Updated global soil property and class maps LandGIS Open data and services - please contribute

167 views
Skip to first unread message

Tomislav Hengl

unread,
Jan 7, 2019, 7:12:58 AM1/7/19
to Global Soil Information
In addition to the newsletter we have sent few days ago, I am forwarding you some more information about our new LandGIS system (https://landgis.opengeohub.org; our general platform for global soil, vegetation, biodiversity, climate... mapping) focused more on global and regional soil mapping.

We have made significant progress (in comparison to https://doi.org/10.1371/journal.pone.0169748) with improved mapping of the global distribution of soil properties and classes (for complete text refer to: https://github.com/Envirometrix/LandGISmaps#soil-properties-and-classes). Improvements include:
  1. New RS-based layers have been added to the list of covariates including Copernicus FAPAR time series of images, MERIT DEM multiscale DEM derivatives, ALOS radar images, MODIST LST aggregates and IMERGE precipitation images,
  2. Selection of the ensemble model is now run via the SuperLearner package framework (https://cran.r-project.org/web/packages/SuperLearner/vignettes/Guide-to-SuperLearner.html), which allows for incorporating spatial subsetting in the Cross-Validation of models,
  3. An estimate of the prediction uncertainty / error is provided via the ensemble model standard deviation,
  4. Soil type maps are now based on over 360,000 training points (compare with ca 70,000 in the previous run),
  5. For each point quality flag (completeness) is used as case.weights in the training process,
  6. Artifacts / extrapolation problems, especially in the soil carbon maps, are now dealt with by using the most up-to-date (ESA) land cover map,
Most importantly, several bugs in the predictions (https://doi.org/10.1371/journal.pone.0169748) have now been resolved:
  1. The previous bug of about two times higher total SOC (https://doi.org/10.1002/2017GB005678) has now been resolved. A new complete and consistent soil carbon content (https://doi.org/10.5281/zenodo.1475457), bulk density (https://doi.org/10.5281/zenodo.1475970), and soil carbon stock maps (https://doi.org/10.5281/zenodo.1475453) can be download under the CC-BY-SA license.
  2. Soil property predictions are now based on better (spatially) represented organic soils of peat-lands, mangrove forests and similar.
  3. Cross-validation and selection of the models for ensemble predictions is based on spatial cross-validation (see: https://github.com/Envirometrix/LandGISmaps/blob/master/soil/soil_properties/LandGIS_soil_property_maps.R#L47) which results in more realistic estimates of average prediction errors.
We have also started registering all data versions (i.e. adding a DOI to every version of the data and code) so that improvements and problems can be back-tracked. To list all LandGIS layers available for download via Zenodo and access all previous versions you can try using: https://zenodo.org/search?page=1&size=20&q=LandGIS

To access LandGIS at individual locations, please use LandGIS REST API (https://landgisapi.opengeohub.org/). This allows access to any layer uploaded to the LandGIS Geoserver and available via the LandGIS app (https://landgis.opengeohub.org). The following code, for example, will fetch values of the soil type (USDA great group) at a point:

which gives:

Project description
0 "https://opengeohub.org/about-landgis"
Description of all codes
0 "https://github.com/envirometrix/landGISmaps/"
response
0
lon -95.5811
lat 30.2543
sol_grtgroup_usda.soiltax_c_250m_s0..0cm_1950..2017_v0.1.tif 30
info
0
X 30
Number 30
Group "paleudalfs"
Great_Group_2015_match "Paleudalfs"
Suborder "Udalfs"
Order "Alfisols"

FYI, we plan to constantly improve predictions and functionality of LandGIS. Updates of maps are now expected every 2–3 months. Updates of soil property and soil class predictions, however, depend on the amount of feedback and amount of new point data we receive from you, hence please consider contributing point data to increase accuracy of the global predictions. The more quality point data we get from you, the better outputs we can produce.

If you are working on producing soil property, or soil class, maps for your region or area, or if you are part of a national or regional soil mapping team and if the data you produce can be released under some Open Data license, then please contact us (https://opengeohub.org/about). We are currently working collaboratively with several countries to jointly develop mapping systems to produce finer resolution (e.g. 100 m) soil maps by combining the best of global and local models within Open Source Machine Learning frameworks. The merging of the global and local predictions will be based on some of the following approaches:
  • Manon, C., Dobarco, M.R, Arrouays, D., Minasny, B. and N.P.A. Saby. (2019). “Merging Country, Continental and Global Predictions of Soil Texture: Lessons from Ensemble Modelling in France.” Geoderma 337: 99–110. https://dx.doi.org/10.1016/j.geoderma.2018.09.007
  • Ramcharan, A., Hengl, T., Nauman, T., Brungard, C., Waltman, S., Wills, S., & Thompson, J. (2018). “Soil Property and Class Maps of the Conterminous United States at 100-Meter Spatial Resolution.” Soil Science Society of America Journal, 82(1), 186-201. https://dl.sciencesocieties.org/publications/sssaj/abstracts/82/1/186
If you discover a bug, artifact or inconsistency in the LandGIS maps, or if you have a question please use some of the following channels:
Contact us if you would like to learn how to use LandGIS and how to publish your own global maps via LandGIS (https://opengeohub.org/submitting-global-layers-inclusion-landgis). Our special interests are land degradation (including loss of soil carbon https://doi.org/10.5281/zenodo.1475449, soil erosion, soil salinization, loss of biodiversity, deforestation...), mapping land potential and soil productivity, estimating impacts of climate change on future soil and vegetation conditions, promoting Open Data and promoting Open Source software through workshops and training courses.

Let's join forces and make together an "OpenStreetMap-type" system for environmental data!

Yours,

-- 
T. (Tom) Hengl
Technical support / Vice Chair
OpenGeoHub Foundation
Mail: OpenGeoHub Foundation, Roghorst 206, 6708KT Wageningen, NL
skype:tom.hengl?chat

Reply all
Reply to author
Forward
0 new messages