After review of all the beach segments studied in 2001-2008, it was clear that the presence and absence of lingering subsurface oil were being influenced by a number of factors. The initial degree of oiling is a very important influence on the presence of lingering oil, because the heavier the oiling, the deeper the oil would penetrate into the subsurface, which occurred only where the shoreline was permeable. Low exposure to wave action slowed the natural reworking of the oiled sediments. The importance of "armoring," which is where the finer gravels on the beach surface are transported away by waves, leaving a layer of coarse gravel on the surface that is very stable, was a new discovery following the spill (Hayes et al. 2010). Armored gravel beaches occur in Prince William Sound because of the wide range in the sizes of the gravel and in the variations in wave energy. In addition, the ruggedness of the shoreline creates intertidal and nearshore bedrock outcrops that act as natural breakwaters, creating micro-sheltered habitats and bending the waves in ways that pile up sediment behind them. Accumulations of boulder-sized rubble along sheltered bays provide semi-permeable sediments with very limited wave energy. Li and Boufadel (2010) found that, in beaches with low freshwater seepage from the land, there were two layers in the beach: an upper layer that was permeable and a lower layer that was 100 times less permeable. The dissolved oxygen in the lower layer was low, which slowed microbial degradation of oil that penetrated into this lower layer.
Spatially explicit machine learning models were developed to identify potential areas where subsurface oil is still present on the shorelines of Prince William Sound and the Gulf of Alaska affected by the Exxon Valdez oil spill. The models were based on data collected at 314 shoreline segments that were surveyed between 2001 and 2008. These data allowed identification of geomorphologic and hydrologic factors that have contributed to the persistence of subsurface oil within Prince William Sound and the Gulf of Alaska two decades after the Spill. The model used surrogates for these identified factors to make predictions about where oil subsurface oil is likely to be present at unsurveyed locations. The model results suggest there are a limited but significant number of as-yet uninvestigated locations in the study area that are likely to contain subsurface oil. Furthermore, the model results can also be used to prioritize shorelines for investigation with known uncertainty.