Wind drag in oil spilled ocean surface and its impact on wind-driven circulation

The drag coefficient is a key parameter to quantify the wind stress over the ocean surface, which depends on the ocean surface roughness. During oil spill events, oil slicks cover the ocean surface and thus change the surface roughness by suppressing multi-scale ocean surface waves, and the drag coefficient is changed. This change has not been included in the current ocean circulation models. In this study, such change in sea surface roughness is studied by satellite remote sensing via synthetic aperture radar (SAR) data to quantify the changes in the wind effect over the oil-covered ocean surface. The concept of effective wind speed is introduced to quantify the wind work on the ocean. We investigate its influence on the wind-driven Ekman current at the ocean surface. Using observations from the Deepwater Horizon oil spill (2010) as an example, we find that the presence of oil can result in an effective wind speed of 50%~100% less than the conventional wind speed, causing overestimates by 75%~100% in the wind driven Ekman current. The effect of such bias on oil trajectory predictions is also discussed. Our results suggest that it is important to consider the effect of changes in the drag coefficient over oil-contaminated areas, especially for large-scale oil spill situations.


Introduction
Oil and oil-based products are heavily relied on by modern humans. In 2017, 98 million barrels of oil were consumed each day (British Petroleum 2018). One-third of the oil and gas extracted worldwide comes from offshore sources (Lange et al. 2014). In the Gulf of Mexico alone, there are more than 7000 platforms installed. Meanwhile, the operation and transportation of the oil is not worry free. Spills, leaks, and explosions happen every day. Most of them have minor impacts, but some have major consequences. For example, on 20 April 2010, the Macondo oil well experienced a massive blowout, causing the explosion and collapse of the Deepwater Horizon (DWH) oil rig, releasing approximately 184.8 million gallons of oil (Crone and Tolstoy 2010), creating a devastating threat to the marine environment and human society along the coast of the Gulf of Mexico. With society's increasing demand for energy resources, more gas and oil production has increasingly shifted to deeper-water regions of the ocean in recent years, leading to a greater potential threat to the environment. When a marine oil spill does occur, it is vital to understand and predict the trajectory of the spill, to help and guide mitigation efforts and to assist responders with the allocation of their limited resources. A marine oil spill model is necessary to estimate such information.
A number of oil spill model systems have been developed by different nations and stakeholders to help predict the transport and weathering of oil that is released in an oil spill, for example, the MEDSLIK-II (Mediterranean Decision Support System for Marine Safety) (De Dominicis et al. 2016) and GNOME™ (General NOAA Operational Modeling Environment). Spill models are typically structured as an integrated series of algorithms describing individual fate and transport processes, which include the physical processes of transport, spreading, and multiple biological and chemical weathering processes, such as evaporation, dissolution, emulsification, degradation. Spaulding (2017) recently presented a general overview of the progress in the understanding and modelling of these multiple processes. Among all these processes, ocean circulation is a major component of the oil spill model system. Indeed, ocean circulation is fundamental to planning mitigation strategies and to determining both the landfall of oil and the movement of oil toward biologically sensitive areas in deep and shallow waters (Liu et al. 2011a). During the DWH oil spill disaster in 2010, the ocean circulation component alone was used for oil trajectory forecasts as a rapid response measure to guide the efforts to mitigate the effects of the spill and to guide ship surveys (Liu et al. 2011b). Various efforts have been taken to improve the modelling of the transport of oil as released by a spill. For example, by adjusting the spatial resolution of the ocean circulation model, De Dominicis et al. (2016) were able to make recommendations about high spatial resolution for better performance. Liu et al. (2011b) suggested that a multiple model ensemble technique is useful to reduce uncertainties in the forecasts of oil trajectories. Based on the experience of oil spill modeling for the DWH oil spill, Liu et al. (2011aLiu et al. ( , 2011b suggested that it is important to improve the wind-forcing functions to drive the ocean models.
As we know, wind-forcing is a key driver for the ocean circulation, by transferring kinetic energy and momentum into the ocean from the atmosphere, by exerting stress with magnitude: where u 10 is wind speed at 10 m above the sea surface, u s is the sea surface currents, ρ air is the density of air, which is 1.22 kg m −3 , and C d is the drag coefficient between air and ocean.
