Sie sind auf Seite 1von 13
Marine Pollution Bulletin 115 (2017) 7 – 19 Contents lists available at ScienceDirect Marine Pollution Bulletin

Contents lists available at ScienceDirect

Marine Pollution Bulletin

journal homepage: www.elsevier.com/locate/marpolbul Review State of the art review and future directions in oil

Review

State of the art review and future directions in oil spill modeling

Malcolm L. Spaulding

Ocean Engineering, University of RI, Narragansett, RI 02882, United States

University of RI, Narragansett, RI 02882, United States article info Article history: Received 16 November 2016

article info

Article history:

Received 16 November 2016 Received in revised form 28 December 2016 Accepted 1 January 2017 Available online 11 January 2017

Keywords:

Oil spill modeling Transport and fate processes Surface and subsurface transport Spreading Evaporation Emulsication Entrainment and droplet size distributions Dissolution Biodegradation Oil sediment interaction Lagrangian elements Spherical particle hydrodynamics

abstract

A review of the state of the art in oil spill modeling, focused on the period from 2000 to present is provided. The review begins with an overview of the current structure of spill models and some lessons learned from model development and application and then provides guiding principles that govern the development of the current generation of spill models. A review of the basic structure of spill models, and new developments in speci c transport and fate processes; including surface and subsurface transport, spreading, evaporation, dissolution, en- trainment and oil droplet size distributions, emulsication, degradation, and sediment oil interaction are present- ed. The paper concludes with thoughts on future directions in the eld with a primary focus on advancements in handling interactions between Lagrangian elements.

© 2017 Elsevier Ltd. All rights reserved.

Contents

1. Introduction

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

8

2. Structure of current generation of oil spill models

 

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

8

3. Lessons learned and guiding principles

 

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

8

4. Oil transport and fate models

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

9

4.1. Surface and subsurface transport

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

9

4.1.1.

Random walk and random ight models .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

4.2. Spreading

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

10

10

4.3. Evaporation

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

11

Dissolution

4.4. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

11

4.5. Entrainment and droplet size distribution

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

12

4.6. Emulsication

 

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

12

4.7. Biodegradation

 

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

13

4.8. Sediment oil interaction

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

13

4.8.1.

Model validation.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

13

5.

Selected future directions in spill modeling

 

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

14

5.1. LEs and spreading .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

14

5.2. Modeling transport of LEs using random walk and random ight methods.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

15

6.

Summary and conclusions

 

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

15

Acronyms

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

16

Acknowledgements

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

16

Appendix A.

Supplementary data

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

16

References

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

16

E-mail address: spaulding@egr.uri.edu.

0025-326X/© 2017 Elsevier Ltd. All rights reserved.

8

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

1. Introduction

State of the art reviews of oil spill models have been performed ap- proximately every 5 to 10 yrs. over the past two decades providing in- sight into the evolution of spill models and their use in supporting spill response and impact assessment (Huang, 1983; Spaulding, 1988; ASCE, 1996; Reed et al., 1999; NRC, 2003; Afenyo et al., 2015). Recently Spaulding et al. (2012) have performed a review to support the devel- opment of the next generation of spill model for the US Bureau of Ocean Energy Management (BOEM). NOAA has also undertaken a re- view and is developing the next version of General NOAA Operational Modeling Environment/Automated Data Inquiry for Oil Spills (GNOME/ADIOS) in support of spill response. The eld has matured to the extent that textbooks are beginning to emerge on Lagrangian modeling techniques that include applications to oil spills ( Lynch et al., 2015). The objective of the present paper is to provide a brief review of the

current state of development of oil spill models and a sense of future di- rections. The review focuses on some highlights of recent developments but is not comprehensive given space limitation. The review begins with an overview of the fundamental structure of spill models ( Section 2) and lessons learned in the development and application of models over the past decade (Section 3). A review of transport and fate process- es included in the models is provided in Section 4. Future directions in spill modeling are provided in Section 5, Conclusions and Summary in Section 6, and references in Section 7. The review mentions modica- tions to address oil ice interactions and modeling of blowouts but does not provide a review in these areas. The reader interested in modeling

of blowouts might wish to review the results of an inter-comparison

study of the most recent generation of blowout models performed on behalf of the American Petroleum Institute (API), through the Joint In- dustry Task Force, D3 Subsea Dispersant Injection Modeling Team for

a selected series of test cases and summarized in Socolofsky et al.

(2015).

2. Structure of current generation of oil spill models

A review of the current generation of spill models (Oil Spill Con- tingency and. Response Model or OSCAR ( Reed et al., 2000 ), Spill Impact Model Application Package/Oil Modeling Application Pack- age or SIMAP/OILMAP ( French McCay et al., 2015; Spaulding et al., 1992 ), GNOME/ADIOS ( Lehr et al., 1992, 2000, and 2002; Zelenke et al., 2012a, 2012b ), and others) shows that the basic structure is essentially formulated using Lagrangian based methods ( Lynch et al., 2015 ) for the transport processes (advection and dispersion) and individual algorithms for the fate processes. All models address the three dimensional surface and subsurface transport and fate processes and can be applied to both surface and subsurface re- leases. The current generation of spill models uses the random walk method, the lowest level of the hierarchy of Lagrangian methods ( Spaulding et al., 2006; Lynch et al., 2015 ), for predicting transport. The wind, wave and currents necessary as input are pro- vided by supporting environmental (hydrodynamic, wind, and wave) models. The hydrodynamic models maybe either 2 or 3 di- mensional, while the wind and wave models are typically two di- mensional, focusing on surface transport processes. As an alternative, information for oc ean currents may come from broad scale measurement systems, such as high frequency radar (HFR) or data-based methods, and winds from offshore buoys. If the oil

is at the surface, it is treated as a series of Lagrangian elements

(LE), each of which is tracked in space and time; and when the oil

is tracked at the subsurface, the LEs are oil droplets tracked by

droplet size class. The algorithms that describe the fate processes

are typically based on underlying fundamental principles and in- formed/calibrated/valid ated by laboratory and eld observations. The fate processes are modeled either by transferring oil mass

between the environmental compartments (sea surface, atmo- sphere, water column, sea bed, and shoreline) or changing the oil's composition or physical characteristics (e.g. density, viscosity, and interfacial tension). In many cases, the algorithms are empiri- cally based, and hence rely on laboratory or eld observations, therefore with the inherent limitation given the lack of availability of this type of information for the wide variations in spill situations. The oil in spill models is either characterized as a bulk oil of a given type (e.g. medium crude, No 2 fuel oil, etc.), or described in terms of the various components, typically related to its distillation, that comprise the oil. The models track the location of the oil through the distribution of oil mass (total or by component) in space (at the sea surface and in the water column, but typically not in the at- mosphere) and time. The componen t distribution is required for performing impact assessment, wh ere oil composition is critical to dissolution and biodegradation and impact on marine life.

3. Lessons learned and guiding principles

Based on a review of the development and application of oil spill models for spill response, impact assessment, including model valida- tion against every major oil spill in the world, some important lessons in the design of spill model have been learned. These include:

Spill models are typically structured as an integrated series of algo- rithms describing individual fate and transport processes. It is best to have each fate process as a separate algorithm with supporting data provided from other algorithms as appropriate. There may be many potential algorithms for each fate process, and the model frame- work needs to accommodate this functionality and allow the use of various approaches.

The model needs to explicitly incorporate quantication of the uncer- tainty in the algorithms and associated coefcients used in the model, with predictions not only representing the mean value but the uncer- tainty as well.

Each algorithm in the model should be thoroughly documented and tested prior to integration into the overall model system. Rigorous at- tention must be paid to ensure that the oil mass balance is preserved in each algorithm and throughout the integrated model.

It is critical to be able to represent the behavior for temporally and spatially dependent releases of oil. For example, in the case of the Deepwater Horizon oil spill the release was originally coming from the end of the collapsed riser, then from the collapsed riser and a se- ries of holes at a kink in the riser, and after the riser cut from one loca- tion above the well head. The release rate, gas to oil ratio, and oil droplet size distribution at each location varied with time, over rela- tively short time scales (Spaulding et al., 2015, 2016a, 2016b).

