ABSTRACT
Quick and correct estimation of earthquake magnitude, within the first few seconds after P‐wave arrival and before the arrival of damaging strong ground motions, is considered to be the main challenge in earthquake early warning systems (EEWSs). In recent years, three frequency‐based approaches, predominant period (), characteristic period (), and log‐average period (), have been applied by EEWSs or studied by scientists to estimate earthquake size. and parameters suffer from biasing toward high frequencies, which yields erroneous results. For magnitude prediction in EEWSs, we propose a new wavelet‐based scale obtained from P‐wave time windows of borehole accelerograms of earthquakes in Japan with magnitudes of 3–8 and epicentral distances less than 100 km. For events with , the new empirical formula obtained from regression of provides more accurate and reliable magnitude estimation than the other , , and proxies. The magnitude estimation error—one of the main causes of false and missed alarms in EEWSs—can be reduced using parameter.