In ocean circulation models, the wind stress formula is well adapted with u 10 directly taken from weather prediction models, and C d from various parameterization schemes designed specifically for ocean surface water (e.g., COARE bulk air-sea flux algorithm) (Fairall et al. 2003). Error from the atmospheric numerical model will induce uncertainty in the modeled wind speed (Mayer et al. 2017). To accommodate such uncertainty in the oil trajectory predictions, NOAA includes a representation of the potential error in the predictions due to the inherent uncertainties in the input wind speeds (e.g., suppose wind speed is in error by 30%) ( fig. 7 in Macfadyen et al. 2011).
Many studies have focused on the effect of the wind on the oil trajectory modeling. Le Hénaff et al. (2012) suggested the importance of including wind-induced current drift in oil trajectory modeling. They suggest that less oil will be advected into the Loop Current when the wind-induced Stokes drift is considered. Weisberg and Liu (2017) found that wind-induced Stokes drift also plays an important role in advecting oil onto the northern Gulf of Mexico beaches. Kim et al. (2014) andDe Dominicis et al. (2016) studied an additional related oil drift factor, directly caused by wind (skin) drag, acting on the oil floating on top of the water.
On the other hand, from eq. (1), the wind work into the ocean also depends on the drag coefficient C d or the related aerodynamic roughness z 0 . Studies in various fields show the key role of surface roughness in air-sea interactions, for example, the dependence of sea surface roughness on wave development (Donelan et al. 1993), the reduced drag coefficient under high wind speeds (Powell et al. 2003), saturated radar backscattered signals in high wind speeds (Shen et al. 2009a(Shen et al. , 2009b, the effect of a sudden rain downpour on wind wave generation (Wu et al. 2017;Cavaleri et al. 2018), the attenuation of swell waves by rain (Cavaleri and Bertotti 2017). The ocean response to the surface wind is significantly altered due to changes of the surface roughness, under different meteorological conditions.
Another way to change the aerodynamic roughness is the presence of different materials on or in the water, for example, oil. It is well known that oil calms waves. Experiments (Barger et al. 1970) show that a biodegradable slick modifies the boundary dynamics and has impact on the turbulent boundary layer. The effects of the slick are modifications to the profile roughness length z 0 . A total absence of capillary waves and breaking waves is observed within the slick. The presence of a surface slick dampens small surface waves and decreases the profile roughness length. The slick interferes with the mechanism of the generation of wind waves. In the latter case, the forces exerted on the sea surface by the atmospheric winds do work on the ocean surface and generate wind waves. In return, reactions to these forces act on the atmosphere to generate turbulence. This process appears to depend on the existence of capillary waves. The introduction of an oleyl alcohol sea slick, as in Barger et al. (1970)'s experiment, with its strong damping and suppression of capillary waves, appears to completely destroy this process. As a result, no wind energy is transferred into the ocean in these oil-covered areas. Alpers and Hühnerfuss (1989) found that oil not only dampens short capillary waves, but that the longer waves are also suppressed, by the so-called Marangoni wave damping theory. Marangoni damping leads to a deformation of the spectral form in the short-gravity-wave region. The wave system responds to such deviations from equilibrium with an increase in the energy flux towards smaller waves. Thus, energy is drawn from the waves with low wave numbers, which in turn leads to damping of longer waves. By introducing the same mechanism, Cox et al. (2016) successfully explained the impossible rescue of a fishing boat in stormy conditions in 1883, whereby with the aid of pouring oil on the ocean surface they were able to reduce the energy flux into ocean waves. It is thus evident that the presence of oil changed the surface roughness. Therefore the question is whether we can quantify the changes of the roughness or drag coefficient. Moreover, how will this change impact the response of the ocean circulation, which uses the surface wind stress as a driving force? We will try to tackle these questions in this paper.
There have been no direct measurements of the changes of drag coefficient (or surface roughness) in oil-covered areas; however, the changes of surface roughness not only affect the air-sea interaction dynamics, but also modify the remote sensing measurements, which sense ocean dynamics via modulation with the small scale waves at the sea surface. Synthetic aperture radar (SAR) observes ocean dynamics through the Bragg-scattering mechanism, where radar waves resonate with similar scale ocean surface waves. For C band SAR, this corresponds to capillary waves with 3-5 cm wavelength. The presence of oil on the sea surface will dampen such length-scale waves, resulting in reduced intensity of radar backscattered signals in the oil spill area. Such mechanism makes oil spills detectable by SAR. SAR has become a very useful tool for monitoring the extension of oil on the ocean surface during oil spill accidents.
