Maps of species composition are important for assessing a wide range of ecosystem functions in forested landscapes, including processes shaping community structure at broader (e.g., climate) and finer (e.g., disturbance) scales. Incorporating recently available remotely sensed datasets has the potential to improve species composition mapping by providing information to help predict species presence and relative abundance. Using USDA Forest Service Forest Inventory and Analysis plot data and the gradient nearest neighbor imputation modeling approach in eastern Washington, USA, we developed tree species composition and structure maps based on climate, topography, and two sources of remote sensing: height from digital aerial photogrammetry (DAP) of pushbroom aerial photography and Sentinel-2 multispectral satellite imagery. We tested the accuracy of these maps based on their capacity to predict species occurrence and proportional basal area for 10 coniferous tree species. In this study region, climate, topography, and location explained much of the species occurrence patterns, while both DAP and Sentinel-2 data were also important in predicting species proportional basal area. Overall accuracies for the best species occurrence model were 68–92% and R2 for the proportional basal area was 0.08–0.55. Comparisons of model accuracy with and without remote sensing indicated that adding some combination of DAP metrics and/or Sentinel-2 imagery increased R2 for the proportional basal area by 0.25–0.45, but had minor and sometimes negative effects on model skill and accuracy for species occurrence. Thus, species ranges appear most strongly constrained by environmental gradients, but abundance depends on forest structure, which is often determined by both environment and disturbance history. For example, proportional basal area responses to moisture limitation and canopy height varied by species, likely contributing to regional patterns of species dominance. However, local-scale examples indicated that remotely sensed forest structures representing recent disturbance patterns likely impacted tree community composition. Overall, our results suggest that characterizing geospatial patterns in tree communities across large landscapes may require not only environmental factors like climate and topography, but also information on forest structure provided by remote sensing.
Read full abstract