Five to ten years of data from an array of 34 total field magnetometers are used to define the temporal and spatial characteristics of secular variation throughout central and southern California. For this period, well-determined rates of secular variation are obtained at each site. These rates are temporally linear but spatially variable, ranging from -45 nT/a near San Francisco to -54 nT/a near the Mexican border. Least squares analysis of all data indicates secular variation decreases in a general southeasterly direction according to F =k1*&thgr; +k2* ϕ+K, where F is in nanoteslas per year, &thgr; and ϕ are the geographic latitude and longitude, and k1, k2, and K are 1.66¿0.13 nT/a deg, -0.13¿0.10 nT/a deg, and -123.2¿0.2 nT/a, respectively. Deviations of as much as 1 nT/a occur on scales of a few tens of kilometers. These apparent small-scale secular variation anomalies result, in part, from differences in local induction and remanent magnetization and may be reduced by determination of a site transfer function. A planar surface fit to the corrected data has the form F= k1*&thgr; +k2* ϕ+K, where k1, k2 and K are now 1.50¿0.08 nT/a deg, -0.23¿0.06 nT/a deg, and -129.2¿0.1 nT/a, respectively. Residual field variations obtained after correction of all data for secular variation are most apparent on the San Andreas fault in southern California between Palm Springs and the Salton Sea and marginally so along the recent Coyote (ML =5.9 of August 6, 1979) and Morgan Hill (ML =5.8 of April 24, 1984) aftershock zones. These residuals could be explained by stress localization in these regions and, particularly in the case of the southern San Andreas anomaly, may indicate the location of a future damaging earthquake. Incomplete correction for complex site effects may be an alternative explanation. In the first large-scale test of global secular variation models we find that the models for this region do not predict either the amplitudes or the mean isogram directions of these data to better than several tens of nanoteslas per year and several tens of degrees, respectively. This may result from a failure to correct for site response effects at some observatories before using the data in global spherical harmonic expansions. Local magnetization response can therefore bias estimates of secular variation and yield apparent impulsive behavior when external fields are perturbed. |