Combining satellite imagery and ancillary data to map snowbed vegetation important to reindeer Rangifer tarandus

Remote sensing provides a viable alternative for mapping vegetation in the Arctic because it allows for the mapping of discontinuous distribution of cover types over different spatial scales. In this paper we present a statistical method to map the distribution of important cover types for the reindeer Rangifer tarandus during summer in northernmost Sweden using IRS 1D-LISS satellite imagery. We exemplify our method with modeling of the distribution of snowbed vegetation, the cover type used most intensively by the reindeer in the study area. An autologistic regression model that incorporates the spatial structure of the data is used to combine the field data and the satellite image data. The terrain effects in the satellite image are accounted for in the regressions using a digital elevation model (DEM). We produced a fine-scaled coverage depicting the probability of occurrence of snowbed vegetation as a continuous variable at the pixel level. The accuracy of mapping snowbed vegetation was 69-77%, depending on the data used. We conclude that small-scale, pixel-wise classification modeling may be useful for depicting sparsely occurring cover types, some of which may be important determinants of range quality for reindeer.