To evaluate soil carbon sequestration and land degradation neutrality policies and measures there is a need for a global web-based platform to inform on the status and trends of soil organic carbon (SOC). Among others, such a platform requires a statistical methodology that allows to predict SOC in space and time from SOC point observations and spatial and spatio-temporal maps of environmental covariates.
This project develops, implements and applies a statistical space-time SOC mapping methodology. In the first phase Argentina was used as a pilot and SOC concentration and stocks were predicted on annual basis from 1982 to 2017. In phase 2 an extension to the globe is made, both using a UNCCD modified IPCC approach and machine learning. Web-based visualization of the resulting time series SOC maps is led by a partner in the project.
- Assemble soil profile data and covariates.
- Develop space-time statistical models and calibrate these using the available data.
- Use the calibrated models to make space-time predictions of SOC concentration and SOC stocks for a chosen time period.
- Quantify the uncertainties associated with the predictions and predicted temporal trends.
- Share resulting time series SOC maps publicly in a web-based interactive portal.
Deliverables (not all freely accessible)
- Project reports phase 1: Description of the machine learning methodology and application to the Argentina pilot.
- Project reports phase 2: Description of the UNCCD modified IPCC approach and application to the globe.
- Journal article 'Machine learning in space and time for modelling soil organic carbon change'.
- Web-based interactive portal Soils Revealed
Partners within the project consortium are ISRIC, the National Institute of Agricultural Technology (Argentina), Woodwell Climate Research Center (USA), Cornell University (USA) and Vizzuality (Spain).
The project is commissioned by The Nature Conservancy (USA).