Mapping gradients of community composition with nearest-neighbour imputation: extending plot data for landscape analysis

摘要: Question: How can nearest-neighbour (NN) imputation be used to develop maps of multiple species and plant communities? Location: Western central Oregon, USA, but methods are applicable anywhere. Methods: We demonstrate NN by mapping woody communities for >100 000 km2 diverse forests woodlands. Species abundances on ∼25 000 plots were related spatial predictors (rasters) describing climate, topography, soil geographic location using constrained ordination (CCA). data from the nearest plot in multi-dimensional CCA space imputed each map pixel. Maps individual community types constructed single surface. computed a variety diagnostics characterize different qualities (mapped) data. Results: Community composition gradients strongly associated with climate elevation, less so topography soil. Accuracy model presence/absence 150 varied widely (kappa 0.00 0.80). Omission error rates higher than commission due low prevalence, areal representation was only slightly inflated. A 78 41% correct 78% fuzzy correct. Errors omission balanced, both rare abundant accurate. Map accuracy may lower some other methods, across large landscapes is preserved. Because vegetation surfaces developed all simultaneously, units contain suites known co-occur nature. species, derived them, will internally consistent at locations. Conclusions: useful modelling approach where needed, such as natural resource management conservation planning or models that project landscape change under alternative disturbance scenarios. More research needed evaluate communities.