Natural Sciences, Earth and Related Environmental Sciences, Multidisciplinary Geosciences, Naturvetenskap, Geovetenskap och relaterad miljövetenskap, Multidisciplinär geovetenskap, Computer and Information Sciences, Other Computer and Information Science, Data- och informationsvetenskap (Datateknik), Annan data- och informationsvetenskap, Physical Geography, Naturgeografi, Artificial Intelligence, Artificiell intelligens

Determining the land cover (LC) data requirements used as input to noise simulations is essential for planning sustainable urban densifications. This study examines how different LC datasets influence simulated environmental noise levels of road traffic using Nord2000 in an urban area of 1 km2 in southern Sweden. Four LC datasets were used. The first dataset was based on satellite data (spatial resolution 10 m) combined with various other datasets implementing an LC classification algorithm prioritizing vegetation. The second dataset was created by applying an LC majority priority rule over every cell of the first dataset. The third dataset was produced by applying a convolutional neural network over an orthophoto (0.08 m spatial resolution), while the fourth dataset was created by manually digitizing ground surfaces over the same orthophoto also utilizing data from the municipality’s basemap. The results show that LC data impact simulated noise levels, with priority rules in LC classification algorithms having a greater effect than spatial resolution. Statistically significant differences (up to 3 dB(A)) were found when comparing the simulated noise levels generated using the vegetation-prioritizing LC dataset compared to the simulated noise levels of the other LC datasets.