Abstract: | Abstract. Vegetation maps serve as the basis for spatial analysis of forest ecosystems and provide initial information for simulations of forest landscape change. Because of the limitations of current remote sensing technology, it is not possible to directly measure forest understory attributes across large spatial extents. Instead we used a predictive vegetation mapping approach to model Tsuga heterophylla and Picea sitchensis seedling patterns in a 3900‐ha landscape in the Oregon Coast Range, USA, as a function of Landsat TM imagery, aerial photographs, digital elevation models, and stream maps. Because the models explained only moderate amounts of variability (R2 values of 0.24–0.56), we interpreted the predicted patterns as qualitative spatial trends rather than precise maps. P. sitchensis seedling patterns were tightly linked to the riparian network, with highest densities in coastal riparian areas. T. heterophylla seedlings exhibited complex patterns related to topography and overstory forest cover, and were also spatially clustered around patches of old‐growth forest. We hypothesize that the old growth served as refugia for this fire‐sensitive species following wildfires in the late 19th and early 20th centuries. Low levels of T. heterophylla regeneration in hardwood‐dominated forests suggest that these patches may succeed to shrublands rather than to conifer forest. Predictive models of seedling patterns could be developed for other landscapes where georeferenced inventory plots, remote sensing data, digital elevation models, and climate maps are available. |