- What is "SoilGrids"?
- What do the filename codes mean?
- How were the spatial predictions generated?
- How were the legends generated?
- Which soil properties are predicted by SoilGrids?
- What is the SoilGrids data-sharing policy?
- How can I access SoilGrids?
- How can I use the Homolosine projection?
- How can I use SoilGrids in a different projection?
- When 100 metre resolution?
- Which soil mask map was used?
- How is SoilGrids related to the GlobalSoilMap IUSS working group and its specifications?
- What happened to the maps of soil types?
- How can I help improve SoilGrids?
- How can SoilGrids help me improve soil maps for my country?
- Who provided soil profile data for the SoilGrids effort?
- What if I did not find an answer to my question?
- Cited sources
SoilGridsTM (hereafter SoilGrids) is a system for global digital soil mapping that makes use of global soil profile information and covariate data to model the spatial distribution of soil properties across the globe. SoilGrids is a collections of soil property maps for the world produced using machine learning at 250 m resolution. It is based on a workflow that allows updates when new soil profile observations or new covariates become available. Predictions are made at six standard depths. SoilGrids uses global models that make use of all available input point data to map a property across the globe. This results in consistent predictions (no abrupt changes in predicted values at country boundaries, etc).
SoilGrids maps are a global soil data product generated at ISRIC — World Soil Information as a result of international collaboration. For more technical and scientific information about SoilGrids contact the development team. For more information about ISRIC and collaboration possibilities please contact the ISRIC Director.
- SoilGrids250m = a set of global maps of soil properties for six depth intervals at 250 m spatial resolution.
- SoilGrids = a system for global digital soil mapping.
- SoilGrids.org = portal to web-services providing access to SoilGrids prediction maps.
SoilGrids spatial predictions (layers) are produced using a reproducible soil mapping workflow, and can therefore be regularly updated as new soil data or covariates become available, after quality control and data standardisation/harmonisation.
Each map in SoilGrids has three components:
- a master VRT file;
- an OVR file with overviews for swift visualisation;
- a folder with GeoTIFF tiles.
Each component name is a triplet separated by underscores:
property_depthInterval_quantile For instance, the file
cfvo_5-15cm_Q05.vrt is the master file for the 5%-quantile prediction of coarse fragments in the 5 cm to 15 cm depth interval. Below each of the components is explained in more detail.
The table below shows the properties currently predicted with SoilGrids, their description and mapped units. All maps produced with SoilGrids store data as integer values to minimise storage space. Therefore, some properties are provided in units that are not so common in soil science, such as permil or ‱. By dividing the predictions values by the values in the Conversion factor column, the user can obtain the more familiar units in the Conventional units column.
|Name||Description||Mapped units||Conversion factor||Conventional units|
|bdod||Bulk density of the fine earth fraction||cg/cm³||100||kg/dm³|
|cec||Cation Exchange Capacity of the soil||mmol(c)/kg||10||cmol(c)/kg|
|cfvo||Volumetric fraction of coarse fragments (> 2 mm)||‱||100||%|
|clay||Proportion of clay particles (< 0.002 mm) in the fine earth fraction||‰||10||%|
|nitrogen||Total nitrogen (N)||cg/kg||100||g/kg|
|sand||Proportion of sand particles (> 0.05 mm) in the fine earth fraction||‰||10||%|
|silt||Proportion of silt particles (≥ 0.002 mm and ≤ 0.05 mm) in the fine earth fraction||‰||10||%|
|soc||Soil organic carbon content in the fine earth fraction||dg/kg||10||g/kg|
|ocd||Organic carbon density||g/dm³||1000||kg/dm³|
|ocs||Organic carbon stocks||t/ha||10||kg/m²|
SoilGrids predictions are made for the six standard depth intervals specified in the GlobalSoilMap IUSS working group and its specifications:
|Interval I||Interval II||Interval III||Interval IV||Interval V||Interval VI|
|Top depth (cm)||0||5||15||30||60||100|
|Bottom depth (cm||5||15||30||60||100||200|
SoilGrids maps have associated uncertainties as any product derived from a modelling approach. The prediction uncertainty is quantified by probability distributions. For each property and each standard depth interval this distribution is characterised by four parameters:
Q0.05- 5% quantile;
Q0.50- median of the distribution;
mean- mean of the distribution;
Q0.95- 95% quantile.
SoilGrids uses state-of-the-art statistical methods for digital soil mapping, relying exclusively on open source tools. The models are tailored per soil property and fitted using documented models. For each property, a global model is calibrated using a spatially stratified 10-fold cross-validation procedure. The model produces values at each map location (cell) and standard depth. The prediction distribution is captured in four different maps reporting its 5%, 50% and 95% quantiles, and the mean.
The ‘mean’ and ‘median (0.5 quantile)’ may both be used as predictions of the soil property for a given cell. The mean represents the ‘expected value’ and provides an unbiased prediction of the soil property. The median yields that value for which there is a 50% probability that the true soil property value is greater and a 50% probability that the true value is smaller. For symmetric distributions the mean and median will be identical, while the mean is greater than the median for distributions that are skewed to the right (such as soil organic carbon concentration).
