Uncovering the seasonal dynamics of terrestrial oil spills through multi-temporal and multi-frequency Synthetic Aperture radar (SAR) observations

• Sentinel-1A temporal profile provided insights on impacted and oil-free vegetation. • Multi-Seasonal and Multi-Sensor SAR Image Composites were used for ML classification. • Incorporation of L-band SAR into MSIC significantly improved discrimination accuracy. • Wet season ML models were generally better using wet season L- C- and X- band MSIC. The phenological characteristics of vegetation exposed to oil pollution can reveal how different vegetation types and species respond to the effects of hydrocarbons in crude oil. This can further inform the recovery status and remediation efforts on polluted sites. In this study, the potential of various SAR frequencies (including multitemporal C band Sentinel-1, X band Cosmo Skymed, X band TanDEM-X, and L band ALOS PALSAR 2) were explored to analyse the characteristics of vegetation affected by hydrocarbons over time. The SAR backscatter characteristics of both oil-polluted and oil-free vegetation were systematically examined across different seasons to deduce the primary effects of oil pollution. Additionally, machine learning random forest (RF) classification and support vector machines (SVM) were implemented on seasonal image composites to assess spatial extent. Results show that stress caused by oil pollution on vegetation is better distinguishable during the wet season in the VV channel than in the VH channel of the multitemporal Sentinel 1. This was supported by the machine learning classification, as overall accuracy (OA) and Kappa (K) were also highest with the wet season SAR image composites. A further incorporation of L- and X-Band multifrequency SAR across the two seasons showed that the wet season composites significantly improved the classification accuracy, with Cropland, Grassland and Tree Cover Area (TCA) recording an increase in OA and K, to 82.3 % and 0.64, 66.67 % and 0.33, and 74.7 % and 0.49, respectively. Findings presented in this study represent a pioneering exploration of the capabilities of multi-temporal and multi-sensor SAR imagery in discriminating oil-impacted from healthy vegetation. This holds significant promise in evaluating the progress of environmental remediation, the regeneration of vegetation, and recovery efforts.