ALGORITHMS FOR THE ESTIMATION AND CHARACTERIZATION OF BEACHING IN OIL SPILL

Federico Santini

Institute of Methodologies for Environmental Analysis, National Research Council of Italy

(CNR-IMAA), Potenza (Italy)

Research theme aims

In the literature, the oil spill detection algorithms use active space segment (radar sensors) for their identification at the sea, also providing indications on the thickness of the pollutant film (PRIMI project). To our knowledge, little has been done to support the emergency phase connected to the beaching of oil patch. The originality of this research consists in dealing with this gap with innovative techniques based on the elaboration of hyperspectral optical data. The validity of the approach is guaranteed by the success of similar applications in the field of hyperspectral remote sensing [1].

Materials and Methods

Since the interest is focused on the identification of beached hydrocarbons, the sites have been chosen for those cases where it was possible to collect in situ and remote data in the period between the beaching and the extinction (natural o consequent to cleaning operations) of the hydrocarbons. The most dramatic case of oil spill is the environmental disaster of the Deepwater Horizon oil plant, affiliated with British Petroleum. An accident involving the Macondo Well, which is more than 1500 meters deep, caused a massive oil spill in the waters of the Gulf of Mexico. The spill started on 20/04/2010 and ended 106 days later, on 04/08/2010, with over 4.4 million barrels of oil spilled on the waters in front of Louisiana, Mississippi, Alabama and Florida [2]. For this analysis, the attention focuses on the area related to the Bay of Barataria. In this area, in fact, several in situ campaigns were carried out for the collection of spectra and ground truths that were used as a validation for some research work [3,4]. In detail, we are interested in verifying the potential of remote sensing in identifying and quantifying beached oil extension. The research is therefore oriented on the investigation of spectral techniques based on the main hydrocarbon absorption features. For this purpose, combinations of spectral indices such as Hydrocarbon Index (HI) by Kühn et al. (2004) [5] have been tested:

                                 (Eq. 1)

where Rx and lx are respectively the radiance and the wavelength of the channel x.

The AVIRIS image acquired on 14/09/2010 was downloaded from the AVIRIS portal (https://aviris.jpl.nasa.gov). The image was then atmospherically corrected and orthorectified. The corrections were carried out using different atmospheric models to optimize the data. Basing on the literature data at disposal, a subset area of the image was selected, to carry out a series of tests to: a) asses the best algorithm for the identification of beached oil; b) identify the best spectral interval for the implementation of the algorithms; c) verify the use of radiance data; d) compare the algorithm performances with literature results e) test the algorithm performances after resampling the AVIRIS data on the characteristics of other sensors (e.g. PRISMA). The tests were carried out defining, on the bases of preliminary results and literature data, ROIs related to beached oil and oil-free areas.

Results and Discussion

Best algorithm and oil spectral feature characterization

In addition to the HI index (eq.1), the Band-depth (Bd) and a three-channel spectral ratio index (3cSR) used by NASA since 2006 were tested:

                              (Eq. 2)

                                                                    (Eq. 3)

For each algorithm, a series of map were produced varying the position of the spectral channel within a wide range in correspondence to the main absorption feature centered at 1.72 and 2.3 μm. To evaluate the capability of the maps in discriminating the oil-impacted areas, the statistical distributions of the values of the ROIs were computed and a separability index (SI) was defined as:

       (Eq. 4)

where goil-impacted(x) and goil-free(x) are the gaussian fits of the value distributions related to the oil-impacted and oil-free ROIs, “x” is the value of the applied index (HI, Bd, 3cSR). Test results suggest that all the algorithms can detect the presence of oil with best results obtained with HI for which a value of SI=.99 is reached against SI=.94 of the Bd and SI= .85 of the 3cSR. Always basing on the SI index, it was possible to characterize the oil spectral features. From the analysis we deduce that the best band choice is [Ra=140=1681,98 nm; Rb=145=1731,79 nm; Rc=148=1761,67 nm] for the feature at 1.72 μm and [Ra=199=2247,58 nm; Rb=205=2307,39 nm; Rc=215=2406,89 nm] for the feature at 2.3 μm.

