ECOCLIMAP-SG-MLv2

60 m resolution land use/land cover map over Europe. An updated version of the ECOCLIMAP-SG-ML [1], produced with a re-trained deep learning model. The new deep learning model was trained by taking coordinates of the map as an additional input, and the resulting map was produced by keeping the score limit to 0.3 and choosing the land cover labels with highest probability. ECOCLIMAP-SG-MLv2 achieves a 0.01 point improvement in overall F1-score for the machine learning generated pixels. At the class level, it shows notable gains(> 0.05 point F1 increase) for Boreal broadleaf deciduous, LCZ5: open midrise,and LCZ10: heavy industry. These improvements come with a trade-off: acomparable decrease (> 0.05 point F1) in accuracy for C4 crops and LCZ8:large low-rise. 1. Rieutord, T., Gleeson, E., & Bessardon, G. (2024). ECOCLIMAP-SG-ML: an ensemble land cover map for numerical weather prediction [Data set]. In Land (Vol. 13, Number 11, p. 1875). Zenodo. https://doi.org/10.5281/zenodo.14046040