TimeSpec4LULC: a global multispectral time series database for training LULC mapping models with machine learning

Abstract. Land use and land cover (LULC) mapping are of paramount importance to monitor understand the structure dynamics Earth system. One most promising ways create accurate global LULC maps is by building good quality state-of-the-art machine learning models. Building such models requires large datasets annotated time series satellite images, which not available yet. This paper presents TimeSpec4LULC (https://doi.org/10.5281/zenodo.5913554; Khaldi et al., 2022), a smart open-source dataset multispectral for 29 classes ready train was built based on seven spectral bands MODIS sensors at 500 m resolution, from 2000 2021, using spatial–temporal agreement across 15 products in Google Engine (GEE). The 22-year monthly were created globally (1) applying different assessment filters Terra Aqua satellites; (2) aggregating their original 8 d temporal granularity into composites; (3) merging + data combined series; (4) extracting, pixel level, 6 076 531 size 262 along with set metadata: geographic coordinates, country departmental divisions, consistency products, availability, human modification index. A balanced subset also provided selecting 1000 evenly distributed samples each class that they representative entire globe. To assess annotation dataset, sample pixels, around world class, selected validated experts very high resolution images both Bing Maps imagery. smartly, pre-processed, targeted towards scientific users interested developing various models, including deep networks, perform mapping.