For effective earthquake and tsunami early-warning it is crucial that key earthquake parameters are determined as rapidly and reliably as possible. This entails that initial estimates of earthquake parameters will be based on a minimal data set and are highly prone to error. These errors may be manageable and rapidly become small as more data is obtained. However, there may be large, persistent errors and bias in the earthquake parameters, indicating a much too high or low magnitude or hypocenter depth, a largely incorrect epicentre, or even a false event. The use of expert and machine intelligence and statistics based on past events can aid in identifying large, persistent errors and bias in earthquake parameters. At the INGV CAT* tsunami alert center, Early-est** is the module for rapid determination of the location, depth, magnitude, mechanism and tsunami potential of an earthquake. Early-est produces fully automatic results and their uncertainties in the shortest possible time using as few as 3 to 5 P onset observations. We present aspects of the intelligence and statistics used in Early-est to identify persistent errors and bias in the earthquake parameters. The R statistical computing and graphics language is used to exploit catalogs of past earthquake locations and phase data. We discuss indicators that are likely to be most efficient and reliable, such as measures of distance and azimuth distributions of detecting stations, the proportion of available stations that detect an event as a function of magnitude, and the goodness of fit of observed amplitudes to a theoretical attenuation law.
* CAT, “Centro di Allerta Tsunami”, part of the Italian, candidate Tsunami Service Provider ** Early-est (EArthquake Rapid Location sYstem with Estimation of Tsunamigenesis)