:.r :■■ :.*> ;.~ i-^ :... 



.' ; i - : : - : 



i i u i :j ::-' : :: :i : 



«1T10«M- rtffilNE FISJCBIE5 SERVICE 



PfiCIFIC ENVIBtNIEHTra. CRCUP 



nCNIERET. CmiFCFHin 



DISTRIBUTION OF 

 OBSERVATIONS 



Figure 2. — Distribution of surface meteorological observations per 1° square. 

 The contour interval is 2,000 observations. Values >3,000 are shaded. 



the subpopulation extracted from the most recent 23 yr of data 

 ( 1950-72). Seasonal bias is evident in a general tendency for fewer 

 observations during winter than in summer, particularly in the 1° 

 squares containing < 100 observations/mo. The long-term means 

 for June and July, for example, are based on approximately twice 

 the number of observations as the mean values for December and 

 January. However, along the shipping route from San Francisco to 

 Hawaii, the summer means are based on from 10 to 20% fewer 

 observations fnan the winter values. Within the relatively densely 

 sampled coastal shipping lanes, the observations are approximately 

 uniformly distributed in time, and exhibit only a slight tendency for 

 the summer bias. 



Inconsistency in the temporal and spatial sampling of surface 

 atmospheric properties may introduce moderate to severe aliasing 

 in the estimates of long-term monthly mean radiative and turbulent 

 heat fluxes. Certain 1 ° square areas contained a disproportionately 

 large number of observations per month with respect to the sur- 

 rounding squares (Fig. 2). For example, the means for the 1° 

 squares located at lat. 40°N, long. 124°W; lat. 46°N, long. 

 124°W; and lat. 48°N, long. 125 P W were primarily based on 3- 

 hourly surface observations collected at the Blunts Reef. Columbia 

 River, and Umatilla Lightship stations from 1970 to 1972. We 

 noted a greater than order of magnitude increase in the numbers of 

 observations per year, from <100 to more than 1,000, at these loca- 

 tions beginning in 1970. According to Quayle 4 , the lightship data, 

 which may be of questionable quality, were not normally archived 

 in the TDF- 11 file and may have been included as the result of spe- 

 cial contractual requests for these observations. Additional regions 

 of dense spatial and temporal sampling at lat. 32°N, long. 124 °W 

 and lat. 50°N. long. 136°W exist as a result of the U. S. Navy radar 

 picket ship program from 1960 to 1965. A particular consequence 



J R. G. Quayle. Chief. Applied Climatology Branch. National Climatic Center. 

 NOAA, Asheville. NC 28801. pers. commun. April 1981. 



of these station-specific biases will be discussed in considering 

 errors associated with computations of latent heat flux. 



Areal averages of track-specific data, such as the TDF-11 

 reports, may be aliased if regions of relatively large spatial or tem- 

 poral gradients are undersampled along narrow shipping lanes. It 

 should be noted that this type of bias is not unique to the California 

 Current region, and may be present in the summaries of Bunker 

 (1976) and Hastenrath and Lamb (1978), who also used data from 

 the TDF-1 1 file for computations of surface heat flux. Weare and 

 Strub (1981) suggested that spatial and temporal biases in surface 

 marine atmospheric data may contribute as much as 10% of the var- 

 iance in long-term monthly mean flux estimates averaged over 5° 

 squares. By averaging over 1° squares, we expect that the mean 

 deviations resulting from spatial bias within a 1° square should 

 contribute substantially < 10% to the long-term variance. Tempo- 

 ral biases in the long-term monthly means caused by nonuniform 

 sampling within a month or by strong diurnal variations in surface 

 atmospheric properties, e.g., cloud cover, should be similar to the 

 estimates calculated by Weare and Strub (1981). These biases may 

 contribute the same order of magnitude variance as errors associ- 

 ated with random or systematic measurement and archiving errors 

 and uncertainties in the bulk formulae. 



Calculations of net short-wave radiation, Q s . reflect uncertainties in 

 the estimates for cloudless sky radiation and sea surface albedo and an 

 inability to accurately parameterize the combined effects of cloud 

 amount, type, height, thickness, and opacity on insolation. We previ- 

 ously discussed the errors in estimating insolation under clear-sky and 

 cloudy conditions based on studies conducted by Reed (1977) and 

 Simpson and Paulson (1979). The uncertainties in Paynes (1972) val- 

 ues for sea surface albedo range from < 7 % for solar altitudes > 25 ° to 

 25% for low solar altitudes, and may contribute from <0.5 to 10% of 

 the error in individual estimates of Q s . 



The monthly distributions of insolation discussed in this report 

 are based on values computed from individual reports of total cloud 

 amount. This procedure assumes that a single estimate of cloudi- 

 ness, regardless of time of day, is representative of the daily mean 

 cloud cover. Synoptic meteorological observations are usually 

 reported at 0000, 0600. 1200, and 1800 Greenwich Mean Time 

 (GMT), which correspond to 1600, 2200, 0400, and 1000 Pacific 

 Standard Time (PST). Even if equal numbers of reports were avail- 

 able from all four synoptic periods, which is certainly not the case, 

 one-half of the estimates of cloud amount would be from nighttime 

 observations. Alternatively, if only daytime estimates of cloud 

 amount were used, as suggested by Tabata ( 1964). then the number 

 of observations would be reduced by a factor of 2 or more, because 

 in certain 1° squares nighttime observations predominate. 



If the distribution of cloud amount during a 24-h period is uni- 

 form, then either daytime or nighttime observations may be used to 

 reliably estimate the mean daily cloud amount. However. Tabata 

 (1964) and Weare and Strub (1981) have shown a tendency for day- 

 time estimates of cloud cover to be 0. 1 greater than nighttime esti- 

 mates at OWS-P and for a 5° square near the coast of Baja 

 California. Further difficulties in correcting insolation for cloudi- 

 ness are caused by the primitive nature of observing and reporting 

 from ships. Because observations of total cloud amount are 

 reported in oktas. or in tenths converted to oktas, random errors of 

 0.125 (1 okta) in individual estimates of cloud amount can be 

 expected. Reed (1977) also suggested the possibility of up to 0.20 

 positive bias in visual estimates of cloud cover in comparison with 

 satellite-derived values. An error of 0.125 in the linear cloud factor 

 in Equation (2) would produce errors in short-wave radiation rang- 

 ing from 8 to 10% for cloud amounts < 0.4 to 18% for total over- 

 cast. We estimate that the error in the long-term mean radiation 



