Remote sensing technology is an essential link in the global monitoring of the ocean surface, and radars are efficient sensors for detecting marine pollution. When used operationally by authorities, a tradeoff must usually be made between the covered area and the quantity of information collected by the radar. To identify the most appropriate imaging mode, a methodology based on receiver operating characteristic curve analysis has been applied to an original data set collected by two airborne systems operating at L-band, both characterized by a very low instrument noise floor. The data set was acquired during controlled releases of mineral and vegetable oil at sea. Various polarization-dependent quantities are investigated, and their ability to detect slick-covered areas is assessed. A relative ordering of the main polarimetric parameters is reported in this paper. When the sensor has a sufficiently low noise floor, HV is recommended because it provides the strongest slick-sea contrast. Otherwise, VV is found to be the most relevant parameter for detecting slicks on the ocean surface. Among all the investigated quad-polarimetric settings, no significant added value compared to single-polarized data was found. More specifically, it is demonstrated, by increasing the instrument noise level, that the studied polarimetric quantities which combine the four polarimetric channels have performances of detection mainly driven by the instrument noise floor, namely, the noise equivalent sigma zero. This result, obtained by progressively adding noise to the raw synthetic aperture radar (SAR) data, indicates that the polarimetric discrimination between clean sea and polluted area results mainly from the differentiated behavior between single-bounce scattering and noise. It is thus demonstrated, using SAR data collected with a low instrument noise floor, that there is no deviation from Bragg scattering for radar scattering from ocean surface covered by mineral and vegetable oil.

Ship detection in the maritime domain is best performed with radar due to its ability to surveil wide areas and operate in almost any weather condition or time of day. Many common detection schemes require an accurate model of the amplitude distribution of radar echoes backscattered by the ocean surface. This paper presents a review of select amplitude distributions from the literature and their ability to represent data from several different radar systems operating from 1 GHz to 10 GHz. These include the K distribution, arguably the most popular model from the literature as well as the Pareto, K+Rayleigh, and the trimodal discrete (3MD) distributions. The models are evaluated with radar data collected from a ground-based bistatic radar system and two experimental airborne radars. These data sets cover a wide range of frequencies (L-, S-, and X-band), and different collection geometries and sea conditions. To guide the selection of the most appropriate model, two goodness of fit metrics are used, the Bhattacharyya distance which measures the overall distribution error and the threshold error which quantifies mismatch in the distribution tail. Together, they allow a quantitative evaluation of each distribution to accurately model radar sea clutter for the purpose of radar ship detection.

Land monitoring using temporal series of Synthetic Aperture Radar (SAR) images requires radiometrically well calibrated sensors. In this paper, the radiometric stability of the new SAR Sentinel-1A “S-1A” sensor was first assessed by analyzing temporal variations of the backscattering coefficient (σ°) returned from invariant targets. Second, the radiometric level of invariant targets was compared from S-1A and Radarsat-2 “RS-2” data. The results show three stable sub-time series of S-1A data. The first (between 1 October 2014 and 19 March 2015) and third (between 25 November 2015 and 1 February 2016) sub-time series have almost the same mean σ°-values (a difference lower than 0.3 dB). The mean σ°-value of the second sub-time series (between 19 March 2015 and 25 November 2015) is higher than that of the first and the third sub-time series by roughly 0.9 dB. Moreover, our results show that the stability of each sub-time series is better than 0.48 dB. In addition, the results show that S-1A images of the first and third sub-time series appear to be well calibrated in comparison to RS-2 data, with a difference between S-1A and RS-2 lower than 0.3 dB. However, the S-1A images of the second sub-time series have σ°-values that are higher than those from RS-2 by roughly 1 dB.

Remote sensing techniques are commonly used by Oil and Gas companies to monitor hydrocarbon on the ocean surface. The interest lies not only in exploration but also in the monitoring of the maritime environment. Occurrence of natural seeps on the sea surface is a key indicator of the presence of mature source rock in the subsurface. These natural seeps, as well as the oil slicks, are commonly detected using radar sensors but the addition of optical imagery can deliver extra information such as thickness and composition of the detected oil, which is critical for both exploration purposes and efficient cleanup operations. Today, state-of-the-art approaches combine multiple data collected by optical and radar sensors embedded on-board different airborne and spaceborne platforms, to ensure wide spatial coverage and high frequency revisit time. Multi-wavelength imaging system may create a breakthrough in remote sensing applications, but it requires adapted processing techniques that need to be developed. To explore performances offered by multi-wavelength radar and optical sensors for oil slick monitoring, remote sensing data have been collected by SETHI (Système Expérimental de Télédection Hyperfréquence Imageur), the airborne system developed by ONERA (the French Aerospace Lab), during an oil spill cleanup exercise carried out in 2015 in the North Sea, Europe. The uniqueness of this dataset lies in its high spatial resolution, low noise level and quasi-simultaneous acquisitions of different part of the EM spectrum. Specific processing techniques have been developed to extract meaningful information associated with oil-covered sea surface. Analysis of this unique and rich dataset demonstrates that remote sensing imagery, collected in both optical and microwave domains, allows estimating slick surface properties such as the age of the emulsion released at sea, the spatial abundance of oil and the relative concentration of hydrocarbons remaining on the sea surface.

SAR sensors are usually used in the offshore domain to detect marine oil slicks which allows the authorities to guide cleanup operations or prosecute polluters. As radar imagery can be used any time of day or year and in almost any weather conditions, the use and programming of such remote sensing data is usually favored over optical imagery. Nevertheless, images collected in the optical domain provide access to key information not accessible today by SAR instruments, such as the thickness or the amount of pollutant. To address this knowledge gap, a methodology based on the joint use of a scattering model (U-WCA) and remote sensing data collected by a low frequency (e.g., L-band) imaging radar over controlled release of mineral oil spill is reported in this paper. The proposed method allows estimation of the concentration of pollutant within an oil-in-water mixture as well as the temporal variation of this quantity due to weathering processes.

In this paper, we proposed to investigate unsupervised anomaly detection in Synthetic Aperture Radar (SAR) images. Our approach considers anomalies as abnormal patterns that deviate from their surroundings without prior knowledge of their characteristics. This method deals with the crucial problems related to the presence of speckle, the spatial correlation structures in SAR images, and the lack of annotated data to train a detection algorithm. Our proposed method aims to address these issues through a self-supervised learning algorithm. First, we propose to mitigate the SAR speckle through the deep learning SAR2SAR algorithm. We then develop an Adversarial Autoencoder (AAE) to reconstruct anomaly-free SAR images from despeckled data taking into account potential spatial correlation structures. Finally, a change detection processing step is applied between the input and the output to detect anomalies. Experiments are performed to show the advantages of our method compared to the conventional Reed-Xiaoli algorithm, highlighting the importance of an efficient despeckling pre-processing step.