In this paper, we utilize SAR data to study the sea surface roughness changes due to oil, with a focus on its impact on wind stress over the ocean surface in oil-covered areas. We start by introducing data and methodology to estimate wind stress changes in an oil-covered area from SAR in Sect. 2, followed by estimations of the Ekman current in oilcovered areas in Sect. 3. We discuss how this process leads to bias in oil trajectory predictions in Sect. 4. Discussions and conclusions are given in Sects. 5 and 6.

Data
We collected RADARSAT SAR data on 20 April and 15 July 2010, during the DWH oil spill event. The spatial cumulative coverage of the oil spilled area is shown in Fig. 1 (data courtesy NOAA), with coverage of the SAR images studied in this paper. RADARSAT (1 and 2) SAR is a C band (5.3G Hz) SAR. It provides all-day (day and night) and all-weather (sunny, cloudy, rainy, etc.) observations of the Earth's surface. It transits radar signals towards the targets, and receives radar backscattered signals. RADARSAT-2 SAR is a full polarization (pol) instrument. It is capable of conducting measurements in all pols (i.e., HH, HV, VH, and HV). Because of its multiple pol and multiple operation modes, RADARSAT-2 SAR data have been widely used in many fields, for example, defence and security, oil and gas, mining, natural resources, disaster management, aviation, agriculture. For marine applications, RASARSAT SAR has been widely used in studies of oil spill detection and monitoring (e.g., Zhang et al. 2011), ocean surface winds (e.g., He et al. 2005), and waves ; hurricanes (e.g., Shen et al. 2006Shen et al. , 2009aShen et al. , 2009bShen et al. , 2014aShen et al. , 2016; ship and wind turbine detection (e.g., Li et al. 2013;Vachon et al. 2014); marine algae detection (Shen et al. 2014b); and many mesoscale ocean dynamics, including internal waves, fronts, eddies, etc. For a detailed overview, readers are directed to refer to Jackson and Apel (2004).
Oil coverage is retrieved from SAR based on image analysis methods and compared with retrievals from NOAA (Streett 2011), which was operationally published during the DWH oil spill response period. The coverage shows the extension of oil spill patches over the sea surface. This pattern is similar to that derived from optical satellite products (Hu et al. 2011).
ERA-interim wind reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) are used, mainly to provide wind direction information for wind speed retrieval from SAR. The ERA-Interim reanalysis is produced by the ECMWF IFS (Integrated Forecasting System), which incorporates a forecast model with three fully coupled components for the atmosphere, land surface, and ocean waves. The system incorporates a state-of-the-art 4D-VAR assimilation scheme to accommodate multiple sources of observations, including in situ observations and satellite measurements. Dee et al. (2011) provide detailed information about the system. We use 10 m standard height wind vector data for the present study, at 3 hourly temporal intervals. The spatial resolution is 0.125°.
Satellite-derived geostrophic currents and Ekman currents from GlobCurrent are used to study the general pattern of ocean circulation when oil is not considered in the circulation model setup. GlobCurrent synthesizes multiple satellite altimeter and wind model inputs to estimate both geostrophic and Ekman-layer velocities (Feng et al. 2018). Altimetryderived current products were found to perform better than data-assimilative models in simulating the drifters during the 2010 DWH oil spill period (Liu et al. 2014). The dataset is particularly useful for this study, because it partitions wind-induced Ekman current from the general circulation driven by geostrophic balance. We focus on the former in this study.
In particular, we focus on two SAR images on 26 April and 21 May 2010, which represent the relatively initial stage of the DWH oil spill 6 days after the initial pipe blow-out, and 1 month after the blow-out, respectively, when the oil spill had expanded to a large spatial scale. Figure 2 shows previews of the RADARSAT-1 SAR image (HH pol) on 26 April and RADARSAT-2 SAR image (VV) pol on 21 May 2010. Features representing the oil spill patches are clear in both images. The nearby NDBC buoy shows wind speed of 11 m/s on 26 April at the time when the first SAR image was acquired (this is the highest wind speed observed when RADARSAT SAR has data) and 6 m/s on 21 May 2010 at the time of the second SAR image. During the DWH oil spill period, the highest wind speed reached 15 m/s on 7 July 2010. Unfortunately, no RADARSAT SAR image was acquired at this time.

