Comparing radar satellites. Use of Sentinel-1 leads to an increase in oil spill alerts in Norwegian waters

Sammanfattning: Radar’s ability to detect oil spills through clouds, fog, and darkness is a significant advantage. Oil spill alert services using radar satellites have been in use in Norway for more than 25 years. After introducing two radar satellites Sentinel-1A and 1B in 2015-2016, the number of oil spill alerts has increased by 250% in Norwegian waters. This increase does not track the slight decreasing trend in oil spills at sea detected by Norwegian remote sensing aircraft. It is also not consistent with the decreasing trends for both the number of oil spills and the yearly oil spill volume registered by international monitoring programs. One unconfirmed theory was that the new satellites could detect smaller-sized oil spills as they offered a higher resolution radar. Approximately 9400 satellite images from 2011-2018, with almost 3900 oil spill alerts, were analyzed to explain the increase in oil spill alerts. To estimate the additional mineral oil discharged into the sea after 2015, 25% of the oil spill alerts were analyzed by connecting oil spill alerts and sources, using historical ship tracks for Automatic Identification System (AIS) and oil and gas facility position data. Using In situ oil spill data from the offshore industy and having detailed information on the ships connected to the oil spills, substance discharged could be estimated. The result shows only a marginal increase in mineral oil (2.65%). Unknown substances not connected to any source, vegetable oils, and animal oils from ship activities mainly contribute to the increase. Hotspot analysis shows significant spatial clustering in oil spill alerts that overlap with offshore industry areas, fishing activities areas, and major shipping lanes. To test the theory that the higher resolution radar sensor detects more minor-sized oil spills, a nonparametric test comparing Sentinel-1A/B to Radarsat-2 on oil slick size. The results show statistically significant differences in performance between Sentinel-1A/B and Radarsat-2 on small-size oil spills. A visual analysis by comparing radar images of Sentinel-1A and Radasrsat-2 on confirmed mineral oil slick shows how differently the two satellites detect small size oil slicks. One challenge using radar for oil spill detections is “oil spill lookalikes” detections. The service provider uses likelihood settings for flagging an alert as mineral oil. The results show that confidence settings are inaccurate and therefore unsuitable for end-user operational applications without additional information. This thesis shows a need for further research regarding mineral oil and oil spill lookalikes. The thesis also concludes that service providers should further develop the use of ancillary source data to improve decision-making for end-users.