Spill models employ a three dimensional, Lagrangian-particle-based strategy to represent the oil. This allows the model to account for the temporally and spatially varying release, transport, and fate of the oil. This strategy is also amenable to sensitivity testing of results to the number of particles in the simulation. Care needs to be exercised in the particle tracking and aggregation/disaggregation rou- tines, to ensure that the total oil mass and the mass by each compo- nent is preserved.

For many applications the chemical characteristics of the oil (i.e., oil weathering ) needs to be tracked in order to reliably estimate the mass balance and concentrations of various components of the oil, both in particulate (oil droplet) and dissolved phases. This partitioning in the environment has important implications for the impacts of oil on the marine ecosystem.

Model output, at a minimum, should include the oil mass balance and the spatial (surface and subsurface) and temporal distribution of the oil.

A geographic information system (GIS) framework should be used since it provides an effective and ef cient tool to visualize model

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

9

input, output, and supporting information.

Use of carefully designed geospatial tools for post-processing and vi- sualization is essential and allows effective communication of model results.

The model should support a wide variety of electronic mapping sys- tems (electronic navigation charts, Google Earth maps, aerial photo based maps, etc.).

The model should undergo thorough validation against actual spill events, to the extent that data allow.

The model should be agnostic to the grid system or spatial location on which environmental data (wind, wave, current, temperature, and sa- linity) are provided, since these are likely to change from application to application.

Environmental data necessary to support model operations (wind, currents, and waves) and to assist in visualizing results is most readily organized in the form of an external Environmental Data Server (EDS) or similar structured data base. The EDS will collect, archive, and manage data as required for the model.

The model needs to explicitly consider its potential use to assess the impact of response options (e.g. mechanical recovery, dispersant treatment). The processes that are affected by the application of dis- persant need to be considered in the algorithm if dispersants are ap- plied (e.g. oil droplet size distribution algorithm used needs to be able to account for the impact of applying dispersants to the release.)

The model should use the most recent data management and commu- nication standards in the integrated ocean observing systems (IOOS) community.

To maximize its use, the model should be targeted for operation on the current generation of personal computers and extension to inter- net enabled operation (cloud).

Given the dramatic difference in spatial scale (100's of meters for blowout model, and kilometers for hydrodynamic model), it may be useful to have the blowout model used to characterize the releases of oil and gas ( Johansen, 2000; Spaulding et al., 2000, 2015; Chen and Yapa, 2002; Yapa and Chen, 2003, 2004, Yapa et al., 2010;

Zheng et al., 2002; Zhao et al., 2014, 2015, and 2016a,b) for subsurface releases as a separate standalone system, with linkage to the large scale spill model via specication of oil and gas concentrations (and droplet sizes) at the appropriate location and time.

4. Oil transport and fate models

Following the basic framework outlined above, a review of advances in spill modeling for the transport and fate processes are provided below. Transport may be either at the surface or subsurface. The princi- pal fate processes of interest are: spreading, evaporation, entrainment (and oil droplet size formation), emulsication, dissolution, biodegrada- tion, photo-oxidation, and sediment oil interaction (Fig. 1). Spill models employ a Lagrangian particle (element or spillet) based strategy to rep- resent the oil at the surface, and droplets to represent the subsurface oil. This allows the model to account for the temporally and spatially vary- ing release, transport, and fate of the oil. The chemical characteristics of the oil are also tracked in order to reliably estimate the mass balance and concentrations of various components of the oil, both in particulate (oil droplet) and dissolved phases.

4.1. Surface and subsurface transport

Surface and subsurface oil transport is modeled using a Lagrangian formulation where the total currents are represented by a vector sum- mation of currents due to large-scale ows, wind and wave induced cur- rents, and turbulent dispersion. Input data on the surface/subsurface currents are provided by a hydrodynamic model. For surface currents, high frequency radar (HFR) system data can be used. The wind-wave in- duced transport at the surface is modeled using an algorithm developed to estimate the percent of drift of the surface oil for a given wind. Dis- persion is modeled as a random walk process with the dispersion coef- cients either derived from the hydrodynamic model or directly specied by the user.

derived from the hydrodynamic model or directly speci fi ed by the user. Fig. 1. Oil

Fig. 1. Oil fate processes in the open ocean.

10

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

Most existing spill models use a random walk strategy to model the movement of the spill at the surface and subsurface (e.g. GNOME (General NOAA Operational Modeling Environment) ( Zelenke et al., 2012a, 2012b ), which is the oil spill trajectory model used by NOAA OR&R Emergency Response Division (ERD) responders during an oil spill); OSCAR (SINTEF's Oil Spill Contingency and Response Model, see Reed et al. (2000) ); OILMAP ( Spaulding et al., 1992 ); SIMAP ( French McCay et al., 2015); and LTRANS (Larval TRANSport model system, see North et al. (2008) for an example of its application). The transport algorithms in spill models, while correctly predicting the large scale structure of spills, are unable to predict the intrusions and ne scale structure that can substantially alter the spill distribution and subsequent transport. In the last decade, there has been a very ac- tive program in trying to understand the dynamics of non-linear chaotic systems (e.g. Haller, 2001, 2002, 2005; Mathur et al., 2007; Tang et al., 2007; Peacock and Dabiri, 2010 ). This work has naturally led to the use of these mathematical techniques to study geophysical (ocean and wind) ow problems. This research area is broadly referred to as Lagrangian Coherent Structures (LCS) analysis and seeks to nd the un- derlying structures in atmospheric and oceanic ows that represent di- viding lines or barriers to the transport of materials released into these environments. These methods have been applied to solve a wide variety of coastal problems, including pollutant and drifter transport in Monterey Bay ( Couliette et al., 2007; Lekien et al., 2005, 2007; Shadden et al., 2009), eddy dynamics in the Gulf of Mexico (Kirwan et al., 2003; Toner et al., 2003 ), and red tides off the west Florida shelf (Olascoaga et al., 2008; Olascoaga, 2010; Olascoaga and Haller, 2012). This approach has recently been applied to the DWH spill and shown considerable promise in being able to help understand the surface transport of oil and the elongations, stretching, and intrusions observed in the spill ( Mezic et al., 2010; Beegle-Krause et al., 2011a, 2011b; Olascoaga and Haller, 2012 ). Mezic et al. (2010) shows LCS (hypergraph) maps on May 25 and June 8, 2010 and the respectively observed surface oil two days later on May 27 and June 10, 2010. Com- paring the LCS maps and the correspondingly observed oiling two days later in the key areas, one can see that the LCS maps clearly identify the major paths of spill elongation and stretching. Beegle-Krause and Peacock's work has been focused on larger-scale control of the DWH oil spill's trajectory, such as blockage of oil from entering the Loop Current, leading to the Tiger's tail events in May 24 26, 2010. To date no one has developed a method or protocol that would allow LCS elds to be used directly as input to a spill model or as an alternate to the existing Monte Carlo based transport method. Following the work of Carlson et al. (2010), Spaulding et al. (2016a) have implemented their methodology to estimate the upper bound for dispersion coef cients used in a spill model. These dispersion coef - cients are determined to ensure that barriers to spill transport are iden- tied and accurately accounted for in the spill model. The rst step is to uniformly seed the ow eld (at an initial time point), obtained from a source such as hydrodynamic model, HF radar, satellite feature tracking, or another equivalent source, and to predict the Lagrangian motion of the particles with progressively increasing levels of dispersion for a pe- riod of several days (2 to 4). These predictions are then processed to de- termine the barriers to transport using relative dispersion (RD) or the standard LCS Finite Time Lyapunov Exponent (FTLE) methods. The two methods give results that are in excellent agreement. In the second step, a dispersive component is added to the Lagrangian model (random walk or random ight) to represent sub- grid scale turbulent transport processes. The RD of Lagrangian trajecto- ries is computed for increasing values of the dispersion coefcient until the mixing barrier is no longer present. The dispersion coef cient at which the mixing barrier disappears provides a dynamical estimate of the upper bound of its value. The progressive elimination of the mixing barrier is clearly observed in numerical simulations. As an example, the RD maps for the Northern Gulf of Mexico based on SABGOM hydrody- namic model ( Xue et al., 2015 ) predictions for the surface layer on

May 27, 2010 were computed for dispersion coefcients varying from 0 to 10 m 2 /s (Fig. S1). The progressive erosion of the barriers is clearly evident. At 10 m 2 /s, the barrier no longer effectively exists. To provide

a more quantitative estimate of barrier strength and its disappearance,

the probability distribution for all particles in the eld is calculated and the kurtosis (skewness) of the distribution determined. The kurtosis is large for low values of dispersion coefcient and decreases

with increasing dispersion. The kurtosis reaches very low values at high dispersion and the distribution becomes Gaussian; as in a classic diffusion problem. The above analysis was repeated assuming the horizontal dispersion

is approximated by the classic Smagorinski formulation (Smagorinsky,

1963) (as used in many hydrodynamic models) and obtained compara- ble upper bounds to the dispersion coefcient. The transport barriers that emerge from this type of analysis naturally evolve from the current eld used as input to the method. The importance of barriers associated with SST (sea surface temperature), SSH (sea surface height), large scale ows (Loop Current) and associated eddies, frontal zones, etc. are integrated and reected in the dynamics of the ow elds.