Wind retrieval from SAR
RADARSAT-2 SAR data are firstly calibrated into normalized radar cross section data (NRCS, unit: dB) following the procedure suggested by the RADARSAT user Manuel (®MDA). A bilinear smoothing procedure is applied in each 3 × 3 image pixels to suppress speckle noise. The data are then used as inputs to a wind speed retrieval module (He et al. 2005). A large portion of RADARSAT-2 SAR images in our data collection are in dual-pol mode. However, data with cross-pol (VH or HV) are excluded from this study, due to the low signal level in oil-covered areas and the reduced sensitivity of cross-pol radar signals for wind speeds under 10 m/s (Shen et al. 2014a). For co-pol data (HH or VV), wind speed is retrieved based on the CMOD5 geophysical model function (GMF) (Hersbach et al. 2010) with wind direction obtained from ECMWF interim dataset.

1D Ekman current model
Wind-induced Ekman current from GlobCurrent is used to investigate the relative contribution of Ekman current in the ocean circulation of the Gulf of Mexico during the Fig. 1. Cumulative spatial coverage of DWH oil spills overlaid by two RADARSAT SAR images studied in this paper. DWH oil spill event. Moreover, a 1D Ekman current model is used to theoretically investigate the Ekman current response to different wind speeds, to help quantify the difference between the surface current and the associated drift between water-and oil-covered areas.
The model is based on the geophysical fluid dynamical equations of motion. Under steady state conditions, in the boundary layer of the upper ocean, horizontal gradients are small compared with vertical gradients. The motion of these currents is controlled by a balance between Coriolis force and friction, where A z is the vertical eddy viscosity, and we approximately use A z = 0.01 m 2 /s, according to Chu (2015).
The solution to these boundary layer equations for a northward-directed flow, for example, wind stress, is as follows: Therefore, based on wind stress exerted over the ocean surface, the associated Ekman current can be estimated. As an example, a surface wind speed of 10 m/s at 30°N will result in about 0.11 m/s Ekman current at the ocean surface, in a direction 45°to the right of the driving winds.
Satellite derived observations of an oil spill event can only observe the sea surface. Information about oil within the water column cannot be estimated. Therefore, we focus on the Ekman current at the sea surface (z = 0) hereinafter.

Wind stress retrieval in oil coverage areas
Wind stress is usually expressed as The wind profile law for neutral stratification is where u Ã = (τ/ρ) 1/2 is the friction velocity, κ = 0.4 is the von Karman constant, and z 0 is the aerodynamic roughness at the sea surface.
For practical applications in oceanography, an empirical formula is often used to calculate wind stress from wind speed, as given by eq. (1).
There is a unique relationship between z 0 and the neutral drag coefficient C d , Therefore, the key parameter C d for calculating wind stress is a measure of the aerodynamic surface roughness z 0 .
It is critical to understand the sea surface roughness, as wind work on the ocean depends on roughness. The process appears to depend on the existence of capillary waves on the ocean surface. Any surfactants on the ocean surface can change the roughness of the sea surface, and thus change wind work into the ocean.
The parameterization of C d is currently a very active topic in the ocean and atmospheric science communities. Although a universal consensus has not been achieved, the most widely cited relationships are possibly those proposed by Smith (1980Smith ( , 1988, Large and Pond (1981), Yelland et al. (1998), and in particular the Coupled Ocean-Atmosphere Response Experiment (COARE) algorithm (Fairall et al. 2003). New results have routinely emerged to optimize the relationship between drag coefficient and wind speed (e.g., Foreman and Emeis 2010). The precise dependence of the drag coefficient on one or more variables (wind speed, wave age, wave slope, etc.) is an ongoing area of research (Sullivan and McWilliams 2010). The present study makes no attempt to advance the study of this specific field; instead, we adopt the most widely cited algorithm, COARE, to study how the presence of oil on the ocean surface can change the wind stress, which can be evaluated by other formulas.
In water, the surface roughness is governed by the amplitude of the short capillary gravity waves (Wu 1969). Field observations indicate that the contribution of the mean square slopes from capillary-gravity waves is a significant portion of the total mean square slopes of the ocean surface (Hwang 1997). A relation between surface roughness and wind stress was first suggested by Charnock (1955). SAR is sensitive to ocean surface roughness. For moderate incidence angles on the ocean surface, the spectra of capillary waves that are in resonance with radar waves determine the strength of the radar backscattered signals, and concomitantly, the retrieved wind speed. Therefore, both radar signals and surface roughness are related to wind stress. It is therefore important to quantify the wind stress from radar measurements.