The 0.05 and 0.95 quantiles present the lower and upper boundaries of a 90% prediction interval and may be used as a measure of prediction uncertainty following the GlobalSoilMap IUSS working group and its specifications. This interval presents a value range that contains the true soil property value for each cell (which one would measure from a soil sample taken at the centre of the cell) with 90% probability.
Quantiles of the distribution were computed with Quantile Regression Forests (Meinhausen, 2006) as implemented in the ranger package in R. The mean was computed using the default random forests algorithm.
For each soil property a standard global legend was created, based on the global sample histograms derived from the compilation of standardised soil profile data provided by WoSIS. Soil legends are available via the GSIF package for R and can be used to program data visualization.
Histogram for soil pH (in H2O) and corresponding colour legend. Note: breaks in the legend colours have been selected using histogram equalization (i.e. by using constant quantiles) to ensure maximum contrast in the output maps.
Legends for Organic Carbon Stocks (ocs) and Organic Carbon Density (ocd) were created with equal size intervals.
SoilGrids contains predictions and associated prediction uncertainties for basic soil properties, following the GlobalSoilMap IUSS working group and its specifications: pH (in water), texture fractions, coarse fragments, bulk density, total nitrogen, organic carbon concentration and cation exchange capacity. SoilGrids provides also predictions for 'complex' soil properties such as organic carbon densities at the six standard depths and organic carbon stocks for topsoil (0-30cm) and subsoil (30-100cm, in development). The list of targeted soil properties will be gradually extended, based on user requests and the availability of soil observations.
Since 2019, SoilGrids products are provided under the CC BY 4.0 (publicly accessible environmental data; see also the ISRIC software and data policy). SoilGrids is a contributes to other public global soil data projects. For a review of global soil mapping initiatives and data sets see: Grunwald et al. 2011, Omuto et al. 2013 and Arrouays et al. 2017. The SoilGrids objective is: "global soil data anywhere, anytime, for everyone".
The latest SoilGrids release can be accessed through the following services:
- WMS: swift access for visualisation and data overview.
- WCS: best way to obtain a segment of a map and use SoilGrids as input to other modelling pipelines.
- WebDAV: download the complete global map(s) in VRT format.
A new web mapping platform be available at SoilGrids.org in early 2020.
To deal with an increasing number of inputs and computation demands, SoilGrids has since 2019 been computed on an equal-area projection. After a thorough comparison, the Homolosine projection was identified as the most efficient in an open source software framework (de Sousa et al. 2019). This projection is fully supported by the PROJ and GDAL libraries; therefore, it can be used with any GIS software that rellies on them.
The actual Spatial Reference System (SRS) of the SoilGrids maps is composed by the Homolosine projection applied to the WGS84 datum. This SRS can be added to the PROJ database (the file named
epsg) with the following string:
# ISRIC - Homolosine
<152160> +proj=igh +datum=WGS84 +no_defs +towgs84=0,0,0 <>
The verbose Well Known Text (WKT) version of this SRS is:
The European Petroleum Survey Group (EPSG) never issued a code for this projection. However, some programmes like MapServer require any SRS to be associated with such a code. For that reason a pseudo EPSG code was created to refer to the SoilGrids SRS:
The Homolosine projection is not mandatory in any way. The WMS and WCS publish the SoilGrids maps in the following alternative SRSs:
EPSG:4326- the popular Marinus of Tyre projection (aka Plate Carré) applied on the WGS84 datum. Not recommended, as it expands the surface area of the globe by 60%.
EPSG:54009- Mollweide projection (aka Homolographic) applied to the WGS84 datum (pseudo EPSG code issued by ESRI).
EPSG:54012- Eckert IV projection applied to the WGS84 datum (pseudo EPSG code issued by ESRI).
The VRT-mosaics can themselves be easily reprojected using the gdalwarp tool. Considering the size of each mosaic, it is best to require a VRT also as output. For example:
gdalwarp -t_srs EPSG:3035 -of VRT ./sand_60-100cm_Q0.5.vrt ./sand_60-100cm_Q0.5_3035.vrt
Producing global soil information requires extensive infrastructure and resources. ISRIC has so far delivered complete and consistent global soil information products at 1 km and 250 m resolution. SoilGrids1km and SoilGrids250m are a step towards 100 m global soil property maps. The GlobalSoilMap IUSS working group aims at delivering maps at a finer target resolution of 100 m, and various countries have already done so. It is also on the agenda of the SoilGrids team.
The soil mask map provides an approximation of global coverage of soils, i.e. where soil occurs. For the current SoilGrids release, the global soil mask map was derived from the latest ESA land cover map, with the classes Urban (code 190), inland water (code 210), glacier (code 220) and bare surface (code 200) masked out. Predictions have been produced only for soils with vegetation cover and soils without vegetation cover. No estimate is provided for permanent ice areas since they are subject to extreme climatic conditions. The areas that have been masked out are often under-represented in soil surveys, making it difficult to fit a reliable statistical model.
The global soil mask map was derived from the latest ESA land cover map.