In practice, only a short simulation period is required to determine the upper bound for the dispersion coefcients. The approach can be ap- plied to any hydrodynamic model prediction and can even accommo- date alternate specications of the ow eld from other sources, such as high frequency (HF) radar or the results of a feature tracking proce- dure, as often available from an analysis of remotely sensed imagery.

This method can be applied to hydrodynamic elds to determine the ap- propriate range of values for dispersion coefcient that are used as input to the spill model. Unlike other LCS based methods, this approach allows

a direct input to the spill model and gives the user control over cross barrier transport.

4.1.1. Random walk and random ight models The vast majority of spill models use a random walk technique to model the transport of the surface oil. In this approach, Lagrangian ele- ments representing the oil are advected by the surface current eld and dispersed by a random walk process that is used to represent horizon-

tal/vertical dispersion. Dispersion is a sub-grid scale process that repre- sents turbulent transport not resolved by the hydrodynamic model. A random walk model has a rst order memory of location. An alternate

is to use a random ight model ( Lynch et al., 2015), which includes a

rst order memory of current speed and direction. Spaulding et al. (2006) provide a detailed description of the model and its implementa- tion and testing for search and rescue problems. Use of this higher order model is possible if one has accurate estimates of the currents and dispersion.

4.2. Spreading

Spreading is typically modeled using the empirically based, thick slick algorithm developed by Mackay et al. (1980a, 1980b, 1982). The model predicts the area of an individual Lagrangian element of oil as a function of time. A spreading rate coefcient is used to scale the results to agree with eld observations. Surface spillets (or Lagrangian ele- ments, LEs) can lose oil mass to the atmosphere via evaporation and into the water column due to entrainment. The spreading ceases when all the volatile fractions of the oil have been lost or when the spill reaches the minimum thickness dened in the oil database. One of the principal criticisms of Mackay (Fay, 1969; Hoult, 1972a, 1972b) like formulations is that the spills are predicted to be circular in shape, with constant thickness no matter what the wind and wave conditions. Observations show that spills are often thicker at the downwind end and have shapes that are elongated ( Galt and Overstreet, 2011; Reed et al., 1999). Recently Galt and Overstreet (2011) have proposed an alternate for- mulation based on the observation that entrainment removes oil from the sea surface, which is transported subsurface in a strongly sheared

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

11

ow, and that some portion of the dispersed oil is returned to the sur- face by the buoyancy of rising oil droplets. In this approach, oil at the sea surface is dispersed in the water column by breaking waves as oil droplets, with injection depths equivalent to approximately 1.5 times the breaking wave height. The oil droplet size distribution is dependent on the oil viscosity and the breaking wave energy. The dispersed oil droplets rise back to the surface, with the times dependent on their size; the larger the droplets the faster the rise times. Currents in the upper water column are strongly sheared due to wind and wave forcing of the near surface layer. Droplets close to the surface travel down-wind the fastest and those at the maximum depth of penetration the slowest.

These processes lead to oil thickening at the leading edge of the spill and

a thinning at the trailing edge. Galt and Overstreet (2011) have imple-

mented this basic strategy as an adjustment to the conventional model- ing of spreading, using Delvigne and Sweeney (1988) formulation for

entrainment and their empirically based oil droplet size distribution. It

is important to note that in Galt and Overstreet's formulation, oil droplet

penetration depth is critical to the entrainment formulation as this is key to predicting the degree of thickening and thinning of oil. In the presence of steady winds, wave induced (Stokes) drift and surface wind stress can interact to form Langmuir cells. These cells ap- pear in counter-rotating, helical vortex pairs aligned with the wind di- rection. The ows induced by the cells result in convergence zones at

the sea surface and down welling between the cells. These cells are re- sponsible for the widely observed surface windrows. Oil is thickened in the convergence zones and the down-welling may be sufciently strong to entrain the oil. The oil transport in the convergence zone at the sur- face down wind is enhanced by the presence of the cells. Leibovich (1997) developed a detailed numerical model of the wave current interaction process described based on his earlier work (Leibovich, 1983). The computational burden for the model makes its routine use for practical problems prohibitive. Fortunately, Galt and Overstreet (2011) have developed a simplied methodology by empir- ically tting these model predictions. This model requires input on the wave and wave conditions at the sea surface and the Ekman depth, or optimally, the prole of wind and wave-induced currents in the near surface layer (to the Ekman depth). Use of Galt and Overstreet (2011) method is an advance over the existing method since it explicitly considers an alternate path for the spreading of oil. It also has the ability to predict the observed thickening of oil at the downwind edge of the spill. It only plays a role in spreading

if entrainment of oil is active (breaking waves are present), so it needs

to be used as a supplement to the current spreading model. It requires work in implementation and will need to be adjusted dependent on the entrainment and droplet size models selected. Galt and Overstreet's (2011) method for dealing with transport associated with Langmuir cells is an advancement, and has never been implemented in a spill model. It is one of the very few processes that can result in the thickening of oil in open water.

4.3. Evaporation

In many spill models, the so-called pseudo-components (chemical component groups) ( Payne et al., 1984; Payne and McNabb, 1985 a) are evaporated according to an evaporative exposure algorithm,

where the ux to the atmosphere is specic to the component's molar volume, vapor pressure, and molecular weight. Jones (1997) developed

a simpli ed pseudo-component (PC) model, relating molar volume,

vapor pressure, and molecular weight to the boiling point of the compo- nent used. In this method, only the boiling points and initial volume fractions of the components need to be speci ed to implement the model. Six evaporating components are typically employed to separate- ly represent BTEX, two PAH fractions, volatile aliphatics, and two semi- volatile aliphatic fractions. Each pseudo-component evaporates according to its vapor pressure, diffusivity, and molecular weight. Recent developments (e.g. French McCay et al., 2015 ) have greatly

extended the number of components used in the analysis to more accurately represent the oils of interest. Fingas (2011a, 2011b) and supporting papers ( Fingas, 1996, 1997, 1998, 1999) has advocated for an empirically based strategy for model- ing evaporation from spills. In his review of the literature and laboratory experiments he has performed, he concludes that oil evaporation rates are not controlled by boundary layer processes between the slick and overlying atmosphere but instead by mass transfer within the slick. Based on this analysis, he concludes that oil evaporation rates are not dependent on wind speed, turbulence levels in the atmospheric bound- ary layer above the spill, slick thickness, or spill area. Most spill models include all of these dependencies via the mass transfer coefcient that controls the rate of evaporation from the surface slick. Fingas proposes an empirical formulation where the evaporation rate is only dependent on temperature and time. He assumes the time dependence is either the square root of time or the natural logarithm of time, dependent on oil type. For diesel fuels and similar, the dependence is square root of time, while for all other oils it is dependent on the natural logarithm of time. He provides these relationships for hundreds of oil types (Fingas, 2011a, 2011b), but no information on how well these relation- ships t the data is provided. While appealingly simple, Fingas's empirically based method has some potential aws including the following: (1) the approach requires that curve ts be generated for all oil types of interest and these types of data are not routinely available. (2) some dispute that his ndings as being overly reliant on laboratory-based experiments, which are not representative of conditions at sea (tray experiments in wind tunnels with limited spatial and temporal scales), and (3) the method does not provide estimates of the evaporative losses of various fractions of oil (volatile and nonvolatile) and hence does not allow for those com- ponents to be tracked in the spill model. This of course has direct im- pacts on the dissolution rate for oil and estimates of the dissolved fraction of oil in the water column and the associated toxicity effects.

