near real time oil spill detection and monitoring using ...

NEAR REAL TIME OIL SPILL DETECTION AND MONITORING USING SATELLITE OPTICAL DATA Grimaldi C.S.L.1, Coviello I. 2, Lacava T.2, Pergola N. 2,1 and Tramutoli V. 1,2 [1] Department of Engineering and Physics of the Environment (DIFA), University of Basilicata, Italy Via dell’Ateneo Lucano, 10 – 85100 Potenza (Italy). [2] Institute of Methodologies for Environmental Analysis (IMAA), National Research Council, Italy C.da S. Loja – 85050 Tito Scalo (Italy). ABSTRACT Timely detection and continuously updated information are fundamental in reducing the ecological impact of the different sources of sea pollution. Satellite remote sensing, especially from meteorological platforms having a high temporal resolution and an easy data delivery, can be profitably used for a near real time sea monitoring. Recently, a new methodology for oil spill detection and monitoring, based on the general Robust Satellite Technique (RST) approach, has been proposed. This technique has shown, by using AVHRR Thermal Infrared (TIR) data, a good capability in automatically detect, with high level of reliability, oil spill presence. In this paper, such an approach has been exported for the first time to MODIS TIR data. Preliminary results obtained for an oil spill event occurred during Lebanon war in 2006, are shown and discussed. Index Terms — Oil spill, remote sensing, RST, AVHRR, MODIS. 1. INTRODUCTION Oil spill disasters seriously threaten marine ecosystem: reducing such kind of technological hazard is essential for protecting the environment as well as for cut down economic losses [13]. Satellite remote sensing might effectively contribute to mitigate oil spill environmental impact, provided that reliable and effective detection techniques will be developed, and relevant information and products may be timely delivered and shared. The most preferred instruments in satellite oil spill surveillance are SAR (Synthetic Aperture Radar) sensors [2] [7] [14] for their all-weather and all-day capabilities [8] [9], as well as their high spatial resolution. Unfortunately, the actual operational use of SAR sensors in timely detection of oil spill at the global scale, is currently limited by their revisiting time (from several days to several weeks depending on the latitude) and their expensive costs;

978-1-4244-3395-7/09/$25.00 ©2009 IEEE

moreover, the reliability in oil spill detection is also limited by the influence of wind speed in acquiring signal and the presence of natural films or rain cells that give an oil spill similar signal (look-alikes). The present SAR revisiting time limitation will be overcame when COSMO-Skymed mission will be fully deployed (expected for the 2010), and a SAR constellation of four satellites will guarantee a refresh time until 12 hours [1] but several open questions regarding costs and delivery policy of such data, might limit their use in an operational context. For the above mentioned reasons, passive optical sensors, on board meteorological satellites, may represent, at this moment, a suitable SAR alternative and, for the next future, an useful complement for oil spill detection and monitoring. They, in fact, offer the best temporal resolution nowadays available from space (from several hours to few minutes, depending on the characteristics of the platform/sensor). This circumstance is very important to mitigate an hazard as the oil spill one for which time factor is crucial to possibly mitigate environmental damages. Reliable optical satellite techniques for an automatic oil spill detection in Infrared and Visible spectral regions have been for long-time mostly missing. They can localize the presence of an oil spill only after an alert and require the presence of an experienced operator. In particular, the techniques in the Mid (MIR) and Thermal (TIR) Infrared spectral regions [6] [20] exploit oil and water different thermal inertia to detect oil spill sea pollution. Oil thermal inertia, in fact, is lower than sea water one, and so oil polluted areas usually show higher brightness temperature in TIR images collected in daytime than sea water, the opposite during the night [8] [9]. Spurious effects, mainly due to local and environmental conditions (presence of clouds, local warming effects, cold/warm sea currents, etc.), might anyway reduce the sensitivity of identification. In latest times, an innovative technique, based on the general RST - Robust Satellite Techniques, [17] [19]approach, originally named RAT - Robust AVHRR Technique – [16], has been proposed [3] [4] [5] for an

