We have developed a UAS system that collects multispectral data in order to characterize oil slick thicknesses and emulsification ratios. This system consists on a UAS that carries multiple cameras that integrate 10 wavelength band sensors ranging from Ultra-Violet (UV) to Long Wave Infrared (LW-IR). This system has been originally tested at OHMSETT and at the MC-20 site in the Gulf of Mexico. More recently this UAS was put in operation during the Lake Washington Wellhead blowout in Louisiana. In here we present examples of how this operational tool allowed oil spill responders to efficiently deploy containments of the floating oil (booming) and to monitor the collection of the oil on real time. Moreover, using a rapid classification algorithm, the multispectral data collected by our UAS allowed us to make a detailed high resolution classification of the oil detected on the shorelines of the affected areas. The UAS also delivered near real time oil detections that were used during the spill by the NOAA oil spill science coordinators through the ERMA system. This UAS has proven its ability to detect oil on ‘hard to reach areas’ and it offers a valuable option for the evaluation of affected areas impacted by the spill. We compared the SCAT surveys with the UAS oil detections and conclude the importance of adding this UAS tool as part of the operational assessment of the spill to determine the level of impact of the spill on the nearshore environment.
Although detection of oil in the form of a slick on the ocean surface based on a variety of remote sensing techniques is well understood under most conditions, estimation of the thickness of oil layers is technically challenging because it requires the verification via in-situ measurements at the spill site [Garcia-Pineda et al., 2009–2012–2015–2017]. Because it is not always practical, safe, or affordable to conduct extensive in situ measurements in the aftermath of a spill, assessments rely heavily on remote sensing data analysis and more recently on the usage of Unmanned Aerial Systems (UAS).
Oil spill response efficiency is currently limited by time, information, manpower, and equipment on hand (Leifer et al., 2012). One of the most important initial steps in response to an oil spill at sea is the detection of the extents of the oil slick and more importantly the identification of actionable oil (i.e. thickness) distribution within it. This is of great importance for efficient direction of response efforts. When released at sea, many types of hydrocarbons rapidly spread out to very thin layers of floating oil, and this is why accurate determination of which areas contain higher volumes of oil is vital for efficiently guiding oil spill response efforts either for contention, extraction, dispersion, or burning operations.
The critical component needed to enable adaptive responder fleet direction is a technology capable of rapid detection and characterization of the oil slick properties with near real-time generation of oil slick characterization maps that can be downlinked directly by the responding vessels available via the internet to all interested participants. The UAS system we use is designed to be deployable from one of the responding vessels, which will also be capable of broadcasting its real-time video to multiple vessels in the responding fleet. Therefore, all responding participants can benefit from the UAS system.
The controlled experiment was conducted at the National Oil Spill Response Research & Renewable Energy Test Facility (Ohmsett) in June 2018. Ohmsett features an above-ground concrete test tank of 203 m long by 20 m wide by 3.4 m deep. The tank was filled with 2.6 million gallons of clear saltwater.
Different from previous Ohmsett experiments, a black tarp (Figure 1a, black rectangular target of 20'x30'under the squares) was set up at the bottom of the Ohmsett tank to mimic water reflectance in the natural marine environment. Because the Ohmsett tank is shallow (2.4 meters deep) and the water is clear, sun light in the visible wavelengths can easily penetrate to the bottom of the tank and then get reflected back through the water column, captured by the hand-held SR-1900 spectroradiometer and multi-band cameras. The high reflectance of the tank bottom thus leads to very high reflectance, representing an unrealistic background as compared with the real marine environments. Using the SR-1900 spectroradiometer, we measured a reflectance of ~25% of the tank water (Figure 1b) in the green wavelengths, which is more than 10 times higher than typical reflectance of coastal waters (i.e., Tampa Bay water, ~2%) and open-ocean water in the Gulf of Mexico (<1%) in the corresponding wavelengths. By introducing a black tarp featuring < ~6% reflectance in the visible wavelengths, the water background becomes much darker, with a surface reflectance much closer to reflectance in typical coastal and open-ocean environments (Figure 1b). All oil square measurements when then conducted on top of the tarp location.