The methodology for wind speed retrieval from SAR has advanced and is widely accepted (e.g., Dagestad et al. 2013). Operational wind products are produced from SAR, for example, by the National SAR Wind Program in Canada (Khurshid et al. 2012) and by the Alaska Coastal SAR Program in the United States (Monaldo et al. 2015). Wind vectors are related to the radar backscattered signals by a GMF. Under moderate wind conditions, CMOD5.n is the most commonly used GMF, which is in the form of eq. (2) and σ 0 = CMODðu 10 , θ, ϕÞ where σ 0 is the normalized radar cross section (unit: dB), u 10 is wind speed at 10 m reference height above the sea surface, θ is wind direction, and ϕ is radar incidence angle. The wind stress can be obtained by applying the wind speed from eq. (8) into eq. (6).

Remote sensing of effective wind speed in oil areas
When oil is present, it dampens the ocean surface roughness. We define the effective wind speed u 10e as that which causes an equivalent sea surface roughness over a pure uncontaminated ocean water surface, compared with that generated by the actual in situ wind speed over the oil-covered area. By this definition, the roughness difference between the real wind speed in the atmosphere and the effective wind speed is balanced by the damping effect from the presence of oil.
Therefore, the effective wind speed can be obtained from the radar measurements as follows: σ 0-oil = CMODðu 10e , θ, ϕÞ (9) By this approach, we calculate the equivalent wind speed that "effectively" functions on the ocean surface, and which is the only mechanism by which wind work (energy, momentum, etc.) is put into the water. It is this wind speed that is to be considered by the ocean circulation model, to forecast the oil spill trajectory. Figures 3 and 4 present the results of wind retrieval from the RADARSAT SAR images on 26 April 2010 and 21 May 2010. For comparison, wind reanalysis data from ECMWF interim are shown in Figs. 3a and 4a and Figs. 3b and 4b. Generally, the SAR retrieved wind speeds are consistent with the wind reanalysis and buoy measurements (not shown) from NDBC buoys. On 26 April, the wind is blowing to the south, the coastal shading effect of the Mississippi Delta and associated strengthened river plume are seen as areas of decreased wind speeds in the SAR images. Such effects are diminished on 21 May when the wind direction turned to the north, when the counter-wind flowing of the river plume contributed to increased wind input into the ocean. Significant changes in the effective wind speeds over the oil-covered areas are seen in the wind speed fields retrieved from SAR. However, the wind speed over the ocean surface itself cannot explain this change, as wind speed from ECMWF (Figs. 3a, 3b and 4a, 4b) indicate homogeneous wind fields. The reduced effective wind speed suggests decreased ocean response to the wind forcing over the oil-covered areas. The difference between the SAR-retrieved effective wind speed and the ECMWF wind speed is generally within ±1 m/s in water-covered ocean areas (Figs. 3d and 4d) but more than 5 m/s lower in oil-covered areas. The effective wind speed over oil-covered areas is about 50% less than that over the uncontaminated ocean waters on 26 April and 100% less on 21 May.

Suppressed Ekman current due to oil presence
Based on the effective wind speed obtained in the previous section, we are able to estimate the changes in the Ekman current in the oil-covered areas, as compared with the oilfree situation. This consideration is presently a missing component in the ocean circulation module of all oil spill models.