IV - 709

IGARSS 2009

automatic, reliable and timely detection of oil spilled areas and for their near real time continuous monitoring using AVHRR (Advanced Very High Resolution Radiometer, aboard NOAA satellites) TIR data (channel 4 and 5, centred at 10.8μm and 11.4μm, respectively). In spite of the AVHRR coarse spatial resolution (1.1 km at nadir), this technique has demonstrated good performances both in terms of sensitivity (to the presence even of very thin/old oil films) and reliability (up to zero occurrence of false alarms). In this work, the same approach has been implemented also using data acquired by MODIS (Moderate Resolution Imaging Spectroradiometer) sensor aboard Terra and Aqua EOS (Earth Observing System) satellites. MODIS, like AVHRR, guarantees an adequate revisiting time (generally from 1 to 4 passes per day at mid-latitudes) for a frequent monitoring of oil spill risk affected areas. MODIS, in fact, acquires data also in the TIR region of the electromagnetic spectrum (1 km of spatial resolution), in particular, ch31 and ch32 are centred very close to AVHRR ch4 and 5, respectively (channel 31: 10.780 - 11.280 μm; channel 32: 11.770 - 12.270 μm). Main aims of such an attempt are: i) verify the full exportability of the RST (that is sensorindependent being based only on the availability of sufficiently long time series of satellite records) on MODIS data; ii) improve, by means of the full integration of AVHRR and MODIS records, the monitoring sample rate reducing, in this way, the limit possibly related to cloud cover. 2. METHODOLOGY The RST approach exploits the analysis of long-term multitemporal satellite records in order to obtain a former characterization of the measured signal, in term of expected value and natural variability, providing a further identification of signal anomalies by an automatic, unsupervised change detection step using specific threshold for time and place of observation. In its first application for oil spill monitoring [3] [4] [5], the signal under investigation was the one measured in AVHRR TIR channels in order to exploit the different thermal inertia between oil polluted areas and sea water at these wavelengths. Oil spill detection was obtained by using the RETIRA index (Robust Estimator of TIR Anomalies) firstly proposed by Tramutoli et al. [18]: '

⊗ ΔT x ( r , t ) =

[ΔTx (r , t ' ) − μ ΔTx ( r )]

σ ΔT ( r )

located over sea in the investigated area. μǻTx (r) and σǻTx (r) are, respectively, the average over time and standard deviation values of ǻTx (r,tǯ) computed on a homogeneous data-set of cloud-free satellite records collected at location r in the same observational conditions (i.e. same month of the year, same time of day) of the image under investigation. By this way, ⊗ΔTx (r, t’) gives the excess of the current signal ǻTx(r,tǯ) compared with its historical mean value and weighted by its historical variability at each considered location. It should be noted that the use of the differential variable ǻTx(r,tǯ) is expected to reduce the possible contributions (e.g. occasional warming) to the signal due to year-to-year climatologically changes and/or season timedrifts which usually affect near surface temperatures at a regional spatial scale [5]. The same index, obviously, can be implemented by using relevant MODIS TIR channels (ch31 or ch32, respectively). Remembering oil spill spectral signatures in TIR channels described above, we expect to find, in correspondence of polluted zones, high positive (negative) values of the index during the day (night). 3. THE LEBANON “JIYYEH POWER PLANT” OIL SPILL The test case proposed for the study is an oil spill event occurred in July 2006 during the Lebanon war between Lebanon and Israel. On July 13 and 15, 2006 the oil-fuelled power plant of Jiyyeh, located directly on the coastline, approximately 30 km south of Beirut, was hit by bombs. Part of the storage tanks caught fire and were burning for several days. Approximately 30,000 tons of heavy fuel oil was spilled into the Mediterranean Sea as a result of the blast [10]. Due to south-westerly winds and the sea currents, the oil spill was partly carried out to sea and partly dispersed along the coast of Lebanon from the Damour region south of Beirut to the Syrian border in the North [10] (Figure 1).

(1)

x

where ǻTx(r,tǯ)=Tx (r,tǯ)-Tx (tǯ) with x being the AVHRR TIR channel (ch4 or ch5, respectively) is the difference between the current (t = tǯ) signal value observed at location r, Tx (r,tǯ) and the spatial average Tx (tǯ), computed in place on the image at hand considering only cloud-free pixels

Figure 1. Oil spill localization (a) and extension (b) (Report UNEP, 4 August 2006 available at: http://www.indybay.org/newsitems/2006/08/05/18294908.php)

