气候变化领域集成服务门户

Mapping forest properties with supervised remote sensing has historically and increasingly remained vital to research and management efforts, and the demand for such products will only increase as better tools and data increase the usability of such maps. Multispectral imagery by the Landsat program has been an invaluable resource for forest type characterization for several decades. As an alternative to traditional classification approaches dominating these efforts, we instead employed an ordination-regression approach to mapping forest composition as floristic gradients across a similar to 5000-km(2) forestland in southeastern Ohio's Central Hardwoods. Plot data (n = 699 plots; 99 species/genera) from a comprehensive sample of both overstory and understory woody plants across structurally- (open to closed canopy) and topographically-variable forest conditions were projected onto a non-metric multidimensional scaling (NMDS) ordination solution. Floristic gradients, via their ordination scores, were related to spectral reflectance provided by a multitemporal Landsat 8-Operational Land Imager (011) image and various terrain variables using Random Forests models. Approximately 61%, 49%, and 25% of the floristic variation among the three axes of the NMDS ordination were related to the remotely-sensed variables during regression modeling. The axes were predicted onto three images and merged to a RGB color composite for the final floristic gradient map, displaying multivariate vegetation variation across the landscape in terms of variation in color. The color values, by referencing ordination space position within the original solution, provide a statistical approximation of the taxonomic composition of individual forest stands in relation to the plot data. We found this approach highly effective and an attractive alternative to traditional classifications. It is time-efficient, more realistic in that compositional turnover is expressed in continuous fields rather than arbitrary breaks, and less subjective, overcoming the generalization problem inherent in categorizing vegetation assemblages a priori. Moving forward, our model will be a valuable tool in developing suitable management options on individual forest stands for the restoration of desired species, adapting to a changing climate, and improving wildlife habitat in forestlands across the Central Hardwoods.