According to Ekman transport theory, wind stress on the ocean surface is proportional to the square root of the wind speed; and Ekman current speed is linearly related to wind stress. The black line in Fig. 5a shows an example of Ekman current variations with wind speed, at latitude of 20°N. For wind speed of 10 m/s, the associated Ekman current is about 13 cm/s; for wind speed of 18 m/s, the Ekman current is about 38 cm/s. For the oil spill case on 26 April 2010, our study shows that the effective wind speed is about 50% of the actual wind speed over the ocean surface. The associated Ekman current speed is about 4 cm/s. This leads to a difference of 9 cm/s for the Ekman current, which is 70% less. For the case on 21 May 2010, the wind speed surrounding the oil spills was about 5 m/s, while the effective wind speed in the oil spill areas was close to 0, leading to 100% reduction in the wind energy input. Under such circumstances, no Ekman current is generated, leading to 100% decrease in the Ekman current in oil-spill covered areas in this case (Figs. 5a and 5b). For wind speed of 18 m/s, if we also consider 50% reduction in effective wind speed, Fig. 5 indicates that the Ekman current will be reduced from 38 to 27 cm/s, corresponding to a reduction to 72% of the Ekman current. Figure 5 also shows that as the wind speed gets stronger, there is an equivalent reduction in the effective wind speed due to the oil coverage of the ocean (e.g., 50% for the black line in Fig. 5b), which results in stronger suppression of the wind-induced Ekman current.  Figure 6 provides spatial distributions of the differences for both the Ekman current estimated from ECMWF winds and SAR effective wind speeds. In oil-covered areas, the Ekman current is generally reduced. As previously stated, about 50%-100% less Ekman current is found in oil-covered areas. On 26 April, there is a significant decrease in the Ekman current, on the order of 10 cm/s due to the relatively high winds (∼10 m/s). On 21 May, an Ekman current difference of 2-3 cm/s is found in the oil-covered areas. When combined with the geostrophic current, the overestimation in Ekman currents not only changes the total current speed, but also changes the current direction. Figure 7 presents the geostrophic currents from the ®GlobCurrent dataset at the SAR observation times. In oil-covered areas, the geostrophic current on 26 April (Fig. 7a) is relatively weak, on the order of 10 cm/s. The difference for the Ekman current is also on the order of 10 cm/s. Therefore, when the weakened Ekman current is neglected, there is an overestimation by 50% in the total current. On 21 May 2010, the geostrophic current in the oil-covered area is about 0-20 cm/s (Fig. 7b). However, the difference in the estimated Ekman current is about 2-3 cm/s (Fig. 6). Such differences will not result in significant changes in the total current field.

Discussion
Laboratory experiments conducted in a wave tank suggest decreased surface roughness for water surfaces covered by crude oil (Charnotskii et al. 2016), resulting in less wind energy input into the water. Field experiments (Hünerfuss et al. 1989) suggest that oil dampens not only capillary waves on the ocean surface, but also lower frequency waves through the Marangoni wave damping effect (Alpers and Hühnerfuss 1989) and wave-wave interactions (Cox et al. 2016). Previous studies have made it clear that the presence of oil modulates the ocean response to winds. The present study shows that ocean circulation, especially Ekman current, can be significantly suppressed in the oil-covered areas. Such changes in ocean circulation can be important for oil trajectory predictions during a large-scale oil spill event.
SAR observes the ocean surface via the Bragg scattering mechanism, where the radar waves are resonant with the capillary waves on the ocean surface. By directly injecting kinetic energy into the capillary waves, wind information can be retrieved from SAR with high accuracy (Dagestad et al. 2013;Shen et al. 2016). Under moderate conditions, wind data retrieval from SAR has reached the operational application stage, for example, as demonstrated by the National SAR Wind Program of Canada (Perrie 2015) and the Alaska SAR wind project in United States (Monaldo et al. 2015). Oceanic processes, such as currents, can also contribute to ocean surface roughness, by modulating the effective wind speed exerted over the ocean surface and by inducing divergence or convergence as related to ocean surface waves. For radar remote sensing, roughness changes induced by ocean currents are generally much weaker than their wind-induced counterparts. This is the reason why SAR can be applied for detection of ocean current features only under low wind conditions. The key point of the present work is to introduce the "effective wind speed" over oilcovered areas. Although the wind speed may not be weak in the atmosphere, the wind work exerted on the ocean surface relies on ocean roughness, which is subject to change when covered by oil or dominated by oceanic processes. The "effective wind speed" discussed in this study practically includes both effects. By linking the dampened radar backscattered signals to the reduced surface roughness in oil-covered areas, we have retrieved the effective wind speed. The difference between the effective wind speed and the wind speed in the atmospheric boundary layer is that the former represents only the component that actually goes into the water. This is notable because radar only "feels" the wind that actually works on the ocean surface, instead of the estimated wind in the atmospheric model.
It is the effective wind speed that should be used as input to drive the ocean circulation models when oil is present on the ocean surface. This approach emphasizes the value of satellite-derived wind products as inputs for numerical ocean modelling. Should the satellite-derived effective wind not be available, the decreased drag coefficient over the oilcovered areas should be applied. Our study is based on RADARSAT SAR images obtained on 26 April and 21 May 2010. We show that there can be 50%-100% differences between the effective wind speed and the winds estimated by ECMWF. By considering the wind stress eq. (2) in Sect. 3, this result represents about 75%-100% decrease in the drag coefficient C d .
