Promoting the use of renewable energy and conserving biodiversity are conflicting issues that need addressing. While the development of offshore wind facilities/turbines is accelerating, many seabirds have been exposed to collisions with wind turbines. We must identify high collision areas and avoid the construction of wind turbines in these spaces to reduce these conflicts. One solution is to develop useful finer scale sensitivity maps. In this study, we created a fine-scale map of collision risk by spatial modelling using information from bird flights at sea and explored the relative importance of each geographic variable relevant to the risk. Between 2016 and 2019, we collected 3D-location data from 117 black-tailed gulls (Larus crassirostris) of three colonies in two areas and 21 slaty-backed gulls (L. schistisagus) of four colonies in one area of northern Hokkaido, Japan. The spatial models that explain the occurrence of M-zone flight, which is the flight within the heights of high collision risk (20–140 m height), were constructed at a 1 km mesh using a random forest algorithm, a machine-learning tool. The model satisfactory predicted the spatial distribution of M-zone flights using geographic variables and species (correlation coefficient: 0.57–0.94), although data had some degrees of variation between species, years, colonies, and areas. Our model can be applied to other regions, as long as we have general topological information and the locations of colonies and harbors. The distance to the breeding colony and the nearest harbors were important, and the collision risk was 6–7 times higher within 15 km from the colonies and 5 km from harbors. Black-tailed gulls used different sites for foraging and commuting between years, whereas slaty-backed gulls used relatively consistent sites. These variations between species and among years suggest that collecting bird data over multiple years is necessary and effective for creating a generally applicable sensitivity map.