Probabilistic Seismic Forecasting for Sustainable Urban Preparedness Around the Lembang Fault, Indonesia: Catalog Audit, Bayesian Spatio-Temporal ETAS, and Forecast-Informed Prioritization for Bandung City

Probabilistic earthquake forecasts can support urban preparedness only when catalog quality, parameter uncertainty, predictive performance, and the limits of interpretation are considered together. We examine the Lembang Fault in West Java by linking a regional Bayesian ETAS analysis with a city-scale preparedness-prioritization assessment for Bandung City. The audited catalog within 50 km of the study area contains 1983 earthquakes, including 1797 shallow events at depths of 30 km or less. Because the catalog combines several reported magnitude types, only 114 events could be confidently expressed on a consistent moment-magnitude scale (Mw). Rather than forcing uncertain magnitude conversions, the primary 2018–2024 analysis retains the original reported catalog magnitudes, denoted here as Mcat, and includes 562 shallow events above a conservative working threshold (M0 = 2.5). A smaller subset of 36 events that can be expressed consistently as moment magnitude (Mw ≥ 3.4) is used for a sensitivity analysis. Four MCMC chains showed satisfactory numerical convergence. The long-term asymptotic branching diagnostic had a median of 7.09, whereas the finite-observation-window reproduction diagnostic had a median of 0.845, supporting interpretation of the model for conditional short-horizon forecasting rather than long-term stationary behavior. The primary model predicts a median of six earthquakes over 30 days, with a 95% predictive interval of 0–23 events and a 96.3% probability of at least one event. In an out-of-sample test for 2024, the model substantially overpredicts the central event count, with a median prediction of 149 events compared with 72 observed, although its spatial forecast performs better than a uniform spatial benchmark. For Bandung City, the 30-day seismicity forecast (F) is combined with 2024 population exposure (E), a provisional socioeconomic vulnerability component (V), and critical-health-facility exposure (C) in an equal-weight Dynamic Preparedness-Priority Index (DPPI). Bojongloa Kidul, Coblong, and Babakan Ciparay have the highest baseline DPPI values. DPPI is intended to identify relative preparedness priorities and should not be interpreted as an estimate of absolute seismic risk, expected damage, or economic loss. Overall, the framework provides a transparent connection between probabilistic seismicity forecasting and urban preparedness planning while keeping uncertainty, data limitations, and model boundaries explicit.