SATELLITE MONITORING OF OIL SPILLS IN THE KAZAKH SECTOR OF THE CASPIAN SEA: RESULTS AND INTERPRETATION

Abstract

A comprehensive methodology for space-based monitoring of oil pollution in the Kazakh sector of the Caspian Sea has been developed. The approach integrates automated processing of Sentinel-1 SAR imagery, deep learning techniques, and hydrodynamic modeling. The primary objective is the reliable detection of oil slicks, their differentiation from natural look-alike phenomena, and forecasting their dispersion across the marine area. Using this methodology, 2,326 satellite scenes acquired between August 2023 and August 2025 were processed. Semantic segmentation was performed using the Mask2Former transformer architecture trained on a specialized SAR dataset of the Caspian Sea. To forecast pollution dynamics, the MIKE 21 Oil Spill model was applied, incorporating GFS meteorological data and bathymetry derived from GEBCO. A total of 52 oil slicks with a combined area of approximately 269 km² were identified, predominantly in regions of intensive shipping and port infrastructure. A detailed analysis of a major slick detected on 16 August 2025, based on multi-temporal Sentinel-1A and Sentinel-2 data as well as hydrodynamic modeling, demonstrated that its characteristics correspond to the discharge of oily (bilge) water rather than an accidental crude oil spill. The developed methodology enables a transition from pollution detection to scientifically grounded forecasting of slick behavior. Future developments include the integration of Automatic Identification System data to identify violating vessels and to establish an operational environmental monitoring system.