A seamless spatiotemporal machine learning framework for automated prediction, uncertainty assessment, and analysis of land use / land cover (LULC) dynamics is presented. The framework includes: (1) harmonization and preprocessing of high-resolution spatial and spatiotemporal covariate datasets (GLAD Landsat, NPP/VIIRS) including 5 million harmonized LUCAS and CORINE Land Cover-derived training samples, (2) model building based on spatial k-fold cross-validation and hyper-parameter optimization, (3) prediction of the most probable class, class probabilities and uncertainty per pixel, (4) LULC change analysis on time-series of produced maps. The spatiotemporal ensemble model was fitted by combining random forest, gradient boosted trees, and artificial neural network, with logistic regressor as meta-learner. The results show that the most important covariates for mapping LULC in Europe are: seasonal aggregates of Landsat green and near-infrared bands, multiple Landsat-derived spectral indices, and elevation. Spatial cross-validation of the model indicates consistent performance across multiple years with 62%, 70%, and 87% accuracy when predicting 33 (level-3), 14 (level-2), and 5 classes (level-1); with artificial surface classes such as 'airports' and 'railroads' showing the lowest match with validation points. The spatiotemporal model outperforms spatial models on known-year classification by 2.7% and unknown-year classification by 3.5%. Results of the accuracy assessment using 48,365 independent test samples shows 87% match with the validation points. Results of time-series analysis (time-series of LULC probabilities and NDVI images) suggest gradual deforestation trends in large parts of Sweden, the Alps, and Scotland. An advantage of using spatiotemporal ML is that the fitted model can be used to predict LULC in years that were not included in its training dataset, allowing generalization to past and future periods, e.g. to predict land cover for years prior to 2000 and beyond 2020. The generated land cover time-series data stack (ODSE-LULC), including the training points, is publicly available via the Open Data Science (ODS)-Europe Viewer.Global climate change and increasing urbanization worldwide intensify the need for a better understanding of human heat stress dynamics in urban systems. During heat waves, which are expected to increase in number and intensity, the development of urban cool islands could be a lifesaver for many elderly and vulnerable people. The use of remote sensing data offers the unique possibility to study these dynamics with spatially distributed large datasets during all seasons of the year and including day and night-time analysis. For the city of Basel 32 high-quality Landsat 8 (L8) scenes are available since 2013, enabling comprehensive statistical analysis. Therefore, land surface temperature (LST) is calculated using L8 thermal infrared (TIR) imagery (stray light corrected) applying improved emissivity and atmospheric corrections. The data are combined with a land use/land cover (LULC) map and evaluated using administrative residential units. The observed dependence of LST on LULC is analyzed using a thermal unmixing approach based on a multiple linear regression (MLR) model, which allows for quantifying the gradual influence of different LULC types on the LST precisely. Seasonal variations due to different solar irradiance and vegetation cover indicate a higher dependence of LST on the LULC during the warmer summer months and an increasing influence of the topography and albedo during the colder seasons. Furthermore, the MLR analysis allows creating predicted LST images, which can be used to fill data gaps like in SLC-off Landsat 7 ETM+ data.
