Detecting migrant vessels in the Mediterranean Sea: Using Sentinel-2 images to aid humanitarian actions

Abstract

The process of migration by sea is often accompanied by a great deal of risk for the migrants. The need for reliable and on time information on migrant movements is essential, especially as the available information is often limited or inconsistent. The aim of this paper is to show how freely available Sentinel-2 optical images over large areas can support humanitarian actions with timely and accurate geospatial information by providing the exact location of vessels at sea at the time of satellite acquisition. With the proposed detection method, we would like to distance ourselves from border surveillance and fight against clandestine migration as the first associations when dealing with movements of people using satellite technology. Instead, we would like to provide a better understanding of the situation for relieve authorities. Using Sentinel-2 data we have developed an automatic vessel detection and classification procedure. By first removing land from the images with the use of Modified Normalized Difference Water Index (MNDWI) we obtained a sea mask. We then applied a “vessel index”, which eliminates most of the atmospherics influences that can significantly weaken the detection results from optical images, and binomial logistic regression on the sea mask to obtain segments of possible vessels. We calculated a group of geometrical and spectral attributes of detected segments and removed all non-vessel segments based on the value of segments area. Later we implemented classification using decision tree classifier. Finally, we performed an accuracy assessment of vessel classification. The results demonstrate that the methodology gives a reliable outcome in a timely and consistent manner but can overlook smaller vessels (the length less than 20 m). Freely available satellite technology can, therefore, offer an efficient and effective solution for frequent monitoring and tracking of vessels in real time across large areas. This approach would give effective results when complementary with other methods for search and rescue of migrants to help reduce the intolerable death toll of refugees while crossing the sea.