4.4. Dissolution

Dissolution is the process by which soluble hydrocarbons (most of which are aromatics) enter the water from a surface slick or from sub- surface oil droplets. The lower molecular weight hydrocarbons tend to be both more volatile and more soluble than those of higher molecular weight. For surface slicks, since the partial pressures tend to exceed the solubility of the lower molecular weight compounds, evaporation accounts for a larger portion of the mass lost than dissolution (McAuliffe, 1989), except perhaps under ice. Dissolution and evapora- tion are competitive processes. The dissolved component concentration of hydrocarbons in water under a surface slick shows an initial increase followed by a rapid decrease after some hours due to the evaporative loss of components. Most soluble components are also volatile and di- rect evaporation (volatilization) from the water column depletes their concentrations in the water. Dissolution is particularly important where evaporation is low (from dispersed oil droplets and in ice-cov- ered surfaces). Dissolution can be signicant from subsurface droplets due to the lack of atmospheric exposure and the higher surface area per unit of volume. The model developed by Mackay and Leinonen (1977) , which is based on Raoult's Law, is widely used for dissolution from a surface slick. The slick (spillet) is treated as a at plate, with a mass ux (Hines and Maddox, 1985) related to solubility and temperature. It as- sumes a well-mixed layer with most of the resistance to mass transfer lying in a hypothetical stagnant region close to the oil. For subsurface oil, dissolution is treated as a mass ux across the sur- face area of a droplet (treated as a sphere) in a calculation analogous to the Mackay and Leinonen (1977) algorithm. The current state of the practice is well represented in the dissolution algorithm as presented by French McCay et al. (2015).

12

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

4.5. Entrainment and droplet size distribution

Delvigne and Sweeney (1988) (DS), Delvigne (1993), and Delvigne and Hulsen (1994), developed an empirical algorithm based on labora- tory and ume experimental observations, to model the entrainment rate of oil from the sea surface. The formula is dependent on breaking wave energy dissipation rate, fraction of the sea surface covered by oil and breaking waves, and an empirical coefcient that depends on oil type and weathering (emulsication) status. The oil injection depth is based on empirical observations, and the droplet rise velocity is based on Stokes law. The oil particle size distribution is determined from an empirical power law t to laboratory and tank experimental observations. Reed et al. (2009) performed laboratory studies to investigate the impact of oil viscosity on entrainment. They also extended Delvigne and Sweeney's approach to higher viscosity, non-Newtonian regions, replacing an ad-hoc correction in the original formula and used the Reynolds number to represent viscous effects. The method was calibrat- ed with laboratory experiments and compared to prior work. Based on laboratory measurements, they determined that a lognormal distribu- tion provided the best t to the droplet size distribution data and hence recommended its use. Their formulation compares well with an empirical formulation custom t to observations recommended by Delvigne and Sweeney (1988). Recently Johansen et al. (2015) have replaced the oil droplet model used in the DS method with a droplet size model developed by Johansen et al. (2013) with dependence on both the Weber (We) and Reynolds ( Re ) numbers. The droplet size model was originally validated with large scale laboratory data for blowout releases, with and without dis- persant treatment (Johansen et al., 2013; Brandvik et al., 2013). Recent work by Zeinstra-Helfrich et al. (2015a,b, and c) has not fundamentally advanced the work of Delvigne and Sweeney. Some of the key problems identied with Delvigne and Sweeney's (1988) formulation over the past three decades (also see Johansen et al., 2015 for a review) include the following:

1. Model formulation is dimensional The empirical model was developed and t to laboratory data using a series of dimensional parameters. The model coefcients hence de- pend on the units of the parameters used in the model. In practical applications, this often presents a challenge. It is also challenging to incorporate new insights into the model as advancements are made.

2. Incorporation of droplet size into entrainment estimate The entrainment formulation estimates the amount of oil entrained into the water column by droplet size class, Δ d, and hence has the droplet size (d droplet diameter) distribution is embedded in the entrainment formulation. This integration makes it impossible to es- timate bulk entrainment, without consideration of droplet formation processes. Conceptually it is simpler to separate these two processes into two independent components.

3. No explicit procedure to address droplet resurfacing The formulation has no explicit procedure to address resurfacing of dispersed droplets. This is typically handled by an ad-hoc adjustment so that droplets that are deemed to surface immediately are not dispersed in the rst place.

4. Use of power law droplet size distribution A power law formulation is used to represent dispersed oil droplet sizes. While the power law ts the original data used in the model development reasonably well, it is not typically used. Lognormal or Rosin-Rammler ( Rosin and Rammler, 1933 ) distribution provide a better t to the oil droplet size data ( Reed et al., 2009; Li et al., 2016a). The approach has not been updated to include advances in oil droplet size modeling (see Li et al., 2016a for a review).

5. Application of dispersants The formulation has no provisions for cases when dispersants are ap- plied to the spill. If this is to be modeled then the droplet size classes

need to be adjusted. The dispersant to oil ratio (DOR) and dispersant efciency (DE) need to be addressed as separate calculations. 6. Ad hoc adjustment for oil viscosity Viscous effects are addressed through an ad hoc adjustment of one of the coefcients in the empirical model formulation and not explicitly incorporated into the model formulation. This was addressed by Reed et al.'s (2009) modi cation making the viscous effects dependent on the Re number. Recently Li et al. (2016b) and French McCay et al. (2015) have revisited the entrainment problem and recast the empirical model in terms of three dimensionless groups (dimensionless oil ux, We, and Oh). The dimensionless ux rate at the surface is expressed in terms of the Weber number (We) and Ohnesorge number (Oh) and is dependent on the oil ux, slick thickness, oil density, and fraction of sea surface covered with breaking waves. The Weber number is the ratio of disruptive momentum (hydrodynamic) forces to restor- ative interfacial tension forces and is a function of seawater density, gravity, wave height, the oil-water interfacial tension, and the char- acteristic length scale (determined by Rayleigh-Taylor instability maximum diameter (Grace et al., 1978)). The Ohnesorge number is the ratio of viscous to interfacial tension forces and is a function of the oil dynamic viscosity, the oil density, oil-water interfacial tension, and maximum diameter. The empirical constants in this formulation were determined by tting the equation to laboratory and eld ob- servations reported by Delvigne and Sweeney (1988), Delvigne and Hulsen (1994), and Reed et al. (2009). These experiments were fo- cused on estimating the amount of oil entrained in the water column in breaking wave events. Delvigne and Sweeney (1988) and Delvigne and Hulsen (1994) experiments use low viscosity oils, while Reed et al. (2009) focused on higher viscosity oils. Recently Li et al. (2016a) has presented a uni ed oil droplet size model that was developed for a variety of turbulent conditions based on non-dimensional analysis of disruptive and restorative forces, which is applicable to oil droplet formation under both sur- face breaking-wave and subsurface-blowout conditions, with or without dispersant application. This new model was calibrated and successfully validated with dispersant-treated and non-treated oil droplet size data obtained from controlled laboratory studies of dis- persant-treated and non-treated oil in API (industry)- and BSEE (government)-sponsored subsea dispersant tank tests. The model was also validated against results from eld surveys, including the Deep Spill experimental release and the Deepwater Horizon blowout oil spill. This model is an advancement over prior models, as it explic- itly addresses the dispersed phase viscosity effects resulting from dispersant application and constrains the maximum stable droplet size based on Rayleigh-Taylor instability that needs to be invoked for a release from a large aperture. To address the problem of resurfacing of dispersed oil droplets, a protocol has been developed that balances the return of oil droplets to the sea surface by buoyancy and the energetic dispersion that occurs in the near surface region in the presence of breaking waves. This replaces the ad hoc method used in Delvigne and Sweeney

(1988).