", "keywords": ["multiple linear regression", "Landsat 8", "land use/land cover", "Science", "atmospheric corrections", "Q", "0211 other engineering and technologies", "land surface temperature", "02 engineering and technology", "15. Life on land", "01 natural sciences", "LST analysis", "13. Climate action", "11. Sustainability", "land surface temperature; thermal infrared data; LST analysis; atmospheric corrections; land use/land cover; multiple linear regression; urban; Landsat 8", "thermal infrared data", "urban", "0105 earth and related environmental sciences"]}, "links": [{"href": "http://www.mdpi.com/2072-4292/9/7/684/pdf"}, {"href": "https://www.mdpi.com/2072-4292/9/7/684/pdf"}, {"href": "https://doi.org/10.3390/rs9070684"}, {"rel": "related", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items/Remote%20Sensing", "name": "related record", "description": "related record", "type": "application/json"}, {"rel": "self", "type": "application/geo+json", "title": "10.3390/rs9070684", "name": "item", "description": "10.3390/rs9070684", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items/10.3390/rs9070684"}, {"rel": "collection", "type": "application/json", "title": "Collection", "name": "collection", "description": "Collection", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main"}], "time": {"date": "2017-07-04T00:00:00Z"}}, {"id": "10.5281/zenodo.7641616", "type": "Feature", "geometry": null, "properties": {"updated": "2026-09-20T16:20:58Z", "type": "Dataset", "title": "Soil organic carbon change data under human land-use changes", "description": "The dataset includes site observational data of soil organic carbon (SOC) changes that were used for the meta analysis of estimating the impacts of human-induced land use/land cover (LULC) change on SOC. The LULC types include agriculture (conventional tillage, animal plowing, shifting cultivation, and no-till), pasture, grazing (light, moderate, and heavy intensity), and secondary vegetation cover. Data include the sampling location, depth range of the sample, soil texture, clay content (if reported), the observed SOC change (in percentage, compared to a reference undisturbed site), and the data source.", "keywords": ["soil organic carbon", "land use/land cover change", "Soil degradation"], "contacts": [{"organization": "Pei-Ling Wang, Johannes Feddema,", "roles": ["creator"]}]}, "links": [{"href": "https://doi.org/10.5281/zenodo.7641616"}, {"rel": "self", "type": "application/geo+json", "title": "10.5281/zenodo.7641616", "name": "item", "description": "10.5281/zenodo.7641616", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items/10.5281/zenodo.7641616"}, {"rel": "collection", "type": "application/json", "title": "Collection", "name": "collection", "description": "Collection", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main"}], "time": {"date": "2025-09-30T00:00:00Z"}}, {"id": "3164960866", "type": "Feature", "geometry": null, "properties": {"license": "Open Access", "updated": "2026-09-20T16:22:50Z", "type": "Journal Article", "created": "2021-05-27", "title": "A spatiotemporal ensemble machine learning framework for generating land use / land cover time-series maps for Europe (2000 \u2013 2019) based on LUCAS, CORINE and GLAD Landsat", "description": "AbstractA seamless spatiotemporal machine learning framework for automated prediction, uncertainty assessment, and analysis of land use / land cover (LULC) dynamics is presented. The framework includes: (1) harmonization and preprocessing of high-resolution spatial and spatiotemporal covariate datasets (GLAD Landsat, NPP/VIIRS) including 5 million harmonized LUCAS and CORINE Land Cover-derived training samples, (2) model building based on spatial k-fold cross-validation and hyper-parameter optimization, (3) prediction of the most probable class, class probabilities and uncertainty per pixel, (4) LULC change analysis on time-series of produced maps. The spatiotemporal ensemble model was fitted by combining random forest, gradient boosted trees, and artificial neural network, with logistic regressor as meta-learner. The results show that the most important covariates for mapping LULC in Europe are: seasonal aggregates of Landsat green and near-infrared bands, multiple Landsat-derived spectral indices, and elevation. Spatial cross-validation of the model indicates consistent performance across multiple years with 62%, 70%, and 87% accuracy when predicting 33 (level-3), 14 (level-2), and 5 classes (level-1); with artificial surface classes such as 'airports' and 'railroads' showing the lowest match with validation points. The spatiotemporal model outperforms spatial models on known-year classification by 2.7% and unknown-year classification by 3.5%. Results of the accuracy assessment using 48,365 independent test samples shows 87% match with the validation points. Results of time-series analysis (time-series of LULC probabilities and NDVI images) suggest gradual deforestation trends in large parts of Sweden, the Alps, and Scotland. An advantage of using spatiotemporal ML is that the fitted model can be used to predict LULC in years that were not included in its training dataset, allowing generalization to past and future periods, e.g. to predict land cover for years prior to 2000 and beyond 2020. The generated land cover time-series data stack (ODSE-LULC), including the training points, is publicly available via the Open Data Science (ODS)-Europe Viewer.Abstract
