Distribution modelling of a century with tree- and forest line changes

Altitudinal tree- and forest lines (TFLs) are two boundaries (but often abbreviated with one word to save space) in the transition zone that separates closed forest from treeless tundra in alpine regions. Due to the last century’s trend of advancing TFLs there is a growing need to understand and predict their distributions. In this study, we aimed to: (1) identify climatic predictors of TFLs in south Norway dominated by mountain birch (Betula pubescens ssp. czerepanovii); (2) analyse elevational changes and estimate distributional changes from 1917 to 2017; and (3) discuss the most likely explanations for the observed changes. The maximum entropy algorithm was used for distribution modelling of past and present TFLs with wall-to-wall coverage of 40 explanatory variables (EVs) with 100x100 m resolution and presence-only data collected in situ from the study area covering 69 000 km2 from 60°26 to 62°43 N and 6°58 to 12°13 E. Stepwise forward selection with the likelihood ratio test for nested models was used to obtain present TFL models with and without topographical variables, evaluated by AUC-ROC and AUC-PR with independently collected evaluation data. Model coefficients were estimated for past TFL models with fixed EVs derived from modelling of present TFLs and evaluated by 4-fold cross-validation. Inverse distance weighting with the elevation of the highest local predictions from past and present TFL models without topographic variables as interpolation attributes was used to obtain interpolated raster layers. Through comparison with a digital elevation model, areas above and below TFLs were identified, and the resulting binary maps were used to estimate changes in distribution. In addition, elevational changes were analysed statistically. We found that: (1) the present treeline distribution was predicted by mean temperature of the warmest quarter, maximum temperature in November, slope inclination and snow water equivalent in March, while mean temperature of the warmest quarter, minimum temperature in November and slope inclination predicted the forest line distribution; (2) TFLs significantly moved upslope from 1917 to 2017 with treelines and forest lines moving on average 0.53 and 0.36 m/year, respectively. The estimated reduction of 6 688 km2 in areas above the treeline (27.6% decline) from 1917 to 2017 was much higher than the estimated 1 137 km2 reduction of areas above the forest line (5.3% decline) but might be affected by the data quality of past TFL observations; (3) the observed changes are most likely a result of climate and land use changes, but it is hard to separate their relative influences. Potential consequences of the observed changes for climate and biodiversity are discussed briefly.