Abstract

Abstract

A serious obstacle to the total uptake of open Earth Observation data (Copernicus Sentinel images, NASA ’s Landsat and similar) in daily lives is the steep data analysis curve required to get from raw images to Analysis-Ready, Decision-Ready/Relevant, not to mention Forensics ready data. The combined complexity of high data volumes, atmospheric disturbances (clouds, haze) and inconsistent coverage and diverse and complex signal physics (e.g. radar images vs optical images; sudden changes in land use) has resulted in the number of EO data applications remaining rather marginal. For example, in Europe, it is estimated that only a small fraction of farmers and forest managers use Sentinel images for decision-making. The recently generated Google DeepMind AlphaEarth (10 m global for 2017–2025) and Tessera embeddings being complete, consistent and ARD , provide an opportunity to decrease the steep data processing curve and enable thousands of applications. In our work, we have also consistently focused on making EO data more ARD and more usable, primarily by aggregating Landsat 1997–2025 values to bi-monthly (Consoli et al., 2024). In the current approach (Landsat ARD global mosaics V2 monthly) developed a 4–step process to derive improved quality mosaics: (1) first, we aggregate monthly reflectances across the whole time-frame (cca 30 years) and use these normalized values to detect outliers, (2) we then derive monthly median values with filtered reflectances (already significantly reduces clouds, snow and noise), (3) we then gap-fill values using convolutional filter and consistent land cover classes, and (4) we finally gap-fill all remaining values using modeling. For these steps we use a data fusion approach with annual ensemble land cover data at 30 m, together with MODIS EVI monthly (complete, consistent) and geometric temperature (a function of latitude and day of the year) as covariate layers to help improve gap-filling. Although using embeddings seems to also solve the issues of clouds, snow and noise, the advantage of having monthly mosaics is that they are easier to interpret and trace back potential errors and artifacts. The resulting monthly global cloud free mosaics are then consistent, gap-free, should contain minimum artifacts and can be used directly for modeling and a diversity of land monitoring applications (from above-ground biomass, vegetation height, yield and soil property mapping). We will present some initial results and discuss how we could combine forces to make open EO data reach more people, enable more applications and save more lives.

Bio

Tom has more than 25 years of experience as an environmental modeler, data scientist and spatial analyst. Tom has a background in soil mapping and geo-information science (PhD at Wageningen University / ITC ). He continuously runs hands-on-R training courses to promote use of Open Source software for spatial analysis / spatial modeling purposes.

He is currently the project leader of the Open-Earth-Monitor project (https://doi.org/10.3030/101059548) and Director at the OpenGeoHub foundation. Tom is recipient of the Clarivate Highly Cited Researchers for 2021, 2022, 2023, 2024 and 2025. Several of his paper have received the best paper awards including the “Finding the right pixel size” (https://doi.org/10.1016/j.cageo.2005.11.008), “Soil property and class maps of the conterminous USA ” (https://doi.org/10.2136/sssaj2017.04.0122), his articles published in PeerJ are among top 10 most cited of all time; his PLOS One paper (https://doi.org/10.1371/journal.pone.0169748) is listed among the most cited in the field.