Squares of 1m x 1m containing crude oil, emulsified oil, Sargassum, and clean water were set on top of the tarp. Sargassum samples were collected from a recent Gulf of Mexico field trip and shipped to the Ohmsett facility. Squares containing different volumes of HOOPS crude oil and emulsified oil were aimed to create different thicknesses of each oil type if oil were to spread evenly in the squares. Unfortunately, due to oil heterogeneity even distributions of surface oil within the squares were impossible even without wind. When wind was blown from one direction, oil may usually pile up in the square, creating a horizontal gradient. This makes it only possible to treat the squared oil as a whole with an equivalent mean thickness. Both clean Sargassum and oiled Sargassum (oiled by either crude or emulsified oil) were set in the squares. There is one square containing clean water to serve as a reference. Figure 1a shows a sample image of the experimental set up.
Imaging sensors response and oil thickness classifications
Figure 2 shows the thermal response and multiband response of the multi-band cameras to different volumes of HOOPS crude oil and Sargassum in the squares. The oil in each square is not distributed uniformly but shows significant patchiness. Therefore, even if a mean thickness can be calculated from the known volume and surface area, the real thickness in the oil patch can be very different, making it difficult to associate the SR-1900 measured thickness with the calculated mean thickness. This is because it is unknown which exact oil patch the hand-held SR-1900 spectroradiometer was viewing.
As shown in Figure 2, oil emulsion and crude oil show different reflectance in the red, NIR and SWIR bands—oil emulsions have high reflectance in these bands while crude oil has very low reflectance (almost zero) in these bands. These observations can be used to develop algorithms to classify oil emulsions from crude oil. For example, of the square setup on June 14, 2018 (Figure 3), crude oil shows dark color in both the true color (R: 668; G: 560; B: 475 nm) and false color (R: 840; G: 668; B: 560 nm) images. Oil emulsions display reddish color in the true color image but brownish to yellowish color in the false color image. Sargassum appears as dark brownish in the true color image while but reddish color in the false color image. The same color appearance of the three different target types can also be found in the square setups in Figures 2,3,4 and 5, respectively. Clearly, with a combination of true color and the false color images, crude oil, oil emulsion, and Sargassum can be differentiated and classified.
Based on the emissivity contrast, the classification of the oil thickness can be achieved as shown on figure 5.
3. Results from Field Test of the Multispectral UAS: Lake Washington Oil Spill
We tested the multispectral array of sensors during the Lake Washington Oil Spill, which started on December 8th of 2018. This spill was reported by the U.S. Coast Guard as an equipment failure on the wellhead casing in Lake Washington/Rattlesnake Bayou. This event presented an excellent opportunity for testing our multispectral UAS on a real setting scenario, as we were aware of possible oil slicks with different thicknesses.
We used the services from a local charter vessel company. We chartered low draft boat. This gave us access to the shallow areas where the spill was occurring. We arrived at the scene of the spill on December 10th, 2018. We were able to quickly identify areas where the thick floating oil could be present and proceeded to prepare our equipment for deployment. The first steps consisted on testing and calibration of the equipment. Tests consisted on aircraft flight checks, camera calibration, and real-time video system check. Figure 6 shows the UAS with all the multispectral sensors collecting a snapshot of the calibration panel.
During this spill event, a massive amount of oil was observed floating on the surface of the area, which could be deleterious to the environment.
Oblique aerial view delivered on Near Real Time to NOAA-ERMA
It is important to point out that during the UAS flights we were able to extract oblique views that were posted on www.erma.noaa.gov in near real time about 15 minutes after the images were taken. This example below (Figure 7) is an oblique image that shows the boat (upper right corner) by the shoreline of an island that has been reached by the oil. This image was posted on The Environmental Response Management Application (ERMA) site on near real time.
