The Impact of System Noise in Polarimetric SAR Imagery on Oil Spill Observations

Abstract

The effects of both system additive and multiplicative noise on X-, C-, and L-band Synthetic Aperture Radar (SAR) data covering oil slicks are examined. Prior studies have attempted to characterize such oil slicks, primarily through analysis of polarimetric SAR data. Here we factor in system noise that is added to the backscattered signal, introducing artifacts that can easily be confused with random and volume scattering. This confusion occurs when additive and/or multiplicative system noise dominates the measured backscattered signal. Polarimetric features used in this study are shown to be affected by both additive and multiplicative system noise, some more than others. This study highlights the importance of considering specifically multiplicative noise in the estimation of the signal-to-noise ratio (SNR). The SNR based on additive noise should at least be above 10dB. The SNR involving both additive and multiplicative noise should at least be above 0dB. The SNR from TerraSAR-X and Radarsat-2 is below 0dB for the majority of the oil slick pixels when considering both the additive and multiplicative noise, rendering these data unsuitable for any analysis of the scattering properties and characterization. These results are in contrast to the reduced impact of noise on oil slicks detected by the Lband UAVSAR system. In particular, we find there is no need to invoke exotic scattering mechanisms to explain characteristics of the data. We also recommend a noise subtraction for any polarimetric scattering analysis.

Synthetic Aperture Radar Remote Sensing of Operational Platform Produced Water Releases

“…For low backscatter regions such as oil spills, the sensor's background noise can be a challenge, resulting in a poor signal-to-noise ratio (SNR). This may affect the backscatter values and multi-polarization parameters, possibly reducing the characterization ability and cause misinterpretations [5,12,13,21].…”

“…Note that the NESZ is subtracted before taking the ratio, i.e., the actual signal above NESZ, not the measured signal is used (see [13]). Figure 2 shows the range of SNR values within the slick regions.…”

“…In [13], a limit of SNR = 10 dB is suggested in order for the data to be useful for polarimetric scattering analysis. Figure 2 shows that the median SNR values for the clean sea lie between 11-18 dB, and all scenes but one have their 5th percentile above 10 dB.…”

“…Still, no polarimetry-based methods have been thoroughly tested for large data sets and found to show consistently good and reliable results, hence being ready for operational use. In addition, sensor noise and its effect on polarimetric measurements and interpretation is a challenge, particularly for spaceborne sensors (see, e.g., [12,13]).…”

“…In PD, the nonpolarized component is removed, and this parameter is, therefore, controlled by the small-scale wave components close to the Bragg wavenumber and should reveal near-surface wind variability and presence of slicks [19]. The PD has been found to give a very good oil-sea contrast for C-and L-band data in, e.g., [8,[19][20][21][22], and to be less sensitive to incidence angle and proximity to noise floor than other multipolarization features [13,21].…”