various data sets. Mean indicator values for moisture, nutrients and light derived from vegetation data are used as response variables. Results show that the topographical variables based on relief features are less powerful predictors than the spectral variables but that combining them enhances the overall predictive power. Stratification of the data according to the tree layer and the shadow areas increases the accuracy of the model.

Actions (login required)

Edit Item