Four invar-wire strainmeters have been operated in shallow trench sites for 19 months beside the San Andreas Fault at Parkfield, California. Temperature and rainfall records were correlated with 1 yr of strainmeter data, and 90 per cent of the strain signal power at periods between 2 and 120 days was predicted entirely from these records, using a multi-channel, Wiener filtering technique. The residual strain series fluctuates with a peak-to-peak amplitude of nearly 10−6 strain. Anomalous strain signals taking place over several days would have to be larger than this to be identifiable. Previous work shows that signals of amplitude 10−7 strain are identifiable if they take place within hours. Deep creep events giving rise to such signals, which may occur as precursors to earthquakes, would need to be very large.
Other workers have shown that shallow, short-base line tiltmeters in California are also very sensitive to meteorological noise. Strainmeter and tiltmeter installations can be made less sensitive to meteorological noise, either by manufacturing instruments with long (∼1 km) base lines, or by using tunnel or borehole sites (≳100 m deep). Proven instruments of these types are costly, unless an underground site was already available. However, if networks of shallow, shortbase line strainmeters or tiltmeters are to be used for earthquake prediction, it is obviously desirable to invest in at least a few installations which are less sensitive to noise of meteorological origin.