Empirical forecasts of the occurrence of earthquakes in spaceÁlvaro González Recent examples of poor performance of earthquake hazard maps highlight the necessity of improved, tested models for the spatial distribution of earthquakes. This contribution will present forecasting maps expressing the probability of occurrence of earthquakes throughout different regions. The maps were first tested retrospectively with several tens of thousands of earthquakes which occurred worldwide and in Southern California. They are currently subjected to global and regional, daily, real-time testing within the Collaboratory for the Study of Earthquake Predictability (www.cseptesting.org). These spatial forecasts are based on the fact that earthquakes tend to occur close to previous ones. So the calculated probabilities are relatively high at the sites of past earthquakes, and decrease with distance from them. Other models of spatial occurrence of earthquakes depend on assumptions: They usually rely on zoning the analyzed region into supposedly homogeneous seismotectonic zones, or on smoothing the observed spatial distribution of seismicity using a theoretical function (smoothing kernel). Here, instead, the procedure relies only on the spatial locations of past epicenters and the empirical distribution of distances between them. It is purely empirical, non-parametric, and does not depend on any theoretical assumption. The results show that the spatial distribution of past seismicity is a very good predictor of the distribution of future seismicity. The forecast map may be updated every time a new earthquake happens, thereby improving its detail and forecasting skill. A remarkable outcome is that it is possible to anticipate in which areas the next earthquake may happen with a specified probability. Those areas can be further refined if temporal correlations are taken into account. Namely, a modified procedure incorporates the fact that the empirical probabilities rise sharply around the location of the latest earthquake, and fade with distance from it. The map elaborated with this method achieved, for example, a significantly improved spatial forecast of the Tohoku earthquake and its aftershocks in the real-time test. As limitations of these procedures, they are purely spatial (do no inform about the timing or maximum magnitude of the future earthquakes) and probabilistic (their performance is evaluated using a long series of events, while they may fail on specific cases). Conversely, because of their simplicity and robustness, these methods may be used as baselines to evaluate the performance of more complex forecasts, or combined with ground motion models to elaborate future hazard maps. |
