Potential climate change effects on tree distributions in the Korean Peninsula: Understanding model & climate uncertainties

- Uncertainty about the SDM projections is a major challenge to reliable projections.
- Uncertainty derives mainly from model and climate uncertainties.
- The accuracies of Silver Magnolia projections showed a high variability among SDMs.
- Future projections of Silver Magnolia considerably varied according to GCMs and RCPs.
- Understanding of such uncertainties is critical for effective conservation planning.

The projections of species distribution models (SDMs) have provided critical knowledge for conservation planning under climate change in the Republic of Korea. However, uncertainty about the SDM projections has been criticized as a major challenge to reliable projections. The present research investigated uncertainty among competing models (Model uncertainty) and uncertainty of future climate conditions (climate uncertainty) driving from different GCMs and CO2 emission scenarios in predicting the future distributions of plants. For this purpose, using nine single-model algorithms and the pre-evaluation weighted ensemble method, we modeled the geographical distributions of Silver Magnolia (Machilus thunbergii Siebold & Zucc.), a warm-adapted evergreen broadleaved tree; furthermore, we predicted its future distributions under 20 climate change scenarios (5 global circulation models (GCMs) × 4 CO2 emission scenarios (RCPs)). The results showed a great variation in the accuracies of nine single-model projections: the mean AUC values of nine single-models ranged from 0.764 (SER) to 0.970 (RF), and the mean TSS ranged from 0.529 (SRE) to 0.852 (RF). RF (mean AUC = 0.970, mean TSS = 0.852) and the ensemble forecast (AUC = 0.968, TSS = 0.804) showed the highest predictive power, while SRE showed the lowest. The future distributions of Silver Magnolia projected with the ensemble SDM clearly varied according to GCMs and RCPs. The twenty climate scenarios produced twenty different projections of the magnolia prospective distribution. GCMs commonly projected the maximum range expansion under RCP 8.5 in 2050 and 2070, but CO2 emission scenarios explaining the minimum expansions differed according to GCMs. In conclusion, our results show that GCMs, CO2 emission scenarios and SDM algorithms produce considerable variations in the SDM projections. Therefore, this research suggests that understanding of model and climate uncertainties is critical for an effective conservation planning in forest management under climate change on the Korean Peninsula.