Assessing Uncertainties in Predicting the Changes in Forest Species Distributions caused by the Climate Change

Abstract

The adverse impacts of climate change on forest ecosystems are expected to increase, and various measures are being proposed to reduce them. To mitigate the negative impacts of climate change with limited time and resources, and to respond effectively, it is necessary to make an accurate impact assessment based on climate change. To do so, it is necessary to understand and quantify the uncertainties that are inevitable in climate change impact assessment.

The concept of uncertainty, which has been mentioned since the fourth report of the Intergovernmental Panel on Climate Change, is specified in the Fifth Impact Assessment Report and is used as a concept to aid decision making. In Korea, efforts are being made to quantify uncertainties in assessing the impacts of climate change. However, these studies are still in the early stages, are limited in scope, and do not consider uncertainties in various aspects.

Therefore, in this study, we analyzed the causes of uncertainties that may occur due to climate change by: 1) measuring the effectiveness of sampling methods and sample size, 2) evaluating the uncertainties in model performance and spatial distribution due to Species Distribution Model (SDM) algorithms, and 3) considering the uncertainties if involved in assuming competition among major species, applying four Representative Concentration Pathway (RCP) scenarios to potential distribution ranges.

To measure the effectiveness of the sampling methods, three sampling methods and seven different sample sizes were considered for the one-way t-test. As a result of the one-way t-test, stratified random sampling methods are shown to well represent the population. In addition, if the sample size exceeds a certain number, for this study, 200 samples, the performance of SDMs does not significantly increases.

We applied eight SDMs that were either statistically based or machine learning based algorithms to model the potential distribution of major species in Korea

the performance of the models differed according to the algorithms. To test the performance of SDMs, the area under the curve (AUC) value and True Skilled Statistics (TSS) value were applied. Machine learning models, especially the random forest (RF) model showed excellent performance while statistical based models (Generalized Linear Model (GLM), Generalized Additive Model (GAM)) showed average performance. When we verified uncertainties in spatial distribution, with thresholds matching the current area of the major species, the uncertainties in the spatial distribution was significant. Ensemble methods need to be applied to minimize uncertainties in the spatial distribution of SDMs.

To consider uncertainties in the competition among major species, the random forest algorithm and Global Agro-Ecological Zones (GAEZ) classification were applied. Modeling results revealed that the multi-species model included higher uncertainties. However, single species models can not include the climate zone changes that we expect in RCP the scenarios. Thus, we need to include the potential introduction of forest species that are suitable in different climate zones.

Through this study, when we establish management strategy for climate change mitigation and adaptation, uncertainties in each step if we predict potential distribution of forest species can be applied to prioritize management target. This can reduce uncertainties in management strategy as well as find effective monitoring points for counteract adverse changes due to the climate change.