Extraction of forest parameters in a mire biotope using high-resolution digital surface models and airborne imagery L.T. Waser, C. Ginzler , M. Kuechle, P. Thee Research Unit Land Resource Assessment, Swiss Federal Research Institute WSL, Birmensdorf, Switzerland (waser, ginzler, kuechler, thee)@wsl.ch E. Baltsavias , H. Eisenbeiss Institute of Geodesy and Photogrammetry, ETH Zurich, Wolfgang-Paulistr. 15, 8093 Zurich, Switzerland; (manos, ehenri) @geod.baug.ethz.ch Abstract— The objective of this paper is to spatially predict tree/shrub genera using generalized linear models (GLM), color- infrared (CIR) aerial images, ADS40 images, digital surface models (DSM)s and field samples. The present study was carried out in the framework of the Swiss Mire Protection Program, where extraction of forest parameters for description of present state of a mire ecosystem and as indicators for changes are of high importance. In a first step, high-quality DSMs were automatically generated from CIR aerial images for two test sites, both located in the Pre-alpine zone of Central Switzerland. In a second step, tree layers were then generated combining canopy height models derived from the DSMs and LiDAR DTM with a fuzzy classification of CIR aerial images. In a third step, on the basis of these tree layers, fractional tree/shrub covers were generated using explanatory variables derived from the DSMs and logistic regression models. Then tree genera were predicted for the pixel values (tree/shrub probability > 0.3) of the fractional covers using a multinomial regression model and additional spectral information as provided by Leica ADS40 data for one test site and CIR aerial images for the other test site. Overall, prediction of tree genera was less satisfactory when only using CIR aerial images. In contrary, up to six different tree genera were predicted with high accuracy using explanatory variables derived from ADS40 images. The study stresses the importance of high-resolution and high-quality DSMs and highlights the potential airborne remotes sensing data for ecological modeling purposes. Tree genera; fractional tree/shrub cover; stand composition; environmental modeling; multi-image matching; LiDAR; CIR aerial images; Leica ADS40 image; forest classification; Swiss Mire Protection Program I. INTRODUCTION This paper focuses on modeling fractional tree/shrub cover and prediction of tree/shrub genera in two mire ecosystems. The study was carried out in the framework of the Swiss Mire Protection Program which aims at conserving mire ecosystems of national importance and outstanding beauty in their present state. This implies no decrease of the mire area and no degradation of vegetation. A monitoring program based on a representative sample of 130 mires was set up in 1996 to examine the effectiveness of the conservation status [1, 2]. The monitoring also implies an assessment of shrub encroachment and increase of forest area exerting vehement impact on the non-forest areas of the mire biotopes. Shrub encroachment is a considerable danger for the biotope and accelerates a degradation of the mire area. Extraction of various forest related parameters (e.g. exact forest/shrub area, canopy height, trees/shrubs in open mire land, tree/shrub genera etc.) is essential to assess magnitude and consequences of this impact. However, such information is often difficult to acquire across mire ecosystems using traditional methods of field survey and aerial photograph interpretation. It is well known that obtaining different forest parameters, such as tree heights or stand composition through ground measurements or vegetation mapping is often not feasible in dense forest, too costly in terms of time and manpower, and also prone to errors [3]. Costs of forest sampling can be reduced substantially by estimating forest and tree parameters directly from high- resolution remotely sensed data. Recent progress in three dimensional remote sensing mainly includes digital stereophotogrammetry, radar interferometry and LIDAR [4]. E.g. by subtracting DTM from the DSM canopy height models can be calculated. DSM can be obtained by means of photogrammetric methods or by LIDAR. Using digital photogrammetry, DSMs are based on ATE algorithms with image correlation. This method is widely used and has proven to provide both, reliable and accurate results [5]. DTMs have been derived from manual photogrammetric or terrestrial measurements for a long time already [6]. Meanwhile several LIDAR systems are commercially available [7], enabling the derivation of DTMs from such data as well. A number of studies reveal the successful application of these methods to assess tree [8] and stand attributes (stand composition, tree height, crown diameter, basal area, and stem volume). Combining some of these attributes can be useful to evaluate growth estimations (including extent of forest area), to detect changes in the forest stands, and determination of tree/shrub genera [9, 10]. There is a growing need for sensitive tools to predict spatial and temporal patterns of plant species or communities [11]. Spatially explicit predictive modeling of vegetation is often used to construct current vegetation cover using information