Abstract
The systematic forecast of earthquakes is carried out regularly in a pre-selected analysis zone. At each iteration, new data on the seismic process are loaded, processed, transformed into grid-based spatiotemporal fields, machine learning is applied, and a forecast is provided with a constant time interval. The result is a map of the alert zone where epicenters of all target earthquakes are expected within the forecast interval. The minimum alert area method is used for the forecast. In the new version of systematic earthquake forecasting, the solution to the problem is divided into two stages. At the first stage, the algorithm identifies alarm intervals containing target earthquakes with epicenters in the analysis area. At the second stage, during alarm intervals, the algorithm predicts alarm zones containing all epicenters of target earthquakes. This allows one to optimize the estimation of the probability of detecting epicenters of target earthquakes in a series of forecasts and the probability that all epicenters of target earthquakes will be within the predicted alert zone in a single forecast. An example of applying the method to the earthquake forecasting in Kamchatka, California, and the island part of Japan are considered.