ninthly) and do not contain an accounting of local wind effects (geos trophic 

 flow). Local wind-induced drift is computed to be 3 1/2 percent of the local 

 wind speed with a 20° clockwise rotation from the wind direction. Following 

 the work of Samuels et al. (1981b), recent model runs incorporate a variable 

 deflection angle which is an exponential function of wind speed. The oil 

 advection algorithm is taken as the vector sum of the surface current vector 

 and local wind-induced drift. Seasonal portrayals of surface currents vary 

 spatially as dictated by available data; examples are contained in Samuels et 

 al. (1982a). Consequences of using various hypothesized versions of surface 

 currents have been examined (Lanfear and Amstutz 1981). 



Local winds are sampled at 3-h intervals, in Monte Carlo fashion, from 

 seasonal wind transition matrices. The matrices are constructed from time 

 series observations, generally measured at coastal stations. The 41 x 41 

 matrices represent eight directions with five speed classes each, and the calm 

 condition. Winds are thus treated as a first-order Markov process. Wind zones 

 are assigned over the modeled area to dictate where each station time series 

 applies. Wind zone definition is determined through comparison of wind roses 

 at sea, derived from ship observations, with those constructed from the coastal 

 station time series. 



Spill advection continues until the spill contacts land, encounters a model 

 boundary, or remains at sea for more than 30 days. Spills are launched through- 

 out the year (500 per season or 2,000 per year) from each launch site; thus 

 the Monte Carlo sampling error does not exceed 2 to 3 percent. Launch sites 

 may be: single points, to simulate platform locations; along lines, to simulate 

 pipelines and tanker routes; or collections of uniformly distributed points, to 

 simulate several platforms in a small area. 



Biological resources (commercial fishing areas, whale migration routes, sea 

 otter ranges, pelagic sea bird feeding grounds, etc.) are represented spatially 

 for times (months) they are considered sensitive to oil spills. Shorelines may 

 be subdivided by type or use (e.g., rocky shores, salt marsh, high- intensity 

 use beaches) and by land segments. Land segments may be designed to be of 

 equal length (typically 20 to 30 nmi) and also by arbitrary criteria such as to 

 be coincident with political subdivisions. 



Conditional probabilities are tabulated annually by launch site identifica- 

 tion and target name, using transit intervals of 3, 10, and 30 days. These 

 conditional probabilities (numerically determined by winds, currents, locations, 

 and temporal sensitivities of resources and locations of spill sites) portray 

 risks from oil spills without consideration of the likelihood of oil spill 

 occurrence. Conditional probabilities, though useful to decisionmakers, are 

 beyond their control. 



Determination of future spill incidence is a complex task. The DOI oil 

 spill risk model projects future spill incidence upon past experience using 

 volume of oil as an exposure variable. Predicted probability distributions for 

 spill incidence are calculated separately for platforms, pipelines, and tankers. 

 Spill occurrence is calculated for size categories equal to and greater than 

 1,000 barrels and 10,000 barrels. The predictive procedure used in the model 

 was initially developed by Devanney and Stewart (1974) using Bayesian techniques, 

 Additional analyses have addressed the data bases and alternative exposure 



94 



