진주시 소나무재선충병 발생 예측 모형 구축 및의사결정나무 기반 구조적 해석

초록

This study aimed to develop a predictive model for pine wilt disease occurrence in Jinju-si, Gyeongsangnam-do, Korea, and to identify structural rules governing disease occurrence using a decision tree approach. Surveillance records from 2017 to 2022 and control project records from 2020 to 2022 were converted into spatial data to construct a dataset consisting of 132,634 points, including 66,317 infected and 66,317 non-infected locations. Six classification models (decision tree, Bayesian network, artificial neural network, logistic regression, bagging, and random forest) were applied to compare predictive performance. Among them, the random forest model showed the highest classification accuracy of 88.9%. To obtain interpretable rules and support management decision-making, occurrence rules were derived using the decision tree model. The results showed that elevation, proximity to firewood boilers, proximity to residential areas, aspect, and slope were the major variables associated with pine wilt disease occurrence. A higher probability of occurrence was observed in low-elevation areas (≤150 m), south-facing slopes, and forests adjacent to residential areas. In contrast, areas located more than 3.3 km from firewood boilers or above 179 m in elevation showed a relatively higher probability of non-occurrence. By using actual local surveillance and control records, this study identified the spatial characteristics and key influencing factors of pine wilt disease occurrence. The rule-based results derived from the decision tree provide interpretable criteria for prioritizing control resources and establishing spatially differentiated management strategies.

키워드