IV - 710

For the computation of AVHRR reference fields an historical data set of about 80 images, all acquired in previous years during the same month of July and in the same temporal range (10:30 – 13:30GMT), have been processed. The ROI (Region of interest) of 200x200 pixels was centred for this event at 34N, 34.5E. For the computation of MODIS reference fields the historical data set for the month of July in the temporal range 08:30 – 11:30 GMT was of about 160 images and the ROI was centred in the same position of AVHRR acquisitions. The difference between the temporal slots of the two datasets is related to scheduled passage of the satellites. We chose NOAA 14, 16 and 18 satellite acquisitions because they are the only daily passage nearest to Aqua and Terra acquisitions available online, affecting, in this way, the different dataset population (two daily MODIS passage respect to one AVHRR). In this paper, for sake of brevity, only the results for the first day in which the oil was detected will be shown: the 28 July 2006 at 10.30 GMT for MODIS and at 11.27 GMT for AVHRR. For the same reason result achieved by using only AVHRR channel 5 (11.4 -12.4 μm) and MODIS channel 32 (11.77 – 12.27 μm) will be shown because performances of RST approach applied to AVHRR channel 4 and MODIS channel 31 have already been discussed in previous papers [5] [12]. 4. RESULTS The results achieved applying the RST approach on the studied oil spill event are shown in Figure 2. Looking at this picture, it is possible to note that anomalous pixels have been identified in both MODIS (Figure 2.a) and AVHRR images (Figure 2.b) along the Lebanon coasts. Their position and distribution is in agreement with oil spill information reported by in situ observations (Figure 1) as well as with those relative to oil spill drift toward North [15] [21]. It is also possible to observe as the presence of oil spill is detected with an accuracy of 100% over the full scene (no false alarms) by RST at a good S/N levels (⊗ǻT > 3). It should be noted that a higher signal to noise ratio means, in the RST context, that observed signal exceeds 3 times (at least) the historically observed local variability. Moreover, it should be noted that the technique does not need of whatever human supervision, automatically generating the products immediately after the acquisition of data. Obtained results confirm the robustness and reliability of the RST, TIR-based, approach independently from the used sensor (AVHRR or MODIS). Results achieved for the two sensors seem in good agreement even if, in this case, the extent of the spilled areas appear underestimated by AVHRR, due to the differences in population of used historical dataset and in satellite view angles over the ROI. In particular, over the investigated region, satellite zenith angles are significantly higher for AVHRR (44deg) than for MODIS scene (27deg), this makes: i) ground resolution cell

larger for AVHRR (about the double of the MODIS one), so that the signal coming form the same oil spill is spreaded over a wider area in the case of AVHRR; ii) atmospheric extinction more effective for AVHRR than MODIS due to the different air masses crossed by the TIR signal. Besides, if using such an approach at higher threshold levels (i.e. ⊗ǻT > 3) allows a detection “for sure” of oil spill presence, it is confirmed [5] that zooming at lower threshold levels around the detected anomalies, it is possible to carry out a detailed mapping of oil spatial distribution also depending on different thickness and/or on different levels of emulsion in sea water. ⊗ ΔT > 2 ⊗ ΔT > 3

(a)

⊗ ΔT > 2 ⊗ ΔT > 3

(b)

Figure 2. Oil spill mapping detected by RETIRA index on the 28 July 2006 MODIS image at 10.30 GMT (a) and AVHRR image at 11.27 GMT (b). Land is masked in black.

5. CONCLUSION In this work, the recently proposed AVHRR TIR, RST– based, technique for oil spill detection and monitoring has been exported to MODIS TIR data. In particular, the Lebanon “Jiyyeh power plant” oil spill event, occurred in July 2006, have been used as test case. The first results obtained analyzing one image of such an event confirm what expected: for both the sensors RST TIR approach detects the presence of oil spills with an accuracy of 100% over the full scene (no false alarms) at good S/N levels (⊗ǻT > 3). This attest, once again, the fully exportability of the proposed technique. Preliminary achievements obtained in this work confirm the expected RST exportability as well as the potential of such an approach, in automatically detecting and mapping oil spill. In fact, the investigation of RETIRA index at lower relative intensities allow us to map oil spills structure and extent as well as on their emulsion level and thickness. Also in this mapping phase the two sensors are in good agreement. A possible step forward is the possibility to exploit the better spatial resolution offered by MODIS VIS data (250m) for a more detailed mapping of the previously detected TIR anomalies [11]. If these results will be further confirmed by analyzing more test cases, the full integration of MODIS and AVHRR sensors will allow us to have more frequent observations (up