Radiance vs. spectral data

No significant results were found applying the algorithms to radiance or reflectance data. The maps are almost indistinguishable as well as the statistical distributions, with a very modest improvement when atmospheric correction are applied (the SI associated to HI1.72 passes from 0.9967 to 0.9984). As a result, the algorithms were applied to radiances to save time. Once evaluated the algorithms and characterized the oil spectral features, other images were downloaded and analyzed to find data with a different impact of beached oil. Other three cloud-free images were selected: 04/10/2010, 04/05/2011 and 15/10/2011. Figure 1 shows how the oil has decreased over time and the HI maps clearly show the decrease of the oil-impacted areas, while the statistical distributions of the oil-impacted and oil-free ROIs tend to overlap with an SI that from 0.99 goes to 0.01.

Figure 1: HI1.72 maps (1st row) and oil maps (2nd row) related to the AVIRIS images acquired on 14/09/2010 (1st column), 04/10/2010 (2nd column), 04/05/2011 (3rd column), and 15/10/2011 (4th column). The last row shows the related oil-impacted (dashed line) and oil-free (dotted line) HI1.72 value statistical distributions with superimposed the gaussian fit (solid lines).

For validation purposes the HI1.72 was applied to a wider region including other areas with documented presence of beached oil. Results were in excellent agreement with what reported on Kokaly et al., 2013 [3].

Simil-PRISMA data

To estimate the potential of the HI index on PRISMA data, a simil-PRISMA image was simulated resampling the AVIRIS data to the spatial resolution characteristic of the PRISMA sensor. Once applied the HI1.72 and reconciled the ROIs to the simil-PRISMA image, the statistical HI1.72 value distributions were computed for the oil-impacted and oil-free areas obtaining the graph of figure 2. Comparing this graph with the one related to the original AVIRIS data, we figure out that the overlap between the distribution is much increased with an SI that is 0.54 for the simil-PRISMA image. Anyhow, selecting the pixels with HI1.72 > 2, we still obtain a good map of the oil-impacted pixels, were many of the areas detected with AVIRIS data are still visible (Fig. 2).

Figure 2: From the left, HI1.72 value distributions of the oil-impacted and oil free ROIs reconciled on the simil-PRISMA image; RGB of the simil-PRISMA image; beached oil map.

Future perspectives

Results show that HI is a powerful algorithm for beached oil detection when applied to both the 1.72 μm and the 2.3 μm spectral features, with a slight advantage for the first one. The results also show that the algorithm can be applied directly to the radiance image. Comparison with literature results and application to temporal series with different beached-oil impact allowed to verify the potential of the approach. Several areas interested by beached oil are still detectable after the resampling to the PRISMA spatial resolution. Accordingly, in the next future, real PRISMA data could confirm the potential of the HI applications for beached-oil detection.

References

  1. Bianchi, R., Cavalli, R.M., & Marino, C.M. (1996). Evaluation of the spatial distribution, in percent coverage of the oil spilled during the Trecate blow-out, based on the analysis of airborne hyperspectral MIVIS data (No. CONF-960203-). Environmental Research Institute of Michigan (ERIM), Ann Arbor, MI (United States).
  2. Crone, T.J., & Tolstoy, M. (2010). Magnitude of the 2010 Gulf of Mexico oil leak. Science, 330(6004), 634-634.
  3. Kokaly, R.F., Couvillion, B.R., Holloway, J.M., Roberts, D.A., Ustin, S.L., Peterson, S.H., Khanna, S., Piazza, S.C. (2013). Spectroscopic remote sensing of the distribution and persistence of oil from the Deepwater Horizon spill in Barataria Bay marshes. Remote Sensing of Environment, 129, 210-230.
  4. Arslan M.D., (2013). Oil spill detection and mapping along the Gulf of Mexico coastline based on imaging spectrometer data. Master of Science Dissertation, Office of Graduate and Professional Studies of Texas A&M University.
  5. Kühn, F., Oppermann, K., Hörig, B. (2004). Hydrocarbon Index–an algorithm for hyperspectral detection of hydrocarbons. International Journal of Remote Sensing, 25(12), 2467-2473.