This data product is currently under evaluation and review. It may contain inaccuracies or be subject to change. Users should exercise caution and discretion when interpreting or relying on this information. The government assumes no liability for any errors or decisions based on this preliminary data. For more details, please see the Government of Canada's Open Commons license (https://open.canada.ca/en/open-government-licence-canada). The Canadian Soil Information Service has developed a detailed dataset of Canada's soils and associated properties using advanced machine learning techniques. The Soil Landscape Grids of Canada is produced using a combination of historical and current data from both soil sampling and remote sensing. The machine learning model is trained using over 10,000 pedon locations from across Canada as well as 70 covariate datasets. This new dataset is pivotal in addressing the gaps left by legacy soil surveys and facilitates more comprehensive assessments nationwide. As new data becomes available and machine learning techniques advance, this information can be updated much faster than with traditional soil surveying methods.

This series of historic flood susceptibility maps comes from an XGBboost machine learning model trained on major floods from 2005 to 2023. The trained model is then run for each year from 2000 to 2023, including unique temporal characteristics of temperature, precipitation, land use land cover and Normalized Difference Vegetation Index (NDVI), to predict the flood susceptibility of any given year.Parent Collection:- **[Collection - Flood Susceptibility Mapping- Cartographic Product Collection](https://open.canada.ca/data/en/dataset/1074f781-85d3-4c86-86cb-fd1c339197dc)**

This series of projected future flood susceptibility maps were generated using an XGBoost machine learning model trained on major floods from 2005 to 2023. The trained model was applied to future climate scenarios for 2050, 2070, and 2100, under two SSP scenarios: 245 and 585. The model uses temperature and precipitation time series to estimate potential future flood susceptibility. These maps represent model projections and should be interpreted as indicators of potential flood susceptibility, not precise forecasts.

OneSoil employs a proprietary machine learning (ML) model based on state-of-the-art instance segmentation to detect field boundaries. Utilizing raw Sentinel-2 data aggregated according to local vegetation season maps and an additional upscaling module to enhance boundary accuracy, we ensure precise results. OneSoil’s data preprocessing involves the utilization of their cloud detector module and local season mapping.

This national map of flood susceptibility or flood prone areas is based on patterns of historic flood events as predicted by an ensemble machine learning model.The recommended use is national, provincial or regional scale and can be used as a guide for identifying areas for further investigation. The Flood Susceptibility Index (FSI) Dataset, while processed and available at 30m cell size, is not recommended for use at the pixel or street level, given the uncertainty in the modelling process and the variability of results as discussed in https://www.mdpi.com/2673-4931/25/1/18 .For additional details on the methods, tests, models and datasets used to generate this data layer, please see https://geoscan.nrcan.gc.ca/starweb/geoscan/servlet.starweb?path=geoscan/fulle.web&search1=R=329493

This series of historic flood susceptibility maps comes from an XGBboost machine learning model trained on major floods from 2005 to 2023. The trained model is then run for each year from 2000 to 2023, including unique temporal characteristics of temperature, precipitation, Land Use Land Cover (LULC) and Normalized Difference Vegetation Index (NDVI), to predict the flood susceptibility of any given year. **This third party metadata element follows the Spatio Temporal Asset Catalog (STAC) specification.**

This series of historic flood susceptibility maps comes from an XGBboost machine learning model trained on major floods from 2005 to 2023. The trained model is then run for each year from 2000 to 2023, including unique temporal characteristics of temperature, precipitation, Land Use Land Cover (LULC) and Normalized Difference Vegetation Index (NDVI), to predict the flood susceptibility of any given year. **This third party metadata element follows the Spatio Temporal Asset Catalog (STAC) specification.**

This series of historic flood susceptibility maps comes from an XGBboost machine learning model trained on major floods from 2005 to 2023. The trained model is then run for each year from 2000 to 2023, including unique temporal characteristics of temperature, precipitation, Land Use Land Cover (LULC) and Normalized Difference Vegetation Index (NDVI), to predict the flood susceptibility of any given year. **This third party metadata element follows the Spatio Temporal Asset Catalog (STAC) specification.**

This series of historic flood susceptibility maps comes from an XGBboost machine learning model trained on major floods from 2005 to 2023. The trained model is then run for each year from 2000 to 2023, including unique temporal characteristics of temperature, precipitation, Land Use Land Cover (LULC) and Normalized Difference Vegetation Index (NDVI), to predict the flood susceptibility of any given year. **This third party metadata element follows the Spatio Temporal Asset Catalog (STAC) specification.**

This dataset presents the first comprehensive, high-resolution (10-meter) wetland inventory map covering the entire 32.2 million square kilometers of the Pan-Arctic region, of which 14 million square kilometers (43%) is terrestrial and 18.4 million square kilometers (57%) is marine. Generated through advanced Earth Observation and machine learning techniques, the map was produced using multi-year (2020–2022), multi-source satellite imagery—including Sentinel‑1, Sentinel‑2, and ALOS PALSAR‑2—as well as various environmental features such as elevation. Over 1,000 wetland polygons were analyzed using an object-based random forest classification workflow on the Google Earth Engine cloud platform, achieving an average overall classification accuracy of 89%.The mapping extent was defined according to the Arctic Council’s Conservation of Arctic Flora and Fauna (CAFF) boundary, resulting in the identification of 2,947,618 km² of wetlands, representing 20% of the land area within the Pan-Arctic region. This dataset establishes a consistent and authoritative baseline for pan-Arctic wetlands, leveraging the latest advances in Earth Observation, machine learning, and cloud computing. The Canadian Wetland Classification System was used and includes the major wetland classes: bog, fen, marsh, swamp, and water.The overall wetlands coverage by country within the CAFF boundary was: Canada (27%), United States of America (i.e., Alaska 39%), Finland (31%), Iceland (8%), Norway (17%), Sweden (26%), Kingdom of Denmark (i.e., Greenland 1%), and the Russian Federation (21%).​Development of this product was undertaken by Natural Resource Canada's Canada Centre for Mapping and Earth Observation and the Canadian Geospatial Data Infrastructure Division in collaboration with the Arctic Council’s CAFF biodiversity group, CAFF Wetland Experts Group, national organisations mandated to monitor wetlands, and Arctic National Mapping Agencies, and Canadian company C-CORE, integrating ground truth data collected from Alaska, Finland, Sweden, and the Kingdom of Denmark through partner agencies and digital image interpretation. More than 60,000 images (2020-2022), primarily covering summer periods, were processed to ensure robust results.This dataset provides essential baseline information for Earth Observation monitoring of climate change impacts and supports critical environmental surveillance for Arctic and remote northern communities.