4.6. Emulsication

Mackay et al.'s (1982) emulsi cation scheme is implemented in many spill models with parameter values for the mousse viscosity coef- cient and an emulsi cation coefcient which determines the rate at which emulsi cation is predicted to occur. The exponential rise in water content is presented by Mackay et al. (1980a, 1980b) and Mackay and Zagorski (1982). The maximum fraction of water incorpo- rated is either provided by the user or specied in the oil database. Fingas (2011a, 2011b) and Fingas and Fieldhouse (2009, 2011) have proposed a Stability Index (SI), based on a regression equation which depends on the density, viscosity, asphaltene, and resin contents of

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

13

the oil to determine if an oil emulsies. These data are readily available for most oils. The SI predictions can then be used to determine the type of emulsion: unstable, entrained water-in-oil state, meso-stable, or sta- ble. A prediction scheme is also given to estimate the water content and viscosity of the resulting water-in-oil state and the time of formation given exposure to a wave environment. The model can provide accurate prediction of the stability class about 90% of the time. Fingas (2011a, 2011b) have used over 400 oils (crude and rened products) to develop the method. The major inaccuracy lies with the unstable types, given the fact that there are three distinct types of oils or fuels in this class, each very different, and because of the possible presence of emulsion brea- kers or asphaltene suspenders in the oils. The predictions were com- pared to the actual data from some common oils and exponential water uptake models dating from the 1980s (Mackay like) and the ADIOS2 model (Lehr et al., 2002). It was found that the present Fingas's model was superior to the prior Mackay's models in predicting oil properties. Fingas (2011a, 2011b) shows a comparison of his proposed model's ability to predict the correct stability class and viscosity and water con- tent with the Mackay et al. uptake model. His model is clearly superior to the Mackay implementation based on this comparison. An appealing feature of Fingas et al.'s approach is that they have tested the method against a wide range of oil types. The method is also based on data that is routinely available for most oils of interest.

4.7. Biodegradation

An in depth review of the literature by Li and French McCay (2016) indicated different biodegradation mechanisms for soluble and insoluble compounds: while microbial degradation acts primarily on the dissolved phase for soluble and semi-soluble aromatics, the biodegradation of insoluble aliphatics and high-molecular-weight PAHs takes place mainly on the particulate phase. Therefore, for the practical application of biodegradation in an oil fates model, biodegra- dation of the soluble and semi-soluble components (AR1 to AR9) can be modeled by applying the rates to the dissolved fraction, considering their much greater dissolution rates than biodegradation rates. The sparingly soluble aliphatic components (AL1 to AL8) and high- molecular-weight aromatics are present mostly in the droplets dis- persed in the water column (or as oating oil). Thus, the AL1AL8 com- ponents can be assumed to biodegrade predominantly in the dispersed phase and the attachment of microorganisms to dispersed oil droplets is the main mechanism leading to biodegradation. Therefore, in practice, biodegradation of the sparingly soluble or non-soluble aliphatic components in the water column need only be quanti ed for the entrained and/or dispersed oil. The derived biodegradation rates of the AL components are for small droplets (mostly b 70 μm), representa- tive of well-dispersed oil entrained and/or dispersed in the water column. The fraction of the well-dispersed oil from either surface break- ing wave conditions or a subsurface blowout release can be quantied by applying a dispersed oil droplet size distribution model (e.g. Li et al., 2016a). The potentially enhanced biodegradation effect of surface and/or subsurface application of dispersants, under surface breaking waves and subsurface blowout release conditions (respectively), is simulated by the combination of the smaller droplet sizes re ecting the dispersant effect on interfacial tension (see Li et al., 2016a), resulting in increased dissolution rate, and the biodegradation algorithm reported therein.

4.8. Sediment oil interaction

Aromatics dissolved in the water column are carried to the sea oor primarily by adsorption to suspended particulates, and subsequent set- tling. In most spill models the ratio of adsorbed to dissolved concentra- tions is computed from standard equilibrium partitioning theory using a dimensionless partition coefcient and the concentration of suspended

particulate matter (SPM) in the water column expressed as mass of par- ticulate per volume of water (e.g., mg/L). Khelifa et al. (2003, 2004, 2005a, 2005b, 2005c) have developed a Monte Carlo model to predict the formation of oil-mineral aggregates. The model is based on collision theory between populations for oil droplets and suspended sediment material. The approach follows the work of Smith and Matsoukas (1998). They have extended the model to address problems with sediment oil aggregation in the presence of dispersants and provided new estimates of settling rates for the aggre- gates that are substantially different from those predicted by Stokes law (Khelifa et al., 2010). The Khelifa et al. method is appealing given its extensive develop- ment, validation with laboratory based observations, and extension to consider the impact of dispersants. The computational time to perform the simulations using his Monte Carlo method is unknown and may be an issue. There are insufcient details given in the papers to readily im- plement the model. It might be possible to approximate his results using curve ts to his experimental data using information on the oil concen- tration to sediment concentration ratio as he has proposed (Khelifa et al., 2005a, 2005b, 2005c). Khelifa et al. (2005a, 2005b, 2005c) compari- son of their Monte Carlo model to a simplied analysis by Guyomarch et al. (1999, 2002) supports this suggestion. Recently a new conceptual formulation of oilparticle coagulation ef ciency, A-DROP, has been introduced to account for the effects of oil stabilization by particles, particle hydrophobicity, and oil particle size ratio on oil particulate aggregate (OPA) formation ( Zhao et al., 2016a, 2016b). The model was able to closely reproduce the oil trapping efciency reported in experimental studies. The model was then used to simulate the OPA formation in a typical nearshore environment where wave breaking, surf, and swash zone processes are active.

4.8.1. Model validation In the interest of better assessing the performance of spill models, they typically are compared to observations. These may include the ap- plication to selected spill events, to laboratory data, or to test spills de- signed to verify the model's transport or fate algorithms. As example, French McCay (2003, 2004) and French McCay and Rowe (2004) pro- vide comparisons of model predictions to observations for the Exxon Valdez and North Cape spills. Spaulding et al. (1994) provide model data comparisons for the Baer spill and Prasad et al. (2014) for a vessel collision and pipeline leak in Mumbai. Moldestad et al. (2004) compare predictions of a weathering model to observations from experimental test spills. Tests designed to verify the model's transport algorithms can be found in French et al. (1997) and French-McCay et al. (2007). The most comprehensive validation of a spill model to date has re- cently been completed for the Deepwater Horizon (DWH) spill by Spaulding et al. (2015, 2016a, 2016b) for the neareld of the subsurface blowout and by French McCay et al. (2016) for the far eld fate and transport. These validations used state of the art methods (OILMAP DEEP for the near eld and SIMAP for the far eld) and had access to all publically available observations of the spill. The characterization of the oil was the most complete possible using Black Oil Tables (BOT) to specify the oil pressure, volume, and temperature relationship for the MC252 oil. The blowout model focused on predicting the neareld dy- namics of the blowout from multiple releases (e.g. kinks in the riser and the end of riser), the plume trapping depth, the spatial extent of the intrusion layer, and the oil droplet size distribution and the associat- ed volume trapped or risen to the surface. Model predictions were per- formed over the entire spill period and included time dependent releases, from varying locations (kink and riser), and the subsurface ap- plication of dispersants by wands and tridents. Model predictions were in very good agreement with the observed trapping depth of the oil based on uorescence and dissolved oxygen data, good agreement with observations of oil droplet size distribution based on holocam measurements, and consistent with the limited observations for oil mass trapped at depth by hydrocarbon types and a chemical signature

