Remotely-Sensed Oil and Shoreline Interaction Modeling

Abstract

ABSTRACT No. 2017-233 The broad adoption of remotely sensed data and derivative products from satellite and aerial platforms available to describe the distribution of spilled oil on the water surface was an important factor during Deepwater Horizon (DWH) oil spill both for tactical response and damage assessment. The availability and utility of these data in describing on-water oil distribution provide strong temptation to make estimates about on-shoreline oil distribution. The mechanisms by which floating oil interact with the shoreline, however, are extremely complex, heterogeneous at fine spatial scales, and generally not well described or quantified beyond broad conceptual or spill-specific empirical models. In short, oil on water does not necessarily lead to oil on adjacent shorelines. We combine data derived from NOAA’s National Environmental Satellite, Data, and Information Service (NESDIS) using a variety of satellite platforms of opportunity describing the remotely-sensed, daily composite anomaly polygons representing oil on water over multiple months with ground observations made in the field, collocated in time and space extracted from a newly compiled database of ground survey data (SCAT, NRDA and others) from the northwestern Gulf of Mexico. Because this new compiled dataset is very large (100,000s of observations) and spans a wide range of habitats, geography, and time, it is particularly suitable for inference and predictive modeling. We use these combined datasets to make inference about the relative influence on shoreline oiling probability and loading of distance from on-water oil observation via multiple distance metrics, shoreline morphology, water levels and ranges, wind direction and speed, wave energy, shoreline aspect and geometry. We also construct predictive models using machine-learning modeling methods to make predictions about shoreline oiling probability given observed distributions of on-water oil. The importance of this work is three part: firstly, the relationships between these parameters can assist hind-cast modeling of shoreline oiling probability for the Deepwater Horizon oil spill. Secondly, these data and models can permit similar modeling for future spills. Lastly, we propose that this dataset serve as a nucleus that can be expanded using data from subsequent or future spills to allow iteratively improvements in shoreline oil probability modeling using remotely sensed data, as well as an improved understanding of oil-shoreline interactions more generally.