per month varies from less than 20 (in January off Van- 

 couver Island) to more than 3,800 (in May off Los 

 Angeles). A significant increase in the number of obser- 

 vations is evident along the shipping route between San 

 Francisco and Hawaii. Less than 20% of the vector means 

 are based on fewer than 50 observations per 1-degree 

 square per month. Some temporal bias may also exist, 

 since approximately 50% of the total reports have been 

 taken since 1945. However, the general coherence of the 

 resulting vector fields indicates that the composite wind 

 stress distributions can be used to describe the dominant 

 seasonal cycle in the California Current system. 



The historical surface marine observations used in this 

 study have been obtained primarily from ship logs, ship 

 weather reporting forms, and published ship obser- 

 vations. The reports differ markedly in method and 

 precision of measurement, ranging from observations 

 made aboard 19th century vessels, to those taken aboard 

 modern oceanographic research ships. Possible errors in 

 wind measurement have been summarized by Hansel 

 (1970). These include influences of anemometer height, 

 atmospheric stability, wind gusts, and duration of the 

 observation. Sources of error in wind estimates include 

 the wind effects observed, error in determining the true 

 wind from the observed apparent wind, and the un- 

 avoidable subjectivity of the observer. 



A substantial portion of the historical data consists of 

 wind reports based on the antiquated Beaufort wind 

 force scale (Kinsman 1968). Table 1 lists the conversion 



Table 1. — Wind speed equivalents of 

 Beaufort force estimates in knots 

 and meters per second. 



Beaufort 



Knots 



-1 

 ms 











0.0 



1 



2 



0.9 



2 



5 



2.4 



3 



9 



4.4 



4 



13 



6.7 



5 



18 



9.3 



6 



24 



12.3 



7 



30 



15.5 



8 



37 



18.9 



9 



44 



22.6 



10 



52 



26.4 



11 



60 



30.5 



12 



68 



34.8 



scale between the Beaufort number and the equivalent 

 wind speed in knots and meters per second. These es- 

 timates are equivalent to a wind speed measurement at a 

 height of 10 m. 2 



Based on the documentation for the National Climatic 

 Center's data file (TDF-11), 3 a determination was made 

 of the number of wind observations estimated and those 

 actually measured by an anemometer. Approximately 

 35"% of the reports in this study consisted of Beaufort 



wind force estimates. An additional 53% of the reports 

 were estimated wind speeds which did not correspond to 

 the Beaufort wind force scale. Less than 12% of the total 

 reports consisted of measured quantities. 



In addition to the errors introduced by the necessary 

 calculation of true wind from the measured apparent 

 wind, the reported directions varied in precision. Resolu- 

 tion varied from ±11.25° to ±5° corresponding to obser- 

 vations based on 16 points of a 32-point compass and a 

 36-point compass, respectively. 



The above considerations lead to the conclusion that 

 random observational errors may be as large as real non- 

 seasonal fluctuations in the distributions of surface wind 

 stress. The problem may be formalized by expressing the 

 individual stress estimates t x and T y as the sums of 

 monthly mean values t s and f , deviations from the 

 monthly means ^andr',., and error terms'^ andf v . The 

 second and higher moment statistics will consist of the 

 nonzero correlations between the deviations from the 

 monthly means and the error terms. If the observational 

 errors are greater than the real fluctuations, then the 

 standard deviations of the monthly means will reflect ob- 

 servational noise rather than actual nonseasonal 

 variability. However, provided that these errors are not 

 systematic, the resultant first moment statistics will be 

 the appropriate estimates of the monthly mean wind 

 stress. 



The standard error of the mean provides a more ap- 

 propriate quantitative measure of the relation between 

 the standard deviation of a set of measurements and the 

 precision of the mean value of the data set (Young 1962). 

 The standard errors of the means are defined by: 



(s f ■%) = <s T /ra,s 7 //^) 



(3) 



: Resolution 9, International Meteorological Committee, Paris, 1946. 

 National Climatic Center, Tape Data Family 11, NOAA/EDS/NCC, 

 Asheville, N.C. 



where Sf and Sj are the standard errors of the means 

 of the eastward and northward components, respectively, 

 S T and S T are the standard deviations, and TV is the 

 number of observations. Large values of N and small 

 values of S Tj and Sr x correspond to mean values t x and 

 f, which closely approximate the population means. 



The computed standard errors of the means and num- 

 bers of observations of each 1-degree square area and 

 long-term month are tabulated in Appendix II. Values 

 less than 0.1 dyne cm -2 occur throughout a large portion 

 of the summary area, although spatial and seasonal 

 dependence is evident. Standard errors greater than 0.1 

 dyne cm " 2 occur over large areas north of Cape Men- 

 docino and between 500 and 1,000 km off the coast of 

 Baja California. High values are generally associated 

 with regions of sparse date, although inadequate sam- 

 pling of intense storms may also lead to large values. 

 Typical ratios of the standard error to the mean stress 

 (S^ It ) range from 0.01 off southern California to 1.0 off- 

 shore from Cape Blanco. Near Cape Mendocino, this 

 ratio varies between 0.02 and 0.10. Minimum values ad- 

 jacent to the coast south of lat. 40°N and along the ship- 

 ping lane between San Francisco and Hawaii are well 

 correlated with the distribution of observations shown in 

 Figure 2. 



