Global change : projecting expansion of invasive species and climate change impacts at the tree-tundra ecotone in the Himalaya

Modeling the distribution of species, especially of invasive species in non-native ranges, has multiple challenges. We develop some novel approaches to species distribution modeling aimed at reducing the influences of these challenges and improve realism of projections. We estimated species-environment relationship with four modeling methods, viz., random forest (RF), boosted regression trees (BRT), generalized linear models (GLM), and generalized additive models (GAM), running each of them with multiple scenarios of (1) sources of occurrences and geographically isolated background ranges, (2) approaches of drawing background points, (3) alternate sets of predictor variables. When a species' distribution is in a non-equilibrium state, as is the case for most invasive species, model projections are very sensitive to the choice of training dataset. Contrary to previous studies, we found that model accuracy is much improved by using a global dataset for model training (both presences and background points from the world), rather than restricting data input to the species' native range. Projections outside the training region, especially in invaded regions, can be very different depending on the modeling method used. Globally projecting, we show that vast stretches of currently uninvaded geographic spaces in multiple continents harbor highly suitable habitats for Parthenium. Projections away from the sampled space (i.e. into areas of potential future invasion), can be very different with different modeling methods, raising questions about the reliability of ensemble projection. Data-driven models that efficiently fit the dominant pattern but exclude highly local features in dataset and model interactions as they appear in data (e.g., boosted regression trees) improve generalization of the species distribution modeling. Alpine treelines are responding to current climate change worldwide. To understand tree line dynamics and its potential drivers, we studied the primary two dominant tree species, Abies spectabilis (AS) and Rhododendron campanulatum (RC), on the north facing slope of two mountains in central Nepal. We determined spatial pattern of regeneration potential, mortality and abundance for various size/age classes, and we identified the most important drivers of such patterns. We also conducted a reciprocal transplant experiment on saplings of RC, moving them between species limit and treeline that were spaced apart by 150m. Young plants (<2m tall) of RC have higher density above treeline than below treeline. Mature plants (>2m tall) of RC, on the contrary, show insignificant trend towards higher density below treeline than above. Mortality of RC was always lower above treeline than below, independent of size class. AS saplings have extremely lower density above treeline than below, with mature plants being virtually absent above treeline. Elevation was identified as the only significant predictor of the decrease in density of both species above treeline. The saplings are progressively younger and shorter with distance above treeline. Both species are regenerating faster above treeline than below. These results are consistent with upward shift of the tree line of RC as a result of recent amelioration of temperature. Climatic extremes during spring affect mortality and leaf size whereas growth is affected by summer climate. Individuals from the species limit, if they survive, perform better when moved downhill than they do at home, and also out-perform the locals. Although the upper elevational boundary of RC is shifting upward, these results indicate that strong differences still exist between individuals across a short elevational gradient, with individuals at the extreme limit of the species range being more tolerant to extreme climate conditions but less tolerant of competition compared to individuals only 150m lower in elevation.