Publication

Abstract : As cities evolve into intelligent cyber-physical ecosystems, infrastructure resilience and disaster preparedness are critical. Earthquake forecasting remains a challenging task, requiring advanced machine learning and non-seismic data integration. This study proposes a hybrid SSA-LSTM model that leverages satellite-derived Outgoing Longwave Radiation (OLR) as a seismic precursor. Singular Spectrum Analysis (SSA) is employed to decompose noisy OLR signals and extract dominant temporal features, which are then learned by a Long Short-Term Memory (LSTM) network. The proposed model is evaluated against a conventional LSTM using identical historical seismic and OLR datasets. Experimental results demonstrate that SSA-LSTM consistently outperforms the standard LSTM, achieving lower forecasting errors and improved robustness. Statistical analysis confirms the effectiveness of SSA in enhancing long-term dependency learning. The findings highlight the potential of combining signal decomposition and deep learning for reliable earthquake forecasting.