earticle

영어

This study aimed to improve the spatial prediction accuracy of Species Distribution Models (SDMs) by applying and comparing area-weighted sampling and simple random sampling for Korean Red Pine (Pinus densiflora). We employed Random Forest (RF) and Maximum Entropy Model (MaxEnt) as the SDMs and evaluated their performance differences and spatial suitability under each sampling regime. The results showed that simple random sampling improved the model’s generalization ability by uniformly reflecting presence points across the entire study area, thereby ensuring overall spatial homogeneity. Meanwhile, in the case of area-weighted sampling, samples were concentrated in areas with high actual occurrence density, such as the East coast region and southern forest zones. This enhanced spatial concentration and resulted in higher predictive reliability in these high-density regions, but it also led to a relative increase in True Negatives (TN) and False Positives (FP). These findings indicate that sampling alters not only the numerical balance of the data but also the representativeness of the spatial patterns learned by the model. Consequently, the selection of a sampling method requires a strategic choice, contingent on whether the objective prioritizes global predictive stability or precision within core distribution areas.