scieee AI-readable full text Open interactive document viewer

CUP4SOIL - High-resolution product presentation and data access

Heiden, Uta; d'Angelo, Pablo; Poggio, Laura; Karlshoefer, Paul; van Egmond, Fenny; van der Woude, Thaisa

Publisher
Zenodo
DOI
10.5281/zenodo.17965592
Repository
zenodo.org
Language
en

Abstract

The CUP4SOIL project aims to develop and test a proof-of-concept for a future downstream service to support national and European agencies in reporting on soil health/quality. One of the objectives of the CUP4SOIL project is to generate European-wide data products and indicators that can be used to characterise soil health and quality. Furthermore, these products, including the input data and environmental covariates used, shall be made available for the user community to test and validate data products and to develop example showcases. This contribution provides an overview about input-products, covariates and soil products including image examples and methodologies. In CUP4SOIL, different data types and levels are used and developed to model soil parameters using the digital soil mapping approach (DSM). DSM uses a statistical model to integrate products derived from Copernicus satellites Sentinel-1 and 2, environmental covariates such as digital elevation models, ERA-5 products and soil ground truth information from the LUCAS survey and other available sources. Several data products are derived from the Sentinel-2 time series using the Soil Composite Mapping Processor (SCMaP). It comprises mean reflectance composites as well as specific soil reflectance composites (SRC) that contain undisturbed and bare soils. Additionally, statistical information about the spectral variability of the soils and the confidence interval (CI) of the SRC spectrum as measure of the quality of the SRC spectrum is given. The bare soil frequency is a measure for the visibility of the bare soil within the whole observed time period in percent. This in an indication for the intensity of use of agricultural fields as well as for the process of soil carbon sequestration. EO based soil products were generated for primary soil properties (SOC, pH, texture), derived soil properties and some basic soil health indicators. The soil observations were split in 10 equally sized folds for cross-validation. Random Forest models were obtained with the ranger package, with the option to build Quantile Random Forests (QRF) to obtain pixel-based uncertainty. In order to allow users to evaluate the usefulness of the products, the SCMaP input products, environmental covariates as well as the final soil parameters are provided at a dedicated webpage with functionalities for downloading and data access via web services. In the future, CUP4SOILS will develop showcases using these data products in order to refine the user requirements for future soil products within the Copernicus Land Monitoring Service.

