印尼廖內南部的森林與土地火災具有明顯的季節差異，並受到植被、水分、氣象、土地利用及泥炭地條件等多重因素的共同影響。本研究建立全年、旱季及雨季的野火易發性模型，並評估歷史環境變化指標是否能提供額外的預測資訊。研究整合2010至2024年的多來源衛星與地理空間資料，並將其統一至500公尺網格；燒毀區標籤則由MODIS MCD64A1 BurnDate產品建立。當期條件型與條件變化型兩種預測因子配置，分別使用四種機器學習演算法進行比較。模型採用1:5的火災與非火災樣本比例，以及分層70:30訓練–測試資料切分，並以PR-AUC作為主要評估指標。最終選定的模型為全年與旱季的條件變化型LightGBM模型，以及雨季的當期條件型LightGBM模型；其保留測試集的PR-AUC分別為0.921、0.905及0.966。歷史環境變化指標僅小幅提升全年與旱季模型的表現，且未改善雨季模型，顯示其預測效益具有季節差異。2024年的分析結果顯示，全年與旱季易發性地圖具有較高的空間一致性，而雨季的高易發性區域則較為局部。旱季的高與極高易發性區域面積最大，達8，581平方公里，占分析網格總面積的21.49%。SHAP分析顯示，全年模型的主要影響因子為植被狀況、土壤水分趨勢及降水；旱季模型主要受到降水、植被狀況及油棕土地利用背景的影響；雨季模型則主要受到風速、植被狀況及泥炭地背景的影響。使用2024年印尼國家警察之獨立火災紀錄進行驗證的結果顯示，全年、旱季及雨季分別有77.60%、76.79%及78.57%的匹配紀錄位於高與極高易發性區域。研究結果顯示，分季建模能揭示僅使用全年模型時可能被掩蓋的預測差異與空間差異。;Forest and land fires in Southern Riau， Indonesia， exhibit strong seasonal variation and are influenced by multiple interacting factors， including vegetation， moisture， meteorological conditions， land use， and peatland characteristics. This study developed wildfire susceptibility models for the all-season， dry-season， and rainy-season periods and assessed whether historical environmental-change indicators provided additional predictive information. Multi-source satellite and geospatial data from 2010 to 2024 were harmonized to a 500 m grid， while burned-area labels were derived from the MODIS MCD64A1 BurnDate product. Condition-only and condition-change predictor configurations were compared using four machine-learning algorithms. The modelling design employed a 1:5 fire-to-non-fire sampling ratio and a stratified 70:30 train–test split， with PR-AUC as the primary evaluation metric. The selected models were condition-change LightGBM for the all-season and dry-season periods and condition-only LightGBM for the rainy-season period， with held-out test PR-AUC values of 0.921， 0.905， and 0.966， respectively. Historical environmental-change indicators produced only modest improvements in the all-season and dry-season models and did not improve the rainy-season model， indicating that their predictive benefit was season-dependent. The 2024 results showed stronger spatial agreement between the all-season and dry-season susceptibility maps， whereas high-susceptibility areas during the rainy season were more localized. The dry-season High and Very High classes covered the largest area， reaching 8，581 km²， or 21.49% of the analytical grid extent. SHAP analysis identified vegetation condition， soil-moisture trend， and precipitation as the principal contributors to the all-season model; precipitation， vegetation condition， and oil-palm land-use context as the principal contributors to the dry-season model; and wind speed， vegetation condition， and peatland context as the principal contributors to the rainy-season model. Independent validation using 2024 POLRI fire records showed that 77.60%， 76.79%， and 78.57% of the matched records for the all-season， dry-season， and rainy-season periods， respectively， occurred within the High and Very High susceptibility classes. These findings demonstrate that separate seasonal models can reveal predictive and spatial differences that may be obscured when wildfire susceptibility is represented solely by an all-season model.