SoilGrids aims to follow closely the GlobalSoilMap specifications. It focuses exclusively on global predictions, currently with 250 m being the finest resolution.
The 2019 SoilGrids release does not include soil class predictions. However, a map synthesising previous predictions of WRB reference groups will be released in the first quarter of 2020.
A novel methodology is currently under development towards a unified soil classification map of the world, that would be more accurate and easier to use. However, the lack of reliable soil class observations remains an important hurdle. If you have access to such data, please consider contributing and help making a global soil classes map.
The actual mapping accuracy of each targeted soil property is still limited: the amount of spatial variation explained by the models varies between 30% and 70%. However. unlike other global soil databases, SoilGrids quantifies the prediction uncertainty. The prediction distribution quantiles can be used to assess the impact of uncertainty in soil predictions on scenario / model testing through, for example, uncertainty propagation techniques.
Within the SoilGrids project, there is a clear intention to gradually improve predictions by incorporating more shared soil profile data, more and more relevant covariates, and by improving the models used. The goal is to increase prediction accuracy with each release. Therefore, please consider contributing/sharing soil property or soil class observations to help incrementally improve the SoilGrids prediction maps.
To improve predictions for your country or region, consider contributing soil profile data to the ISRIC WoSIS database so that your point data can be also used to generate improved predictions. Note that ISRIC will always respect the data policy of the data provider and will not publically share any data unless written permission is given to us to do so. Agencies that contribute data will be acknowledged and listed as a contributing organization on the main SoilGrids portal.
For specific regions, SoilGrids predictions can be used as a covariate to improve prediction of soil properties through a digital soil mapping approach. Predictions for an area of interest can be obtained from the SoilGrids WCS and overlaid with point data. The SoilGrids maps can thus be used as covariates to predict or adjust the values of target variables locally, together with locally available covariates not used by SoilGrids. An example of how a regional scale prediction can be conducted is described in Hengl et al. (2015).
The SoilGrids framework is intended to facilitate global soil data initiatives and to serve as a bridge between global and local soil mapping. Contact ISRIC to explore how we may collaborate to generate or update predictions for your area of interest. To get a better understanding of digital soil mapping techniques, you may also join the Digital Soil Mapping course of our annual ISRIC Spring School.
Suggested uses of SoilGrids by national and regional agencies include:
- use as a covariate layer for regional mapping;
- filling gaps in country-based (bottom-up) global soil information services such as, for instance, developed by the Global Soil Partnership.
- providing input soil data for nationally determined contributions (NDC's), such as for instance required by the UNFCCC.
SoilGrids draws on a large collection of geo-referenced soil profile data for the world that are managed in WoSIS (World Soil Information Service; Batjes et al. (2019). Populating WoSIS has been made possible thanks to the contributions and shared knowledge of a steadily growing number of data providers; we gratefully acknowledge their contributions.
If you have a technical question about SoilGrids that is not answered in this FAQ, please post it to GIS.StackExchange, under the tag
soilgrids. ISRIC staff are subscribed to this tag and will be automatically notified of any new question arising. GIS.StackExchange makes it easier for other SoilGrids users to find quality answers to their questions.
SoilGrids was funded from ISRIC's core funding, with additional support from the EU-H2020 CIRCASA project.
A wide range of agencies and experts have provided data for the WoSIS/SoilGrids effort; we gratefully thank them for their contributions.
- Batjes N.H., Ribeiro E, and van Oostrum Ad (2019). Standardised soil profile data to support global mapping and modelling (WoSIS snapshot 2019). Earth System Science Data Discuss. doi:10.5194/essd-2019-164
- Grunwald, S., Thompson, J. A., & Boettinger, J. L. (2011). Digital soil mapping and modeling at continental scales: Finding solutions for global issues. Soil Science Society of America Journal, 75(4), 1201-1213. doi:10.2136/sssaj2011.0025
- Hengl T, de Jesus JM, MacMillan RA, Batjes NH, Heuvelink GBM, et al. (2014) SoilGrids1km — Global Soil Information Based on Automated Mapping. PLoS ONE 9(8): e105992. doi:10.1371/journal.pone.0105992
- Hengl T, Heuvelink GBM, Kempen B, Leenaars JGB, Walsh MG, Shepherd KD, et al. (2015) Mapping Soil Properties of Africa at 250 m Resolution: Random Forests Significantly Improve Current Predictions. PLoS ONE 10(6): e0125814. doi:10.1371/journal.pone.0125814
- Meinshausen, N. (2006). Quantile regression forests. Journal of Machine Learning Research, 7(Jun), 983-999.
- Omuto, C., Nachtergaele, F., and Vargas Rojas, R. (2012). State of the Art Report on Global and Regional Soil Information: Where are we? Where to go? Global Soil Partnership technical report. FAO, Rome.
- de Sousa, L.M., Poggio, L., Kempen, B.: Comparison of FOSS4G Supported Equal- Area Projections Using Discrete Distortion Indicatrices. ISPRS International Jour- nal of Geo-Information 8(8), 351 (2019) doi:10.3390/ijgi8080351