Full text

CUP4SOIL High-resolution product presentation and data access Uta Heiden1, Pablo d'Angelo1, Laura Poggio2, Paul Karlshöfer1, Fenny van Egmond2, Thaïsa van der Woude2 1DLR 2ISRIC ESA SYMPOSIUM ON EARTH OBSERVATION FOR SOIL PROTECTION AND RESTORATION 07.02.2024 DLR.de • Chart 1 Introduction CUP4SOIL general objective Objectives •Prepare a potential Copernicus downstream service to support national and European agencies for reporting on soil health/quality. •Generate European-wide example data products characterising soil health/quality •Develop a user community that tests and validates data products for soil health/quality information •Ensure close cooperation with the ESA WorldSoils project activities and other related projects/initiatives such as the EJP SOIL projects and others etc. … •Intermediate SCMaP soil products •Current possibility of EO-based soil parameter •Deviations from user requirements •Data package to “play around” •Develop show cases Introduction CUP4SOIL and WorldSoils WorldSoils (ESA) CUP4SOIL (EU - FPCUP) Lead GMV DLR Main objective Development of a pre-operational system for SOC monitoring Prepare future soil products within the Copernicus Land Monitoring Service (CLMS) Soil parameter SOC content SOC content pH, bulk density, nitrogen, texture, coarse fragments , … (maybe more) Soil prediction model Spectral soil mapping (bare soil) Digital soil mapping (vegetated areas) - > Including SCMaP products Digital soil mapping (all areas) - > Including SCMaP products Spatial resolution 50 m (Europe) 100 m (Global ) 20 m (Europe) Spatial coverage Europe Europe + Sentinel -2 L2A input data Sen2Cor MAJA Methodology General overview SCMaP Compositing MEAN + STD Bare surface reflectance composite Mean + STD reflectance composite Index calc. + thresholding Bare surface detection Averaging Averaging + Statistics Cloud and haze filter Input data Sentinel-2 Reflectances + Cloud Masks European Threshold Database ESA WorldCover Bare surface mask Bare surface pixel count Bare surface frequency Valid pixel count SCMaP Output •All Sentinel–2 images in L2A format -> processed with MAJA from 2018 –2022 •Larger Europe including Ukraine •Spectral Index based (e.g. Diek et al. 2017, Rogge et al. 2018, Demattê et al., 2018) •Used index: PV+IR2 (Heiden et al. 2022, Möller, M. et al. 2022, Dvorakova, K., et al., 2023) •Regionalised thresholds (Karlshöfer et al., in preparation) •5-years composite products Methodology General overview SCMaP Compositing MEAN + STD Bare surface reflectance composite Mean + STD reflectance composite Index calc. + thresholding Bare surface detection Averaging Averaging + Statistics Cloud and haze filter Input data Sentinel-2 Reflectances + Cloud Masks European Threshold Database ESA WorldCover Bare surface mask Bare surface pixel count Bare surface frequency Valid pixel count SCMaP Output Digital Soil Mapping Methodology Digital Soil Mapping –some notes Digital Soil Mapping •Input data from LUCAS (and other sources in WoSIS if relevant) •Covariates: oData prepared by DLR oData available from Copernicus (DEM, land cover) oGeology/parent material (JRC) oSimple radar products from Sentinel1 •Model: quantile random forest (robust approach allowing pixel-based uncertainty assessment) •Outputs: oPrimary soil properties oUncertainty index oOther uncertainty measures (to be further developed) Intermediate products Cross-validation - SOC Used covariates at the X-Axes: Groups (x -axes) Description no_dlr All covariates excluding DLR/ SCMaP products no_dlr_src All covariates excluding DLR/ SCMaP covariate: Soil Reflectance Composite no_dlr_src_mos All covariates excluding DL/SCMaP covariates using the mosaic of MREF and SRC and SRC itself Intermediate products Cross-validation –pH (water) Groups (x -axes) Description no_dlr All covariates excluding DLR/ SCMaP products no_dlr_src All covariates excluding DLR/ SCMaP covariate: Soil Reflectance Composite no_dlr_src_mos All covariates excluding DL/SCMaP covariates using the mosaic of MREF and SRC and SRC itself Used covariates at the X-Axes: Intermediate products Cross-validation –bulk density (oven dry) Groups (x -axes) Description no_dlr All covariates excluding DLR/ SCMaP products no_dlr_src All covariates excluding DLR/ SCMaP covariate: Soil Reflectance Composite no_dlr_src_mos All covariates excluding DL/SCMaP covariates using the mosaic of MREF and SRC and SRC itself Used covariates at the X-Axes: Example France SCMaP products Bare Soil Pixel Count •Sentinel-2 •2018 –2022 •PV+IR2 •Regionalised thresholds 0 62 Paris Nancy Example France SCMaP products Bare Soil Frequency [%] •Sentinel-2 •2018 –2022 •PV+IR2 •Regionalised thresholds 0 % 60 % Paris Nancy Example France Soil parameter Soil Organic Carbon Content •Topsoil - 0-30cm •g/kg*10 95 % SOC [g/kg*10] Paris Nancy 0 2500 pH [pH*10] Example France Soil parameter pH in water •Topsoil - 0-30cm •pH*10 43 78 Paris Nancy Example France Soil parameter 15 37 Total Nitrogen •Topsoil - 0-30cm •[g/kg*10] Total Nitrogen [g/kg*10] Paris Nancy Example France Soil parameter Bulk density, oven dry •Topsoil - 0-30cm •[TBD] Bulk Density [TBD] Paris Nancy 80 140 Data access Data policy and web portals CUP4SOIL (EU - FPCUP) License CC BY 4.0 Products 5 Years SCMaP and Soil products (2018 –2022) • Mean + Soil reflectance composites • Statistic-Products (Frequency, Valid pixels, etc.) • Bare soil mask Yearly SCMaP products Soil frequency, Soil count, Valid pixels, Soil mask Soil parameters: •SOC, texture, pH, bulk density, etc •Associated accuracy / uncertainty Format Cloud Optimized GeoTiff (COG) 20 m (Europe) Web Platform DLR Geoservice ( Browsing, Webservices, Download, STAC) ISRIC Webportal (Browsing, Webservices, Download) Purpose: •Provide a set of data products for stakeholders to „play around“ and get first experiences •Development of show cases •Special emphasis on validation / accuracy / uncertainty of the products (more products in progress) •Explore the pros and cons of SOC maps fromWorldSoils and CUP4SOIL and other sources (SoilGrids, Holisoils, …) Outlook Summary and future developments •Summary: •DLR and ISRIC partnered to produce: •SCMaP intermediate products •Soil parameter •Test about the beste choise of covariates –direct spectral covariates could improve the modelling •Data will be published and available •Future developments: •Preparing the webserver •Comparison ofWorldSoils SOC with CUP4SOIL SOC •Validation / Uncertainty by „spatial pattern agreement“ („How well does digital soil mapping represent soil geography “) •Further CUP4SOIL presentation on: •IUSS Centennial –May 2024 Heiden et al - High resolution soil quality indicators maps for Europe •EGU 14–19 April, 2024, Vienna, Austria: Laura Poggio et al.: European high resolution soil quality products •IGARSS 7 - 12 July, 2024, Athens, Greece: Uta Heiden et al.: „High resolution soil products at European scale integrating remote sensing information” CUP4SOIL - Thank you very much! [email protected] [email protected] [email protected]