Riverine litter patch detection, quantification and tracking at the Grabovica dam using Sentinel-2 imagery before and...

Extreme weather events are leading to significant societal challenges, including loss of life and extensive damage to the natural environment. Flooding, often caused by high rainfall, can result in significant leakage of both natural and anthropogenic litter into the environment. We utilised Sentinel-2 high spatial resolution observations to detect, quantify and track the riverine litter garbage patch from the 03-04 October 2024 flooding event in Bosnia and Herzegovina. A dynamic hotspot of floating garbage formed at the intake side of Grabovica dam and initial detection was achieved through visual inspection of true colour Sentinel-2 images as well as in-situ photographs. Surface area of the floating garbage patch was estimated using a spectral band anomaly approach, Random Forest and XGBoost machine learning classification techniques. Robustness of the classification algorithms was evaluated through a leave-one-scene-out cross validation strategy. Within 5 days of the flooding event the floating garbage patch reached a surface area coverage of ∼77100 m². The computed average precision scores were relatively very high (>99%) and Leave-One-Scene-Out held-out accuracies were ≥ 95%, suggesting a consistent generalization performance at pixel-level in the ensemble models, with XGBoost showing the best seed-to-seed stability. Quantification at sub-pixel level was achieved using the band anomaly algorithm. Hindcasting for salient floating matter occurrences near the Grabovica dam wall was explored using time series ERA5 reanalysis data and the cloud-free Sentinel-2 images from May to November 2024. Temporal insights about the weather conditions were found to share trends with the satellite observable floating garbage around the Grabovica dam. Transferability of the methods was assessed and demonstrated by inference mapping of floating garbage patches from the same 03-04 October 2024 flooding event on Jablaničko Lake. The study findings showcase the potential observational benefits of integrating Sentinel-2 into monitoring strategies for supporting rapid response action plans after extreme flooding events. Utilizing relevant environmental descriptors derived from satellite observations, policymakers can make informed timely decisions especially in directing clean-up resources, initiating mitigation measures and future planning.