14

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

(e.g. diocytl sodium sulfosuccinate, DOSS) from the dispersants used to treat the oil. French McCay et al. (2016) compared SIMAP model predictions to observations for the oil mass balance, surface (oating) and subsurface oil, sediment oil interaction, and shoreline oiling. The model was fully three dimensional and included subsurface oil droplet transport and dispersion, oil droplet surfacing, surface oil spreading, evaporation of volatiles from surface oil to the atmosphere, stranding of oil on shore- lines, emulsi cation of oil, entrainment of oil as droplets into the water column, re-surfacing of oil, dissolution of soluble components into the water column, volatilization from the water column to the at- mosphere, partitioning of oil between water and sediment, sedimenta- tion of oil droplets, and degradation. The oil was characterized in terms of both soluble and insoluble pseudo-components. Output of the fate model included the location and dimensions of oating oil, mass and concentrations of hydrocarbon constituents in water, and uxes of hydrocarbons to the atmosphere and sediment over time. Neareld pre- dictions necessary as input to the model were provided by the blowout model. The subsurface observational data allowed for comparison to model predictions of both the oil mass and its composition. Model simulations were driven by a variety of different wind(four) and hydrodynamic(six) models that covered the study area and duration of the spill. A network of 18 acoustic Doppler current proler (ADCP) measurement locations and six NOAA National Buoy Data Center (NBDC) meteorological observation buoys were available to validate the wind and current model predictions or be assimilated into the models. The time varying oil release rate and locations and the subsur- face and surface dispersant treatment amounts and strategies were also speci ed based on available data and simulations from the blowout model. Removal of oil at the surface from mechanical recovery or burn- ing were based on eld observations. Both model simulations and ob- servations ( Payne and Driskell, 2015a, 2015b; Horn et al., 2015a, 2015b ) showed that the highest total hydrocarbon concentrations were near the wellhead, decreasing with distance from the spill site as oil droplets rose to the surface, diluted, and degraded. Smaller droplets were transported farther from the wellhead by currents before rising to the surface. Complex patterns arose as rising droplets were transported and sheared by the variable currents at different depths. Once reaching the surface the oil transport was dominated by the near surface currents and winds. The oil mass balance, and observations of the surface and shoreline oiling pattern were consistent with the extensive remote sensing data which provided information on the spatial extent and thickness (volume per unit surface area) of the oil (e.g. Leifer et al., 2012; Garcia-Pineda et al., 2013a, 2013b; Graettinger et al., 2015; MacDonald et al., 2015; Svejkovsky et al., 2012, 2016; Sun et al., 2016). The subsurface oil concentrations were consistent with the lim- ited observations with both model predictions and observations show- ing substantial patchiness in space and time. Model predictions, in agreement with observations, showed very limited sediment oil inter- action. Spill model predictions showed considerable sensitivity to pre- dicted model currents at depth and to winds and currents at the surface as evidenced by tests performed using the varying meteorolog- ical and ocean current model predictions as input. Predicted currents at depth showed substantial differences in both spatial and temporal pat- tern and speeds compared to available (ADCP) current observations. The poor performance of the current model predictions at depth were by far the weakest link in predicting subsurface oil transport.

5. Selected future directions in spill modeling

Presented below are some thoughts on future directions in spill modeling. Given space limitations, two ideas are articulated that will ad- dress fundamental problems within the framework that is employed by many existing spill models. The rst is focused on how spreading is handled with LEs and the second on predictions of oil transport using random walk methods.

5.1. LEs and spreading

Most spill models available today employ Lagrangian elements (LE) (spillets) to represent the oil. Each LE represents a lot (mass or volume) of oil that is released from the source and tracked through time. LEs can be released and tracked either at the surface (lots of oil) or subsurface (normally as droplets); the later as a result of a blowout release or en- trainment of oil from the surface. LEs at the surface are tracked as indi- vidual lots of oil typically characterized by their radius, thickness, and composition (properties) of the oil. Surface and subsurface LEs are moved in space and time as a result of environmental forcing (winds and currents). The oil in the LEs can be moved between the surface and the water column by oil entrainment and in the other direction by resurfacing of buoyant oil droplets. The surface area resulting from the oil in a surface LE is determined by the oil volume/mass and spreading rate for each LE. The mass in each LE generally decreases with time as oil is evaporated (transferred to the atmosphere), entrained (dispersed into the water column), or biodegraded. Entrainment and evaporation are dependent on the oiled surface area and associated oil thickness. The oil mass in a LE can be further broken down and tracked by various subcomponents of the oil if desired. This is often done for subsurface oil. The spreading approach used in most spill models follows the labo- ratory and theoretical work by Fay (1969) and Hoult (1972a, 1972b) or some variant thereof, where spreading is approximated by the force bal- ances in the three stages: inertia- pressure gradient, pressure gradient -viscous, and viscous -surface tension. Most spill models exclude the rst stage, since the time scale for this stage is short compared to the model time step (10's min to hrs) and the third stage since the oil is very thin and readily subject to breakup into patches and hence be- comes discontinuous. Spill models approximate spreading in the inter- mediate regime by using the functional form proposed by Fay and Hoult, with a scaling coef cient based on laboratory or eld experi- ments. The scaling coefcients vary widely depending on the laboratory experiments (1 or 2-D release) and the particulars of the eld experi- ment. The spreading algorithm performs best for instantaneous releases of a xed amount of oil and worst for complicated time dependent re- leases. The former is a very rare event and the later much more common. In implementing spreading for individual LEs, there are several prac- tical challenges in existing spill models that raise concerns about the re- sults: 1. Spill models treat LEs as independent elements. The radius of an LE, determined by spreading and the spacing between LEs may lead to situations where there is overlap between the spreading radii of LEs in close proximity. Since the LEs do not know if there is another LE present or its proximity, there is no feedback among the LEs. Interaction among adjacent LEs is clearly important if the spacing between them is suf- ciently close that they exert forces on one another. Lynch et al. (2015) refers to this as the crowding problem. 2. In LE based spill models, one key underlying principle is that as the number of individual LEs in- creases, the spill model predictions should asymptotically approach the correct solution to the underlying governing equations of motion for the surface oil. This of course is impossible to satisfy if the LEs do not know of the presence of nearby elements. One can clearly see that as the number of elements increases they become increasingly close to- gether, particularly in areas of thick oil, the LEs should inuence one an- other (most spill models perform ad-hoc corrections to the spreading algorithm for each LE to insure that this constraint is met). 3. Given the lack of interaction between LEs most spill models perform poorly in areas of thick oil, typically found near the spill source (either from a surface release or above a blowout where the larger oil droplets reach the surface) and do much better in the far eld where the oil is thinner and more widely dispersed. In order to address this problem one needs to develop a mechanism for LEs to interact with one another that is con- sistent with the underlying equations of motion for the oil and results in the solution converging to the true solution as the number of LEs increase.

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

15

There has been very active development in the last several decades to model complicated free surface ow problems using Smooth Particle Hydrodynamics (SPH) methods (Liu and Liu, 2003), with the core con- cept to model the interaction between discrete particles that represent the uid (system) of interest (Gingold and Monaghan, 1977). The SPH strategy is well suited to the current problem of determining the inter- actions between oil LEs. Maderich et al. (2012) have in fact applied the SPH to very simple one- and two-dimensional oil spreading problems and shown excellent comparison with theoretical solutions. Maderich et al.'s simulations used standard SPH algorithms and employed 10 5 particles (LEs). This approach shows promise to address the crowding problem inherent in LEs method (Lynch et al., 2015). Key attributes of the method are provided in the Supplemental material section.

5.2. Modeling transport of LEs using random walk and random ight methods

The widely used random walk ( ight) method has several shortcomings when implemented in most spill models. These include:

1. Not applicable to spatially or temporally varying dispersion elds . As documented by Hunter et al. (1993) the random walk method is not applicable when either the horizontal or vertical dispersion coef- cients are spatially or temporally varying. In these situations con- ventional applications of random walk methods result in the arti cial accumulation of LEs in areas of low dispersion. For many of the more rudimentary spill models this is not an issue since the dispersion coef cients are user speci ed and constant in both the vertical and horizontal direction (although of different magnitudes). The exception might be in the vertical where dispersion is specied in discrete steps or zones (near sea surface layer, from surface layer to bottom of Ekman layer, below surface mixed layer to the top of bottom boundary layer, etc.). If the output from hydrodynamic models is used, however, the issue becomes more important since these models predict temporally and spatially varying dispersion. The horizontal values typically vary with horizontal shear ( Smagorinsky, 1963) and the vertical values are dependent on the turbulent closure scheme employed in the hydrodynamics model. The horizontal values can change substantially in the presence of eddies and strong current features, while the vertical values primar- ily depend on current shear and water column stratication. The rate of change of vertical dispersion can be substantial in the zone very near the sea surface, where stratication is signicant, or where the blowout plumes are trapped in the water column.

