Effect on the predictability of pattern informatics method related to selection of studied regions

The Pattern Informatics （PI） method is a approach for earthquake forecasting based on statistical physics, and has been widely applied both at home and abroad due to its good performance in medium to long term earthquake forecasting. The algorithm of PI method includes the process of normalization of all grid parameters in the selected region, so the distribution of PI hotspots might be different with the different selected studied regions theoretically. However, the predictability of PI due to the selection of studied regions has not been systematically studied so far. We performed the retrospective forecasting for seven earthquakes above M_S6.0 in the North-South Seismic Zone since 2016 under different size regions. The earthquake catalogue since 1970 is taken from the China Earthquake Networks Center. Both the anomaly learning period and forecast interval are fixed as five years and the moving step is taken as one year in this study. The forecasting efficiency of PI is tested by R score and ROC （receiver operating characteristic） test. The results showed that different region selection might lead to different forecasting results with the same calculation parameters. The R score and ROC tests results for the selected regions with lower seismicity-difference are better than those with higher seismicity-difference. In the selected regions with higher seismicity-difference, target earthquakes in the areas with higher seismicity tend to be predicted more easily than those in the areas with lower seismicity, which is supposed to be caused by the fact that PI hotspots are more obvious in the areas with higher seismicity and they will suppress the anomalous signal detected by PI algorithm in the areas with lower seismicity, resulting in the missing prediction for the target earthquakes in the areas with lower seismicity. For a specific target earthquake, the imagine of PI hotspot around the epicenter will evolve, so the combination of multiple forecasting windows should be considered when the forward events are predicted using PI method. Different from the hotspots evolution trend of other natural tectonic earthquakes, the hotspots of 2019 M_S6.0 Changning and 2021 M_S6.0 Luxian earthquakes in Sichuan appeared and disappeared repeatedly near their epicenters, which may be related to human activities. There are continuous PI hotspots at the boundary of southwestern Yunnan, middle and eastern section of Haiyuan fault, the middle Xiaojiang fault, the southern Longmenshan fault and the northeastern Xiaojiang fault, suggesting that there will be seismic potentials with M_S6.0 or above in these regions.