The ERMA site is a GIS platform that integrates all types of data collections on the field. This platform shares the information with all the emergency responders, and for the first time we were able to use this platform to share data with the responders in near real time (within 15 minutes of being collected on the field). The aerial oblique photography includes the GPS location of the UAS, which provides a perspective of the location of the oil and features on the ground.
ERMA website was used during the response (figure 8). And it is also a recipient of all the UAS data products that are generated from this work.
Each of the UAS multispectral surveys collected could contain several hundreds of individual stills that were stored immediately after the flight on a hard drive. These stills along with the flight information is then used to generate the orthomosaics of each UAS survey. The orthomosaics are generated by several algorithm routines that integrate the individual stills with the location, altitude, direction of flight, and angle of the cameras, all this information is needed to compile all the images into an orthomosaic.
UAS Multispectral Data Used for Oil Classification
The UAS used on this project was also equipped with a down looking sensor that measures the incoming light from the sky. This measurement is crucial for the calibration of the reflectances on each of the surveys. The incoming light is recorded on each stills and it is used during the postprocessing calibration steps. The ability to relate: 1) the digital value of the pixels of the single channels, 2) the incoming light, and 3) the calibration panel, allows us to convert pixel values to reflectance values (Clark et al 2010). Despite the many attempts in the past to determine reflectance response to oil-on-water (e.g., Otremba and Piskozub, 2001 & 2003; Otremba et al., 2013), in practice it has been very difficult to develop inversion algorithms to classify oil type (e.g., crude oil versus oil emulsion) and quantify thickness from the oil's reflectance spectra. Many experiments have been attempted for such inversions, mostly under controlled laboratory environments, yet the results often differ for many reasons (Svejkovsky et al 2006–2012, Wettle et al 2009, Sun et al 2016). In addition to the heterogeneity induced mixed pixel effect, when applied to remote sensing imagery, the results from laboratory-based measurements are always confounded variable solar/viewing conditions, winds, and different water properties from those in the laboratory experiments. This problem is overcome by adding a normalization of the incoming light sensor. Single channels from the multispectral sensor are:
1) UltraViolet (450 nm) with a Bandwith of 20nm
2) Green (560nm) with a Bandwith of 20nm
3) Red (668nm) with a Bandwith of 10nm
4) Red Edge (717nm) with a Bandwith of 10nm
5) Near IR (840nm) with a Bandwith of 40nm
6) LW-IR (7um) Infrared Thermal
By using narrow bands, this multispectral array of sensors are able to better detect the spectral reflectance properties of the objects on the surface, specifically the properties that are most important for vegetation indices and for the reflection of the oil properties. The narrow bands allow higher sensitivity than wide bands by capturing the most relevant segments of the spectral curve. An example of the orthomosaics generated from the independent channels is shown on Figure 9.
Besides the single channel orthomosaics, a Normalized Difference Vegetation Index (NDVI) and Ultra-High Visual (RGB) resolution maps are generated (Figure 9). The incoming light sensor allow us to calibrate the imagery regardless of the sky conditions. During the collection days, sky conditions were cloudy on the 15th, then sunny on the 16th, and finally cloudy again on the 17th of December.
Once the visual inspection and identification of the oil is performed, we bring into the analysis the layer that contains the UAS-UV channel. On this layer the focus is on those areas previously identified as possible oil on the enhanced RGB. The purpose is to identify the pixels on the UV layer that would correspond to thicknesses from rainbow sheen and above observed on the visual. We then use an ‘Oil Thickness Classification Tool’ (OTCT which is a tool developed by Water Mapping used in arcmap). This tools allows us to create training sets (regions of pixels with the aspect and characteristics we hand pick) based on features of interest. We then use OTCT to run a classification over the entire orthomosaic.