IV - 711

to 3 hours starting from sensors having a maximum revisiting time of 9 and 6 hours, respectively) of oil spill risk affected areas, increasing the sample rate of free clouds acquisitions. This will improve the possibility to effectively monitor in near real time oil slick evolution. 6. REFERENCES [1] ASI (Italian Space Agency), “COSMO-SkyMed System Description & User Guide”, 49 pp, available on line at http://www.cosmo-skymed.it/it/index.htm, 2007. [2] C. Brekke and A.H.S Solberg, “Review on Oil spill detection by satellite remote sensing”, Remote Sensing of Environment, 95, pp. 1-13. 2005 [3] D. Casciello, T. Lacava, N. Pergola, and V. Tramutoli, “Robust Satellite Techniques (RST) for Oil Spill Detection and Monitoring”, Fourth International Workshop on the Analysis of Multitemporal Remote Sensing Images, MultiTemp 2007, July 1820, 2007 Leuven, Belgium, 2007a. [4] D. Casciello, C. S. L. Grimaldi, I. Coviello, T. Lacava, N. Pergola, and V. Tramutoli, “A Robust Satellite Techniques for oil spill detection and monitoring in the optical range” In Global Monitoring for Security and Stability (GMOSS), JRC Scientific and Technical Reports, Ed. G. Zeug & M. Pesaresi, EUR 23033 EN, pp. 294-305, 2007b. [5] D. Casciello, T. Lacava, N. Pergola, and V. Tramutoli, “Robust Satellite Techniques (RST) for oil spill detection and monitoring using AVHRR Thermal Infrared bands”, International Journal of Remote Sensing, 2009, in press. [6] A.M. Cross, “Monitoring marine oil pollution using AVHRR data: observation off coast of Kuwait and Saudi Arabia during January 1991”, International Journal of Remote Sensing, 13, pp. 781-788, 1992. [7] G. Ferraro, A. Bernardini, M. David, S. Meyer-Roux, O. Muellenhoff, M. Perkovic, D. Tarchi, K. Topouzelis, “Towards an operational use of space imagery for oil pollution monitoring in the Mediterranean basin: A demonstration in the Adriatic Sea”, Marine pollution Bulletin, 54, pp. 403-422, 2007. [8] M.F. Fingas, C.E. Brown, “Remote sensing of oil spills”, Sea Technology, 38, pp. 37-46. 1997. [9] M.F. Fingas, C.E. Brown, “A review of the status of advanced technologies for the detection of oil in and with ice”, Spill Science & Technology Bulletin, 6 (5), pp. 295-302. 2000. [10] Green Line, “Lebanon-Jiyye Oil Spill July 2006” available at http://www.oilspilllebanon.org/articles/leb_oil_spill_2006_fact_sh eet.pdf, 2006.

[12] C.L.S Grimaldi, I. Coviello, T. Lacava, N. Pergola, and V. Tramutoli, “RST-based oil spill detection and monitoring using optical data”, International Workshop on the Analysis of Multitemporal Remote Sensing Images, Multitemp 2009, July 2830 2009, Mystic, CT, USA (in press). [13] M, N. Jha, J. Levy and Y. Gao, “Advances in Remote Sensing for Oil Spill Disaster Management: State-of-the-Art Sensor Technology for Oil spill Surveillance”, Sensors, 8, pp. 236-255, 2008. [14] A. Kostianoy, K. Litovchenko, O. Lavrova, M. Mityagina, T. Bocharova, S. Lebedev, S. Stanichny, D. Soloviev, A. Sirota, O. Pichuzhkina, “Operational Satellite Monitorino of Oil Spill Pollution in the Southeastern Baltic Sea: 18 Months Experience”, Environmental research, engineering and management, No. 4 (38), pp. 70-77, 2006. [15] REMPEC (Regional Marine Pollution Emergency Response Center for the Mediterranean Sea), “SITREP 1 SPILL IN LEBANON”, Situation Report available at http://www.rempec.org/newsmore.asp?id=153&lang=en, 2006. [16] V. Tramutoli, “Robust AVHRR Techniques (RAT) for Environmental Monitoring: theory and applications” In Earth Surface Remote Sensing II, Giovanna Cecchi, Eugenio Zilioli, Editors, Proceedings of SPIE Vol. 3496, pp.101-113, 1998. [17] V. Tramutoli, “Robust Satellites Techniques (RST) for natural and environmental hazards monitoring and mitigation: ten years and applications” The 9th International Symposium on Physical Measurements and Signatures in Remote Sensing, Beijing (China), ISPRS, vol. XXXVI (7/W20), pp. 792-795, ISSN 1682-1750, 2005a. [18] V. Tramutoli, V. Cuomo, C. Filizzola, N. Pergola, C. Pietrapertosa, “Assessing the potential of thermal infrared satellite surveys for monitoring seismically active areas: The case of Kocaeli (I˙zmit) earthquake, August 17, 1999” Remote Sensing of Environment, 96: 409 – 426, 2005b. [19] V. Tramutoli, “Robust Satellite Techniques (RST) for Natural and Environmental Hazards Monitoring and Mitigation: Theory and Applications” In Fourth International Workshop on the Analysis of Multitemporal Remote Sensing Images, Louven, Belgium,18-20 July, 2007. [20] W.Y. Tseng, “Oil spill detection from NOAA-AVHRR imagery” International Journal of Remote Sensing, 16 (18), pp. 3481-3482, 1995. [21] UNEP (United Nations Environment Programme), “Environmental Update No. 01- Lebanon Crisis -27 July 2006” available at http://www.unep.org/lebanon/pdfs/EESUpdate427July.pdf, 2006.

[11] C.L.S Grimaldi, D. Casciello, I. Coviello, T. Lacava, N. Pergola, and V. Tramutoli, “A MODIS-based Robust Satellite Technique for near real time monitoring of oil spilled areas”, Proceeding of 2008 IEEE Gold Remote Sensing Conference, ESAESRIN Frascati (Roma), Italy, 22-23 May 2008 (in press).

IV - 712