Predicting Future Plant Species Occurrence Using Paleo- and Modern Climate Data with Machine Learning

Since the onset of industrialization, accelerated global warming has significantly impacted global ecosystems, particularly affecting the survival and distribution of plant species. Rising temperatures and increasing climate variability pose substantial threats to the structural and functional stability of ecosystems, interspecies interactions, and the geographical distribution of plant taxa. These changes may ultimately lead to reduced biodiversity and potential species extinction. In response, predicting future vegetation dynamics has become essential for biodiversity conservation and human sustainability. This study aims to analyze the occurrence trends of plant species using machine learning techniques, based on past and present climate data, to project future distributions under climate change.

Temperature and precipitation data from the Gwangyang region in South Korea were used as input variables. The target variable, representing species presence or absence, was derived from pollen and environmental DNA (eDNA) analyses of soil samples collected from the Topyeongcheon area. Separate logistic regression models were constructed for three representative plant species. The statistical significance of regression coefficients was examined, and model performance was evaluated using accuracy, precision, recall, and F1-score metrics.

Future projections were based on climate scenarios from the CMIP6 dataset, incorporating a range of socio-economic factors including demographics, economic development, welfare levels, ecosystem conditions, resource availability, technological advancements, social dynamics, and policy interventions. These scenarios were applied to the models to estimate future plant species occurrence and to conduct comparative analyses across different pathways. The findings of this study provide a quantitative assessment of species distribution under climate uncertainty and offer a foundational basis for long-term biodiversity conservation planning.