2. Problem with estimating concentrations in the water column. In most spill models there is a need to estimate the concentration of contam- inants in the water from constituents in either the oil droplets or gas bubbles that have undergone dissolution in the water. In estimating the rate of dissolution, the concentration of the contaminant of inter- est must be estimated both in the water column and in the droplets and bubbles. It is this gradient of concentration that drives the disso- lution process. In many spill models the concentrations in the water column are assumed to be small (hence the solution is dilute) and hence water column concentrations are not needed and therefore not estimated. The dilute solution assumption is clearly applicable in the far eld where concentrations and the density of LEs is low, but is not appropriate where concentrations and their gradients are large, such as in the blowout plume and trapping area, and the near sea surface zone where oil is entrained by breaking waves. If models do incorporate any feedback from water column concentrations on dissolution rates, the LEs are typically mapped to a grid that covers the area of interest and concentrations are estimated by partitioning the mass of each LE to a grid and then dividing by the grid volume. Most models that perform impact assessment use water column con- centrations of speci c components of the oil and gas and their

associated toxicity to determine impacts on organisms. The transla- tion of LEs to concentrations is critical to these analyses.

3. Problem with modeling dispersion where edge effects or transport bar- riers are present. In conventional spill models, LEs often end up being concentrated along discrete lineal features, such as the coast line, fronts, intrusions, or the like in the horizontal or the sea surface or sea bed in the vertical. These are areas where dispersion rates vary rapidly, but of course are not represented in most spill models, since dispersion coefcients are typically constant. Spill models deal with this situation by a variety of ad-hoc, local adjustments. As an exam- ple, oil droplets (LEs) that are transported on shore are either kept there (stick) or placed some arbitrary distance offshore and the sim- ulation continues. Oil often accumulates along the shoreline in these cases, not due to the physics but to the model treatments of LEs.

4. Sensitivity to numbers of LEs. All LE-based methods are sensitive to the number of LEs used in the simulation. The general rule of thumb is that as the number of LEs increase the model predictions asymptoti- cally approach the solution to the advective diffusion equation. It is easy to show this convergence for simple test cases (point releases in uniform ow) but considerably more difcult for complex, time varying releases that incorporate oil fate processes. For this situation, the normal strategy is to study the sensitivity of model predictions to the number of LEs employed in the simulation. This issue is critical to all applications of LE methods. Hunter et al. (1993) provide back-of- envelope procedures to estimate the number of LEs required. It is clear that this sensitivity is also dependent on the selection of disper- sion coefcient used in the simulation. Spaulding and Pavish (1984) have developed an alternate solution to the conservation of constituent mass equation that addresses the concerns noted above. The approach, described in more detail in the Supplemental materials section, reformulates the dispersive term in the form of its equivalent advective transport. In order to im- plement the method a procedure is needed to rapidly estimate the concentrations in the water column given the spatial distribution of LEs. Three separate approaches might be used here: standard rectangular

gridding, a SPH smoothing kernel (Liu and Liu, 2003), and triangulation tessellation ( Galt, 2012, 2015 ). Each is outlined and its attributes are noted in the Supplemental material section. Implementation of the Spaulding and Pavish (1984) method will require greater computation- al resources but will substantially improve model performance for actu- al spill releases in complicated near coast ow elds being predicted by the current generation of hydrodynamic models.

6. Summary and conclusions

The review has found that the structure of spill models based on tracking Lagrangian elements (LEs) to represent the surface oil (lots of oil, spillets) and subsurface oil (droplets) continues to be the method most often used. The fate processes are then employed to predict the transfer of oil from one environmental compartment to another (evap- oration transfers oil from the sea surface to the atmosphere, entrain- ment transports oil from the sea surface to the water column, etc.) or the change in the properties (e.g. density, viscosity, and surface tension) of the oil (emulsi cation changes in the water content of the oil and thus its viscosity). The more highly developed models are increasingly using more detailed characterization of the oil in terms of its hydrocar- bon composition. The motivation in these cases is better characteriza- tion of the loss of the more volatile components and their impact on marine organisms. Lessons learned in the development and application of spill models have been summarized and presented in Section 2 and not repeated here. The most important of these is that careful tracking of the oil mass balance (total and by components, as appropriate) and transfer among the major environmental compartments (atmosphere, surface, water column, sea bed, and biodegradation) is critical.

16

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

There have been advances in modeling oil spreading via subsurface transport and resurfacing, but these have yet to appear in most spill models. Modeling of evaporation has not signi cantly changed, with strategies to model bulk losses based on generic oil characteristics and pseudo-component based methods that include a more detailed charac- terization of the oil typically used. The later have experienced an in- crease in the number of components tracked to meet the need for impact modeling. There have been signicant advances in modeling of oil entrainment and associated droplet size distributions. The core methods in both cases are empirically based but have bene ted from investigators revisiting the underlying formulations to restructure them using stan- dard dimensional analysis methods. This has reduced the uncertainty in model coefcients and made them more directly related to the phys- ical and chemical processes of importance. These advances have been aided substantially from data that has become available from small and large scale laboratory experiments and real spill events. The droplet size modeling has advanced to the point that unied models are now emerging that apply to both surface (entrainment) and subsurface (blowout) releases and that can explicitly address treatment with dispersant. Recent applications of state of the art models to the DWH spill have shown the model's ability to capture the major transport and fate paths for this spill but highlighted the problem with sparse water column ob- servations and the poor predictive performance of hydrodynamic models for currents at depth. The utility of remotely sensed measure- ments of spill oil location and thickness have resulted in substantial im- provements in providing synoptic maps of surface oil distribution. Several suggestions about important future directions in spill model have been proposed with a focus on a more accurate characterization of the spreading and transport of oil using random walk methods. Once implemented the approach outlined can signicantly advance our abil- ity to model spill spreading and dispersion from temporal and spatially varying releases and in the presence of complex coastal current systems. The proposed directions are readily realizable given the fact that the building blocks for each are available and well developed.

Acronyms

ADCP acoustic Doppler current proler ADROP oil particle aggregate model AL1-AL8 aliphatic compounds (insoluble, high molecular weight) AR1-AR9 aromatic compounds (soluble and semi-soluble, low molec- ular weight) ASA RPS Group Applied Science Associates, RPS Group API American Petroleum Institute BOT Black Oil Table BTEX Benzene, Toluene, Ethylbenzene, and Xylenes BSEE Bureau of Safety and Environmental Enforcement BOEM Bureau of Ocean Energy Management EDS Environmental Data Server

DOSS

diocytl sodium sulfosuccinate

DWH

Deepwater Horizon oil spill

FTLE Finite Time Lyapunov Exponent GIS geographic information system GNOME/ADIOS General NOAA Operational Modeling Environment/ Automated Data Inquiry for Oil Spills (GNOME/ADIOS) HFR high frequency radar Holocam holographic imaging camera IOOS integrated ocean observing system LCS Lagrangian Coherent Structures LE Lagrangian elements LTRANS Larval TRANSport model system MC252 Macondo oil spilled during the Deepwater Horizon spill NDBC National Data Buoy Center

NRC

National Research Council

NOAA

National Oceanic and Atmospheric Administration

Oh

Ohnesorge number

OPA

oil particle aggregates

OILMAP DEEP Deep water blowout model OILMAP/SIMAP Spill Impact Model Application Package/Oil Modeling Application Package OSCAR Oil Spill Contingency and Response Model

PAH

PC

RD

Re

SPH

SI

SSH

SST

VDROP oil droplet model

We

Petroleum Aromatic Hydrocarbons pseudo-component method relative dispersion Reynolds number Smooth Particle Hydrodynamics emulsion stability index sea surface height anomaly sea surface temperature

Weber number

Acknowledgements

This work was supported in part by a contract awarded to RPS-Ap- plied Science Associates (ASA) by the Bureau of Ocean Energy Management (BOEM) for a project entitled: Simulation Modeling of Ocean Circulation and Oil Spills in the Gulf of Mexico , under contract Number M11PS00019. The manuscript bene tted greatly from an in depth review performed by Zhengkai Li, ASA/RPS Associate. The views and conclusions contained in this document are those of the author and should not be interpreted as representing the opinions or policies of the U.S. Government, nor does mention of trade names or commercial products constitute endorsement or recommendation for use.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx.

