Estimates of the percentage of moderate to large crustal earthquakes (mainshocks) that have foreshocks (the foreshock rate) vary widely: recent estimates in Southern California using an enhanced catalog range between 19 and 72%. Enhanced catalogs seem to reveal more foreshocks, possibly providing new constraints on nucleation mechanisms, but precise, commonly-accepted foreshock definitions are lacking. To investigate the observed range we quantify the sensitivity of foreshock rates to mainshock selection method, catalog (standard and enhanced), foreshock definition, geographical restriction and magnitude cut-offs. We compare two foreshock definitions: type A - any earthquakes above a magnitude threshold in a space-time window; and type B - an earthquake count in a space-time window that exceeds the 99th percentile of a statistical representation of past seismicity rates (using three distributions: Poisson, Gamma and Empirical). Foreshock rate estimates are increased by (in order of influence): Poisson distribution, type A definition, fixed mainshock selection, and restricting to mainshocks with minimum background rates or spatial completeness magnitudes. Rates are lowered by: magnitude-dependent mainshock selection, Gamma and Empirical distributions, and applying a magnitude cut-off. A large increase in foreshock rate between the standard and enhanced catalog is only observed when using Poisson distributed background rates for type B foreshocks. A lower magnitude of completeness may thus not lead to significantly more mainshocks with detected foreshocks. Our preferred method, using a more robust mainshock selection and quality-controlled data, estimates ~25% of M4+ “mainshocks” in Southern California have foreshocks.

Supervised deep learning models have become a popular choice for seismic phase arrival detection. However, they don’t always perform well on out-of-distribution data and require large training sets to aid generalization and prevent overfitting. This can present issues when using these models in new monitoring settings. In this work, we develop a deep learning model for automating phase arrival detection at Nabro volcano using a limited amount of training data (2498 event waveforms recorded over 35 days) through a process known as transfer learning. We use the feature extraction layers of an existing, extensively-trained seismic phase picking model to form the base of a new all-convolutional model, which we call U-GPD. We demonstrate that transfer learning reduces overfitting and model error relative to training the same model from scratch, particularly for small training sets (e.g., 500 waveforms). The new U-GPD model achieves greater classification accuracy and smaller arrival time residuals than off-the-shelf applications of two existing, extensively-trained baseline models for a test set of 800 event and noise waveforms from Nabro volcano. When applied to 14 months of continuous Nabro data, the new U-GPD model detects 31,387 events with at least four P-wave arrivals and one S-wave arrival, which is more than the original base model (26,808 events) and our existing manual catalogue (2,926 events), with smaller location errors. The new model is also more efficient when applied as a sliding window, processing 14 months of data from 7 stations in less than 4 hours on a single GPU.

We present the Malawi Active Fault Database (MAFD), a geospatial database of 114 active fault traces in Malawi, and in neighboring Tanzania and Mozambique. The MAFD has been developed from a multidisciplinary dataset: high resolution digital elevation models, field observations, aeromagnetic and gravity data, and seismic reflection surveys from offshore Lake Malawi. Active faults longer than 50 km are found throughout Malawi, where seismic risk is increasing due to its rapidly growing population and its seismically vulnerable building stock. The MAFD also provides an opportunity to investigate the population of normal faults in an incipient continental rift. We find that the null hypothesis that the distribution of fault lengths in the MAFD is described by a power law cannot be rejected. Furthermore, a power-law distribution of faults in Malawi is consistent with its thick seismogenic crust (~35 km), and low (<8%) regional extensional strain that is predominantly (50-75%) accommodated across relatively long hard-linked border faults. Cumulatively, the data and inferences drawn from the MAFD highlight the importance of integrating onshore and offshore geological and geophysical data to develop active fault databases along the East African Rift and similar continental settings, both to understand the regional seismic hazard and tectonic evolution.