A seamless spatiotemporal machine learning framework for automated prediction, uncertainty assessment, and analysis of land use / land cover (LULC) dynamics is presented. The framework includes: (1) harmonization and preprocessing of high-resolution spatial and spatiotemporal covariate datasets (GLAD Landsat, NPP/VIIRS) including 5 million harmonized LUCAS and CORINE Land Cover-derived training samples, (2) model building based on spatial k-fold cross-validation and hyper-parameter optimization, (3) prediction of the most probable class, class probabilities and uncertainty per pixel, (4) LULC change analysis on time-series of produced maps. The spatiotemporal ensemble model was fitted by combining random forest, gradient boosted trees, and artificial neural network, with logistic regressor as meta-learner. The results show that the most important covariates for mapping LULC in Europe are: seasonal aggregates of Landsat green and near-infrared bands, multiple Landsat-derived spectral indices, and elevation. Spatial cross-validation of the model indicates consistent performance across multiple years with 62%, 70%, and 87% accuracy when predicting 33 (level-3), 14 (level-2), and 5 classes (level-1); with artificial surface classes such as 'airports' and 'railroads' showing the lowest match with validation points. The spatiotemporal model outperforms spatial models on known-year classification by 2.7% and unknown-year classification by 3.5%. Results of the accuracy assessment using 48,365 independent test samples shows 87% match with the validation points. Results of time-series analysis (time-series of LULC probabilities and NDVI images) suggest gradual deforestation trends in large parts of Sweden, the Alps, and Scotland. An advantage of using spatiotemporal ML is that the fitted model can be used to predict LULC in years that were not included in its training dataset, allowing generalization to past and future periods, e.g. to predict land cover for years prior to 2000 and beyond 2020. The generated land cover time-series data stack (ODSE-LULC), including the training points, is publicly available via the Open Data Science (ODS)-Europe Viewer.
", "keywords": ["Time Factors", "Spatiotemporal", "QH301-705.5", "Data Mining and Machine Learning", "Urbanization", "Uncertainty", "Spatial analysis", "R", "Environmental monitoring", "04 agricultural and veterinary sciences", "15. Life on land", "01 natural sciences", "Europe", "Big data", "Machine learning", "Medicine", "0401 agriculture", " forestry", " and fisheries", "Biology (General)", "Landsat", "Ensemble", "Land use/land cover", "Environmental Monitoring", "Probability", "0105 earth and related environmental sciences"]}, "links": [{"href": "https://doi.org/PMC9308969"}, {"rel": "related", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items/PeerJ", "name": "related record", "description": "related record", "type": "application/json"}, {"rel": "self", "type": "application/geo+json", "title": "PMC9308969", "name": "item", "description": "PMC9308969", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items/PMC9308969"}, {"rel": "collection", "type": "application/json", "title": "Collection", "name": "collection", "description": "Collection", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main"}], "time": {"date": "2021-05-27T00:00:00Z"}}], "links": [{"rel": "self", "type": "application/geo+json", "title": "This document as GeoJSON", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items?keywords=land+use%2Fland+cover&f=json", "hreflang": "en-US"}, {"rel": "alternate", "type": "text/html", "title": "This document as HTML", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items?keywords=land+use%2Fland+cover&f=html", "hreflang": "en-US"}, {"rel": "collection", "type": "application/json", "title": "Collection URL", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main", "hreflang": "en-US"}, {"type": "application/geo+json", "rel": "first", "title": "items (first)", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items?keywords=land+use%2Fland+cover&", "hreflang": "en-US"}, {"rel": "last", "type": "application/geo+json", "title": "items (last)", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items?keywords=land+use%2Fland+cover&offset=5", "hreflang": "en-US"}], "numberMatched": 5, "numberReturned": 5, "distributedFeatures": [], "timeStamp": "2026-09-20T20:04:42.968543Z"}