Does more mean less in epicentre estimation?

The epicentres of explosions at two test sites - Balapan (Shagan River), Former Soviet Union and Lop Nor, China - are estimated using the onset times of P from only three or four array stations at teleseismic distances. The epicentres of the explosions are known to within about 1 km from studies that make use of information from satellite imagery; these estimates are taken to be the true epicentres. With the true epicentres, differences between the true travel times and the times from travel-time tables are estimated. The differences include a component - path effects - that results in epicentre bias. Comparing our estimates using three or four stations with the true epicentres shows that with correction for path effects most of the epicentres are within 5 km of true and even without correction most estimated epicentres are within 10 km of true. The results confirm the conclusion of Evernden that if reading error in P times has a standard deviation of a few tenths of a second, reliable epicentres can be obtained given readings from only a few stations. This implies, what has been noted by others, that for epicentre estimation, better results can be obtained with a few well read P times from a constant network of the most reliable and sensitive stations, than by using uncritically all the available times. Even without correction for path effects none of the explosions (with times free from possible clock errors) falls outside a circular 1000 km2 region; 1000 km2 being the search area allowed for an on-site inspection under the Comprehensive Test Ban Treaty. The results suggest that rather than try and calibrate the whole of the International Monitoring System, being set up to verify the Test Ban, it would be better initially to concentrate on calibrating the few stations with the longest recording history and lowest detection thresholds.