NDVI (Normalized Difference Vegetation Index) Map
NDVI contrasts the red and near-infrared bands of light reflected from plant leaves. It is a general indicator of canopy density and is frequently used to distinguish live green vegetation from soil. This products can be use to evaluate before/after spill effects.
NDVI always ranges from −1 to +1. But there isn't a distinct boundary for each type of land cover. For example, when we have negative values, it's highly likely that it's water. On the other hand, if you have a NDVI value close to +1, there's a high possibility that it's dense green leaves. But when NDVI is close to zero, there isn't green leaves and it could even be an urbanized area. Normalized Difference Vegetation Index (NDVI) uses the NIR and red channels in its formula.
The final product is when the SCAT surveys are integrated on the UAS analysis. Figure 10 shows an orthomosaic with the detection of the oil produced as a visual map (RGB) after the reflectances has been calibrated (using the multispectral sensor). The top panel on Figure 9 shows a clear distinction of oil reaching the shoreline of the marsh island. This map also shows the SCAT lines with the survey lines of regions that were mapped and other regions that were not mapped. Most of the areas inside the marshes are hard to reach even by airboats, being this the reason why SCAT was not able to reach some areas impacted by oil that can be seen by the UAS. Figure 9 bottom shows the classification of the oil as thin (rainbow sheens) and thick (metallic and above).
Discussion and Conclusion: Multispectral UAS Data Synthesis.
UAS imagery collected during the Lake Washington spill revealed oil in multiple areas on the open water and on the shoreline of several small islands with marshes. The method we followed to conduct the identificaiton of the oil, and then the classification of the oiled shorelines was based mainly on the analysis of the UAS-Visual (RGB) and the UAS-UV channel from the multispectral sensor. This is a supervised classification technique that requires understanding of the visual aspect of floating oil.
By looking at the raw UAS visual products it is possible to discern the presence of floating oil depicted by the rainbow sheen appearance on the water. If instead of looking at the raw RGB we look at an enhanced version of the imagery (image stretching), the oil oustands even more and it is easier to identify the floating rainbow sheens. The image stretching varies and its applied depending on the color properties of the image. Different stretching settings could provide multiple results and we apply different settings to see what configuration produce the best enhancement of the features. Figure 9 shows a version of the calibrated RGB and an enhancement version of the same area where the oil is more clearly discernable. The visual inspection of the enhanced imagery is the first step in which we take the orthomosaic and we zoom in-out to the maximum displayable resolution and then we pan the imagery left-right, up-down so we identify those areas that could have possible features related to the oil spill. The purpose of this procedure is to visually identify features that could be associated with different classes of oil thickness.
Floating oil can vary its thicknesses due to the hetereogenity properties of the oil. Oil thicknesses can range from the thinnest silver sheen, rainbow sheen, metallic, transitional dark, true dark (or true color) and thick emulsions. In this case the purpose is to identify those areas that could be reached by a significant amount of oil from rainbow sheens and above. The ultimate product from the multispectral UAS is the analysis of the vegetation on areas where oil has been detected. As shown on Figure 10. This product can be used for injury assesment. The panel on the left shows an ultra high resolution map of the marshes. The panel on the middle shows a index of vegetation density (color coded) that represents the degree of prescence of healthy vegetation on an area near the shoreline impacted by the oil (red and yellow colors). A time series analysis of a map like this would provide an understanding of the injury assessment over time.
This study was made possible in part by funding received from the U.S. Bureau of Safety and Environmental Enforcement (BSEE) and the National Oceanic and Atmospheric Administration (NOAA). Reference in this paper to any specific commercial product, process, or service, or the use of any trade, firm or corporation name does not constitute endorsement, recommendation, or favoring by the BSEE, NASA, NOAA or USGS. The scientific results and conclusions, as well as any views or opinions expressed herein, are those of the authors and do not necessarily reflect the views of NOAA or the Department of Commerce.