Capturing uncertainties in earthquake forecasts

This dissertation investigates the uncertainties in seismicity rate estimates and their implications for Probabilistic Seismic Hazard Assessment (PSHA), focusing on regions with limited earthquake data. Through a series of cumulative studies, this research addresses key aspects of earthquake forecasting models, testing frameworks, and the integration of non-Poissonian rate distributions into hazard models. First, we evaluate a decade-long earthquake forecasting experiment in Italy, demonstrating the importance of long-term earthquake catalogs and fault data in improving forecast performance. The study also highlights the limitations of Poisson-based models, suggesting that negative binomial distributions provide a better fit for the observed seismicity rates. Following, we compare regional and global seismicity models, revealing that global models, which incorporate geodetic strain-rate data, often perform as well as or better than regionally calibrated models. This finding underscores the value of combining diverse data sources in earthquake forecasting. To improve the forecast testing existing framework, we introduce an open, reproducible framework for earthquake forecasting experiments, named Floating Experiment, facilitated by the floatCSEP software. This new experiment format enhances transparency, standardizes procedures, and allows for the decentralized evaluation of forecasting models, fostering continuous improvement and benchmarking. Then, we address the challenge of forecasting seismicity in low-seismicity regions by developing a method that captures spatio-temporal uncertainties using uniform rate zones derived from strain-rate fields. The study demonstrates that non-Poissonian models can better account for the variability in seismicity rates, particularly in regions with sparse earthquake records. Finally, we implement these non-Poissonian forecasts into the New Zealand National Seismic Hazard Model, showing that incorporating rate variability significantly impacts hazard estimates. The results highlight the need for PSHA models to move beyond the traditional Poissonian assumptions to accurately reflect seismic hazard, especially in low-seismicity areas. The findings of this dissertation suggest that current PSHA practices may underestimate seismic hazard in low-seismicity regions due to inadequate temporal and spatial modeling of seismicity rates. By integrating more comprehensive statistical descriptions and leveraging geodetic data, this research provides a pathway for improving earthquake forecasting and hazard assessment methodologies. Future research should focus on testing these methods in diverse tectonic settings, refining the relationship between strain rates and seismicity, and developing time-dependent hazard models. The continued use and development of the Floating Experiment framework will be crucial for advancing the field of earthquake predictability and enhancing the resilience of societies to seismic hazards.