To accommodate such uncertainty in the oil trajectory predictions, we note that NOAA tends to include 30% of the wind speed for the potential error in the prediction (Macfadyen et al. 2011). Based on our analysis, the error could be much larger if the effect of oil on the wind drag is not considered in the ocean circulation model.
By comparing our analysis for the cases on 26 April and 21 May 2010, it is found that the change in the effective wind speed depends on the wind speed. Oil can block all the wind energy from entering the ocean from the atmosphere in low wind conditions, for example, on 21 May, when the wind speed was about 5 m/s. When wind speed is higher, for example, 10 m/s on 26 April, wind is still able to transfer energy into the ocean through the oil interface. Previous studies show that oil properties, such as thickness, category, density, concentration (Charnotskii et al. 2016), can all effect changes in the ocean surface roughness. Therefore, it is useful to study the effects of these parameters on the wind energy input into the ocean.
Changes in effective wind speed lead to different responses for the ocean circulation. As a result, the Ekman current can be directly suppressed. Such changes can eventually affect the oil trajectories, in complex ways. On the one hand, changes in the currents will directly result in different trajectories for the oil particles. Thus, changes occur in the location and timing of the of the oil's spread into a geographical areas. On the other hand, when the location and timing of an oil spill entering large ocean circulation areas changes, the ultimate trajectories of the oil spill can be significantly altered, as was the case for the DWH oil spill event. The Loop Current and associated eddies dominate the ocean circulation in the Gulf of Mexico. During the events related to the DWH oil spill, several eddies were evident surrounding the oil spill area, which are thought to be important for keeping the oilcontaminated water from entering the Loop Current. A bias in the prediction of the timing and location of oil spills can lead to very different patterns for ultimate spill trajectories. As a matter of fact, during the DWH period, many forecasting systems predicted massive oil quantities entering the Loop Current and undergoing vast spreading and land-falling on Florida coastal areas (Le Hénaff et al. 2012), which eventually did not happen. The neglect of the wind drag changes in these modelling exercises over the oil-covered areas may be one reason for these results.
Oil spill patches are able to flatten on the ocean surface under moderate wind speed. When wind exceeds 18 m/s, the oil is likely to be dispersed into the water column (Alpers and Hühnerfuss 1989). Under such situations, the oil will have limited capability to dampen the ocean surface waves. Therefore, the dynamical response of the sea surface to wind will no longer experience sheltering from the oil, and its effect on the Ekman current will diminish. For the DWH oil spill event, wind measurements at NDBC buoy 42040 show that the wind speed never exceeded 18 m/s. The yearly maximum wind speed was 18.3 m/s, which occurred during the winter time. Thus the effect of the oil on changes in the surface wind stress cannot be neglected.
In the oil spill modeling community, wind-induced oil drift is often considered as an extra component to the transport by the ocean currents (Reed et al. 1999;Abascal et al. 2009). The winds contribute positively to the oil drift estimate because the oil itself is subject to wind drag. This kind of practical estimate is not in conflict with the reduced effective wind speed, as discussed in the present study. Rather, these considerations highlight the different physics of wind work that are effective in oil spill drift. By improved understanding of the physics of oil drift in the ocean, appropriate response measures can be taken. Future studies to estimate oil drift trajectories will benefit from conducting detailed numerical simulations that incorporate the effective wind speed proposed in this paper, detailing the contributions from ocean currents as well as from the direct wind drag on the oil.

Conclusions
We have considered the effect of the presence of oil on the ocean surface and its possible modifications of ocean surface stress and associated impact on the dynamical response of the ocean circulation. By changing the surface roughness, oil reduces the wind kinematic input into the ocean interior. Such changes can be quantified, based on measurements from SAR. The concept of effective wind speed is proposed to evaluate the wind stress received by the ocean in oil-covered areas. It is found that the presence of oil on the ocean surface can significantly reduce the wind stress, and thus reduce the associated wind-driven circulation via the Ekman current. By analyzing data from the DWH oil spill event, it is found that the Ekman current can be reduced by 75%-100%, when oil is present on the ocean surface. It is important to consider such effects on oil trajectory predictions, especially in large oil spill events, to better support the measures taken for emergency response.