North American vegetation model for land‐use planning in a changing climate: a solution to large classification problems

Data points intensively sampling 46 North American biomes were used to predict the geographic distribution of biomes from climate variables using the Random Forests classification tree. Techniques were incorporated to accommodate a large number of classes and to predict the future occurrence of climates beyond the contemporary climatic range of the biomes. Errors of prediction from the statistical model averaged 3.7%, but for individual biomes, ranged from 0% to 21.5%. In validating the ability of the model to identify climates without analogs, 78% of 1528 locations outside North America and 81% of land area of the Caribbean Islands were predicted to have no analogs among the 46 biomes. Biome climates were projected into the future according to low and high greenhouse gas emission scenarios of three General Circulation Models for three periods, the decades surrounding 2030, 2060, and 2090. Prominent in the projections were (1) expansion of climates suitable for the tropical dry deciduous forests of Mexico, (2) expansion of climates typifying desertscrub biomes of western USA and northern Mexico, (3) stability of climates typifying the evergreen-deciduous forests of eastern USA, and (4) northward expansion of climates suited to temperate forests, Great Plains grasslands, and montane forests to the detriment of taiga and tundra climates. Maps indicating either poor agreement among projections or climates without contemporary analogs identify geographic areas where land management programs would be most equivocal. Concentrating efforts and resources where projections are more certain can assure land managers a greater likelihood of success.