Emery el al : Do oil and gas platforms reduce recruitment of Sebastes paucispinis to natural habitat' 



393 



35.0N- 



34.8- 



34.6- 



34,4- 



34.2- 



34.0 



35. ON 



34.8- 



34.6- 



34.4- 



34.2- 



340 



-120.8 



120.6 



-120.4W 



120.6 



-120.4W 



Figure 1 



(A) Map of study area near Pt. Conception, California, showing trajectories derived from high-frequency 

 (HF) radar from 1 May-31 Aug 2002. Triangles show HF radar locations and the white square shows Plat- 

 form Irene. Gray curve superimposed on trajectories is the coverage boundary used for 2002. Labeled thin 

 black lines are bathymetric contours. (B) Heavier black lines are 25 sample trajectories that intersect the 

 coverage boundary. Gray curve is the same as in panel A. The trajectories in panels A and B were created 

 from velocity time series that were interpolated with empirical orthogonal functions (EOFs). 



The number of trajectories reaching the coverage 

 boundaries defined in Figures 2A and 3A were reduced 

 by gaps in spatial and temporal radar coverage. For 

 example, of the 670 possible trajectories in 2002, 541 

 (81%) ended within the radar coverage area and 129 

 (19%) intersected the coverage boundary. Changes in 

 spatial coverage on diurnal and longer time scales re- 

 sulted from several factors, such as broadcast interfer- 

 ence, and are a characteristic of HF radars (Paduan 

 and Rosenfeld, 1996). Gaps in the velocity time series 

 were also caused by outages of individual radars. The 

 average durations of these gaps were 4.4 ±22.3 h and 

 5.9 ±7.9 h in 1999 and 2002, respectively. Outages of 

 individual radars also produced a few long gaps in the 

 velocity time series for each year across the entire cov- 

 erage area. In 1999 two long gaps occurred: one from 

 1800 coordinated universal time (UTC) 28 June through 

 2000 UTC 22 July, and a second from 2300 UTC 24 

 July through 2200 UTC 13 August. In 2002 a single 

 long gap occurred from 1700 UTC 16 May through 0100 

 UTC 21 May. These longer gaps were not filled. 



Shorter gaps were filled by interpolation by using em- 

 pirical orthogonal functions (EOFs; (Emery and Thom- 

 son, 1998)). EOFs incorporate the underlying spatial 



structure of all velocities recorded at all locations where 

 data existed at a given time. Any velocity component, u 

 say, at grid point j may be expressed as 



N 



where t = time; 



£7 = the time average at locationj (computed from 



available data at locationj); 

 a, = the time-varying amplitude function; 

 q>,j - the ith spatial EOF mode at locationj; and 

 A^ = the number of modes. 



The first seven modes (i.e., N=7) were used for inter- 

 polation and explained 64% (1999) and 56% (2002) of 

 the variance. EOF interpolation increases the number 

 of trajectories reaching the coverage boundary to 99%. 

 As a test, gaps were also filled with linear, spline, and 

 moving average interpolation, but EOF interpolation 

 resulted in the most trajectories reaching the coverage 

 boundaries. Otherwise, results did not depend strongly 

 on the interpolation method. 



The fraction of filled data with EOF interpolation 

 compared with the total possible data was 4% in 1999, 



