Earthquake Forecasting

We are still a long way off being able to predict earthquakes. In fact, it may never be possible to say when and where an earthquake will occur. Earthquake forecasting, where we give probabilities of earthquakes occurring in a given space/time window, is becoming more accepted. This acceptance is due in part to the efforts of those people at CSEP, a global initiative to objectively evaluate forecast models. On the left you can view two kinds of forecasts, a yearly and a daily forecast, that I have developed in collaboration with some colleagues at Kyoto University. They are provided purely for the interested individual, and we take no responsibility for the accuracy of the forecasts. Don't assume that you are safe from the damaging effects of earthquakes because you live in an area indicated by these models as only having a low probability of an earthquake. More information about the models is given below.

Yearly Forecast of Japan

This forecast has been submitted to the Earthquake Forecast System based on the Seismicity of Japan experiment. Many models have been submitted to this experiment, and the testing center is trying to establish which method best predicts the seismicity of Japan. The picture shows the yearly probability of a magnitude five or larger earthquake throughout Japan. The details can be found in the manuscript submitted to Earth, Planets, Space. Briefly, we assume that earthquakes are likely to occur where they have occured in the past, and recent seismicity changes are indicative of future seismic activity.

Hourly Forecast of Japan

I have looked back through old catalog data, and found that on average five percent of the time an earthquake is followed by a larger earthquake within five days and 10 kilometers. The probability changes for different areas, say offshore and onshore, but for simplicity we use the general five percent. The probability is highest in the hour following the earthquake, and decreases with time. The probability also decreases for larger magnitude earthquakes. For example, the probability that a magnitude three earthquake is followed by a magnitude five earthquake is much smaller than the probability of a magnitude five earthquake being followed by a magnitude 5.1 earthquake. We use these probabilities and generate an hourly forecast for magnitude five or larger earthquakes. Areas close to a recent earthquake are assigned higher probabilities based on when the last earthquake occured and how big it was. I try and update the forecast hourly, to use the most recent seismicity. The previous hour's forecast is also available. The method is based almost entirely on previous research publised in Jones (Foreshocks and Time-Dependent Earthquake Hazard Assessment in Southern California, BSSA, 76 (6), 1985). We gratefully acknowledge the National Research Institute for Earth Science and Disaster Prevention, who provide the earthquake locations via the High Sensitivity Seismograph Network Japan (Hi-net). These preliminary locations are used in the formation of these forecasts.

B-values of the Gutenberg Richter law

Recently, there has been increased interest if the b-value of the Gutenberg Richter law decreases at a location before a large earthquake (such as the Tohoku earthquake in 2011). We are interested in verifying this claim. We plot recent b-values for over 700 locations throughout Japan, and allow the reader to view the historical changes in these values. Please click here for the map and more information.