There is much worldwide interest in accurate estimation of the global soil organic carbon stock. This is because soil carbon storage can help mitigate climate change. For instance, the 4per1000 initiative aims to increase the soil organic carbon stock with 0.4 per cent each year. If this can be achieved, it would compensate for all emissions from fossil fuels. But evaluation of the effectiveness of 4per1000 measures requires that we can accurately estimate the (changes) in the global organic carbon stock. The aim of this thesis research is to contribute to this requirement by using the SoilGrids global soil map to compute the global carbon stock and analyse how uncertainties in the SoilGrids product propagate to the global carbon stock estimate.
You will first get acquainted with the SoilGrids product by studying which statistical and geo-informatics techniques were used to derive the product. Next you will compute the global carbon stock based on 3D global soil property maps of soil organic carbon concentration, bulk density and coarse fragments. You will compare the result with other estimates of the global soil organic carbon stock as published in the scientific literature. Once this is done you will model the uncertainty by building a geostatistical model of the discrepancy between SoilGrids carbon stock predictions and point values of the carbon stock. This will be a fairly simple model but regional variations will need to be taken into account, while another difficulty lies in handling big data issues. It may be useful to tackle the big data issues by first working for a smaller part of the world (e.g. France or USA) and next scale up to the globe. Once the geostatistical model is derived it will be used to create ‘possible realities’ using spatial stochastic simulation techniques. These possible realities are next used as input to a Monte Carlo uncertainty propagation analysis, which ultimately yields an estimate of the global carbon stock while also quantifying the estimation error. If time permits you will also analyse the sensitivity of the results to assumptions and parameter settings.
- Understand the scientific literature on global soil organic carbon stock estimation
- Model the uncertainty in SoilGrids products by developing a geostatistical model of the errors in the SoilGrids carbon stock map and by calibrating this model using point data of observed differences
- Be able to handle big data issues by familiarising with appropriate software and hardware environments
- Run a Monte Carlo uncertainty propagation analysis and interpret results
- Global soil organic carbon stock estimation, e.g. Stockmann et al. (2015)
- Spatial Stochastic simulation and Monte Carlo uncertainty propagation analysis (SMS readers)
- High-performance computing for big-data problems, e.g. Knaus et al. (2009)
- Solid background in geostatistical modelling, such as obtained through the Spatial Modelling and Statistics course
- Experience with programming in R
- Geo-informatics skills, in particular for handling big-data situations
Theme(s): Modelling & visualisation