Sequence-specific updating of European ETAS model: Application to the 2023 Türkiye-Syria earthquake sequence

We analyse the spatio-temporal evolution of the aftershock sequence to the 2023 M7.8 Türkiye-Syria earthquake. Recently, we have calibrated a generic ETAS-based operational forecasting model for Europe, using the unified earthquake catalog developed within the European Seismic Hazard Model (ESHM20; Danciu et al., 2022) for data between 1990 and 2015. Focusing on the earthquake sequence that started in February 2023 in Türkiye, we analyse how our model would have forecasted the temporal and spatial evolution of the sequence. We observed that the productivity of the sequence is substantially higher than forecasted by our generic model. Similar observations have been made in earlier studies on other sequences, and strategies have been proposed to improve existing models based on sequence-specific data (e.g., Omi et al., 2015). Therefore, we conclude that sequence-specific updating is required to achieve an acceptable fit between model and observations.

Here, we investigate the best way to visualize the results of aftershock forecasting and operational earthquake forecasting, and propose a new strategy for sequence-specific updating of model parameters to accurately describe the productivity and the spatial aftershock distribution, while leveraging on the parameters obtained from larger amounts of data within the European model. Our approach strives to avoid biases in the description of the temporal decay due to relying on short-term data. This is done by keeping the parameters describing the temporal decay fixed to the values inverted with our baseline model and calibrating the remaining parameters, using data of the ongoing sequence. As an alternative way to better control productivity, we test model variants for which the a value is fixed to be equal to the GR law exponent b, as proposed by Hainzl et al. (2008). The variants with both fixed and calibrated temporal kernel and productivity are fitted to varying time periods of the Turkish sequence.

We assess the model’s consistency with observations by comparing the forecasts issued by the basic and modified models to the observed events. Preliminary results suggest that keeping the temporal kernel and the productivity parameter a fixed provides better forecasts than the baseline model, already when small amounts of data from the sequence are available. Having identified a promising strategy for sequence-specific model updating, we plan to test whether it is systematically successful by applying it to all earthquake sequences in Europe that occurred after the end of the baseline model training period in 2015. Moreover, we will develop prototypes of communication products that should support professional, societal stakeholders (e.g., decision makers, first responders) to take informed decisions, for example during rescue investigations. Thereby, we will follow evidence-based recommendations derived from the research efforts in the European Horizon-2020 project RISE (Freeman et al., 2023).