<![CDATA[Indonesia's escalating hydrometeorological disaster frequency demands robust predictive frameworks capable of capturing non-stationary climate dynamics. This study aimed to analyze statistical correlations among disaster types and generate long-term flood frequency projections using advanced computational methods applied to national disaster data. Pearson correlation analysis was first conducted to quantify inter-disaster relationships, revealing strong associations between extreme weather, floods, and landslides (r = 0.79–0.86), alongside inverse relationships with drought. The Pruned Exact Linear Time (PELT) algorithm subsequently identified three significant regime shifts in 2012, 2017, and 2022, confirming the progressive non-stationarity of Indonesia's disaster patterns. A Long Short-Term Memory (LSTM) deep learning model was then trained on these regime-structured data to generate predictive forecasts. The model achieved high directional accuracy, successfully capturing the 2025 peak and 2026 decline, with an RMSE of 816.67 and MAPE of 43.77%. Projections for 2027–2029 estimate flood events reaching 2,278, 2,542, and 2,021 incidents respectively, indicating a sustained high-frequency disaster regime that necessitates urgent adaptive infrastructure and evidence-based climate resilience planning.]]>
Copyrights © 2026
