(2)These products are usually collected at a coarse resolution, which makes analysis at a finer scale difficult 12. Even though, users still have some issues regarding appropriate product selection due to the following factors: (1)In most cases, users are unable to find a product that fits their desired LULC class or geographic region of interest 25, 26. Over the last few years, several attempts have been made to overcome these inconsistencies with a harmonised approach capable of providing better control in the validation and comparison over the growing number of existing LULC products 23, 24. (6)Different validation techniques and different ground truth reference data were used in each product, which impedes a reliable accuracy comparison. (5)Different classification techniques, field-data collection approaches, and subjective interpretations were used to create each product. (4)Different classification systems (i.e., LULC legends) were adopted in each product, usually each one focusing on a distinct application. (3)Each product has a different updating frequency (from regularly to never updated products). (2)Different pre-processing techniques, like atmospheric corrections, cloud removal and image composition were used in each product. These reports explain that this disagreement is due to several methodological reasons, including: (1)Given that different satellite sensors with different spatial resolutions were used in each product, the difference in precision from coarse to fine resolution imagery partially determines the final quality of each product. However, despite the acceptable global accuracy of each individual product, a considerable disagreement among products has been reported 4, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22.

Multiple LULC products have been derived from satellite information at the global scale (Table 2), contributing to a better monitoring, understanding, and territorial planning of our planet 5, 6. High resolution LULC mapping plays a key role in many fields, from natural resources monitoring, to biodiversity conservation, urban planning, agricultural management or climate and earth system modelling 2, 3, 4. Land-Use and Land-Cover (LULC) mapping aims to characterize the continuous biophysical properties of the Earth surface as categorical classes of natural or human origin, such as forests, shrublands, grasslands, marshlands, croplands, urban areas or water bodies, etc. Sentinel2GlobalLULC is designed for training deep learning models aiming to build precise and robust global or regional LULC maps. Metadata includes a unique LULC annotation per image, together with level of consensus, reverse geo-referencing, global human modification index, and number of dates used in the composite. Each image is a 224 × 224 pixels tile at 10 × 10 m resolution built as a cloud-free composite from Sentinel-2 images acquired between June 2015 and October 2020. Sentinel2GlobalLULC v2.1 contains 194877 single-class RGB image tiles organized into 29 LULC classes. Here, we introduce Sentinel2GlobalLULC, a Sentinel-2 RGB image dataset, built from the spatial-temporal consensus of up to 15 global LULC maps available in Google Earth Engine. However, there still exists low consistency among LULC products due to low accuracy in some regions and LULC types. Global LULC products are continuously developing as remote sensing data and methods grow. Land-Use and Land-Cover (LULC) mapping is relevant for many applications, from Earth system and climate modelling to territorial and urban planning.

Spacenet classifier decision