References

Afenyo, M., Veitch, B., Khan, F., 2015. A state of the art review of the fate and transport of oil in open and ice covered waters. Ocean Eng. November 2015. http://dx.doi.org/10.

1016/j.oceaneng.2015.10.014.

ASCE, 1996. State-of-the-art review of modeling transport and fate of oil spills. J. Hydraul. Eng. 122, 594609. Beegle-Krause, C.J., Peacock, T., Allshouse, M., 2011a. Exploiting Lagrangian Coherent Structures (LCS) for the calculation of oil spill and search-and-rescue drift patterns in the ocean. Proceedings of the 34th Annual Arctic and Marine Oil Pollution Confer- ence, October 46, 2011, Banff, Alberta, Canada. Beegle-Krause, C.J., Boyer, T.P., Garcia, H.E., Barker, C.H., MacFadyen, A., Payton, D., 2011b. Deepwater Horizon MC 252: understanding the spill below the surface. Proceedings of the 34th Annual Arctic and Marine Oil Pollution Conference, October 46, 2011, Banff, Alberta, Canada. Brandvik, P.J., Johansen, Ø., Leirvik, F., Farooq, U., Daling, P.S., 2013. Droplet breakup in subsurface oil releases part 1: experimental study of droplet breakup and effective- ness of dispersant injection. Mar. Pollut. Bull. 73, 319326. Carlson, D.F., Fredj, E., Gildor, H., Rom-Kedar, V., 2010. Deducing an upper bound to the horizontal eddy diffusivity using a stochastic Lagrangian model. Environ. Fluid Mech. 10 (5):499520. https://link.springer.com/article/10.1007/s10652-010-9181-0. Couliette, C., Lekien, F., Paduan, J., Haller, G., 2007. Optimal pollution mitigation in Monterey Bay based on coastal radar data and nonlinear dynamics. Environ. Sci. Technol. 2007 (41), 65626572. Chen, F., Yapa, P., 2002. A model for simulating deepwater oil and gas blowouts part II:

M.L. Spaulding / Marine Pollution Bulletin 115 (2017) 719

17

Gingold, R.A., Monaghan, J.J., 1977. Smoothed particle hydrodynamics: theory and appli- cation to non spherical stars. Mon. Not. R. Astron. Soc. 181, 375389. Hoult, D.P., 1972a. Oil spreading on the sea. Annu. Rev. Fluid Mech. 59, 341348. Fingas, M. (Ed.), 2011a. Oil Spill Science and Technology, Prevention, Response and Clean- up. Elsevier, Boston, MA (ISBN: 978-1-85617-943-0). Fingas, M., 1996. The evaporation of crude oil and petroleum products. PhD Dissertation. McGill University, Montreal, Canada, p. 181. Fingas, M., 1997. The evaporation of oil spills: prediction of equations using distillation data. Proceedings of the 20th Arctic and Marine Oil Spill Program (AMOP) Technical Seminar. Environment Canada, pp. 120. Fingas, M.F., 1998. Studies on the evaporation of crude oil and products: II. Boundary layer regulation. J. Hazard. Mater. 57, 4158. Fingas, M.F., 1999. The evaporation of oil spills: development and implementation of new prediction methodology. Proceedings of the 1999 International Oil Spill Conference. American Petroleum Institute, Washington, D.C., pp. 281287. Fingas, M.F., Fieldhouse, B., 2011. Water-in-oil emulsions: formation and prediction. Pro- ceedings of the 34th Arctic and Marine Oilspill Program Technical Seminar. Fingas, M., Fieldhouse, B., 2009. Studies on crude oil and petroleum product emulsions: water resolution and rheology. Colloids Surf. A Physicochem. Eng. Asp. 333, 6781 (2009). Fingas, M., 2011b. Models for water-in-oil emulsion formation, Chapter 10. In: Fingas, M. (Ed.), Oil Spill Science and Technology. Gulf Publishing Company, NY, NY, pp. 243273 (2011). Fay, J.A., 1969. The spread of oil on a calm sea. Oil on the Sea. Plenum Press, NY, pp. 53 63. French, D.P., Rines, H., Masciangioli, P., 1997. Validation of an orimulsion spill fates model using observations from eld test spills. Proceedings of the Twentieth Arctic and Marine Oil Spill Program (AMOP) Technical Seminar. Environment Canada, Ottawa, ON, pp. 933961 (1997). French McCay, D.P., 2003. Development and application of damage assessment modeling:

French McCay, D.P., 2004. Oil spill impact modeling: development and validation. Environ. Toxicol. Chem. 23 (10), 24412456 (2004). French McCay, D.P., Rowe, J.J., 2004. Evaluation of bird impacts in historical oil spill cases using the SIMAP oil spill model. Proceedings of the Twenty-Seventh Arctic and Marine Oil Spill Program (AMOP) Technical Seminar. Environment Canada, Ottawa, ON, pp. 421452 (2004). French-McCay, D.P., Mueller, C., Jayko, K., Longval, B., Schroeder, M., Payne, J.R., Terrill, E., Carter, M., Otero, M., Kim, S.Y., Nordhausen, W., Lampinen, M., Ohlmann, C., 2007. Evaluation of eld-collected data measuring uorescein dye movements and disper- sion for dispersed oil transport modeling. Proceedings of the 30th Arctic and Marine Oil Spill Program (AMOP) Technical Seminar. Emergencies Science Division, Environ- ment Canada, Ottawa, ON, Canada, pp. 713754. French McCay, D.P., Jayko, K., Li, Z., Horn, M., Kim, Y., Isaji, T., Crowley, D., Spaulding, M., Decker, L., Turner, C., Zamorski, S., Fontenault, J., Shmookler, R., Rowe, J.J., 2015. Tech- nical Reports for Deepwater Horizon Water Column Injury Assessment WC_TR14:

Modeling Oil Fate and Exposure Concentrations in the Deepwater Plume and Cone of Rising Oil Resulting From the Deepwater Horizon Oil Spill. DWH NRDA Water Column Technical Working Group Report. Prepared for National Oceanic and Atmospheric Administration by RPS ASA, South Kingstown, RI, USA. September 29, 2015. Administrative Record no. DWH-AR0285776.pdf. https://www.doi.gov/ deepwaterhorizon/adminrecord. French McCay, D., Horn, M., Li, Z., Crowley, D., Spaulding, M., Mendelsohn, D., Jayko, K., Kim, Y., Isaji, T., Fontenault, J., Shmookler, R., Rowe, J., 2016. Simulation Modeling of Ocean Circulation and Oil Spills in the Gulf of Mexico Appendix VI Data Collection, Analysis and Model Validation. Prepared by RPS ASA for the US Department of the Interior, Bureau of Ocean Energy Management, Gulf of Mexico OCS Region, New Orleans, LA. OCS Study BOEM 2017 (in Review). Galt, J.A., 2015. Triangular tessellation documentation, part II clustering on boundaries. Genwest Technical Publication 15-001. Genwest Systems Inc., Edmonds, WA. Galt, J. A. and R. Overstreet, 2011. Development of spreading algorithms for the response options calculator (ROC), by Genwest Systems, Inc., (Genwest Technical Publication

11-003).

Galt, J.A., 2012. Triangular tessellation documentation. Technical Report 11-001. Genwest Systems, Inc., Edmonds, Washington. Garcia-Pineda, O., MacDonald, I., Hu, C., Svejkovsky, J., Hess, M., Dukhovskoy, D., Morey, S.L., 2013a. Detection of oating oil anomalies from the Deepwater Horizon oil spill with synthetic aperture radar. Oceanography 26 (2):124137. http://dx.doi.org/10.

5670/oceanog.2013.38.