Comparative performance of time-invariant, long-range and short-range forecasting models on the earthquake catalogue of Greece

Abstract

Time-invariant, long-range, and short-range forecasting models were fitted to the earthquake catalogue of Greece for magnitudes 4.0 and greater to optimize their ability to forecast events of magnitude 6.0 and greater in the period 1966–1980. The models considered were stationary spatially uniform and spatially varying Poisson models, a long-range forecasting model based on the precursory scale increase phenomenon with every earthquake regarded as a precursor according to scale, and epidemic type short-range forecasting models with spatially uniform and spatially varying spontaneous seismicity. Each of the models was then applied to the catalogue for 1981–2002, and their forecasting performance was compared using the log likelihood statistic. The long-range forecasting model performed substantially better than the time-invariant models, and the short-range forecasting models performed substantially better again. The results show that the information value to be gained from modeling temporal and spatial variation of earthquake occurrence rate, at both long and short range, is much greater than can be gained from modeling spatial variation alone.