Nearest neighbors mapping of vegetation gradients for landscape analysis and conservation planning

Background/Question/Methods

Nearest neighbors (NN) techniques are not well known in the ecological community, yet they offer several advantages for spatial modeling of vegetation composition and structure, particularly for applications where maps are needed for multiple plant species and where species co-occurrence is important. We present one version of nearest-neighbor imputation that uses constrained ordination (direct gradient analysis) to map plant communities comprised of multiple plant species. We developed detailed maps of forest vegetation for ~50 million ha in the Pacific Northwest, USA. Data from >25,000 field plots were integrated with GIS layers (feature space) describing climate, topography, substrate, latitude, and longitude. Constrained ordination was used to quantify relations between species relative abundance on plots and the spatial predictors. For each map pixel, the k plots that were nearest in multivariate gradient space were identified, and species data collected on the plots were imputed to the pixel. We translated the spatial predictions for multiple individual species into probability surfaces in two ways: (1) using the distance from species centroids in feature space as an inverse probability of presence; and (2) using the species information from the nearest k plots in feature space, weighted by inverse distance, to calculate a probability of presence.

Results/Conclusions

Gradients in community composition were strongly associated with climate and elevation, and less so with topography and soil. Accuracy of the imputation model for presence/absence of 150 species varied widely (kappa 0.00 to 0.80). Omission error rates were higher than commission rates due to low species prevalence, and areal representation of species was only slightly inflated. Area Under the Curve (AUC) diagnostics, which are uncorrelated with prevalence, indicated good model performance for most species. A map of 78 community types was 41% correct and 78% fuzzy correct. Errors of omission and commission were balanced, and areal representation of both rare and abundant communities was accurate. Although map accuracy from NN imputation may be lower for some species than can be achieved with univariate methods, areal representation of species and communities across large landscapes is preserved. Also, because imputed vegetation surfaces are developed for all species simultaneously, map units contain suites of species known to co-occur in nature. Maps of individual species, and of community types derived from them, will be internally consistent at map locations. This is desirable for applications in natural resource management and conservation planning or for models that project landscape change under alternative disturbance or climate scenarios.