Farming practices and interventions that sequester and store carbon in agricultural soils can significantly contribute to climate change mitigation, as well as support farmer’s adaptation to increasing frequency of extreme events, droughts and floods. Development of policy, as well as modelling the impact of different policy scenarios (as used e.g. in IPCC AR6) necessitate a good evidence base about the current quantity and pattern of soil carbon in agricultural soils.
The aim of this PhD studentship would be to significantly improve on the accuracy of EU-wide mapping of topsoil organic carbon content. For comparison, the current state-of-the-art map produced by the European Commission Joint Research Centre (JRC) explains only 21% of the variance in mineral soils and 6% in organic soils (Brogniez et al. Eur. J. Soil Sci. 2014). The JRC map was produced using LUCAS (land use & coverage area frame survey) soil survey data from 2009, whereas the more recent LUCAS soil survey in 2018 used improved protocols (measuring soil bulk density) and has ready-to-go spatial framework for analysis using Copernicus satellites, specifically Sentienl-2 (see Figure 1).
Specifically, the following objectives will be explored:
- Adapting methods developed for Landsat satellite (e.g. Dematte et al. Sci. Reports, 10, 4461 2020) to produce bare soil reflectance maps using Sentinel-2.
- Using machine learning and/or generalized regression models to predict LUCAS topsoil carbon content from bare soil spectra (cf. Sorenson et al. Geoderma 401, 115316 2021).
- Produce a new European-wide topsoil carbon content map, including uncertainty.
- Analyse patterns of soil carbon content for different pedo-climatic regions of Europe and for various arable crops/rotations.
The project will combine optical imagery from Sentinel-2, topsoil data from >2000 sites collected in LUCAS 2018 soon-to-be-published by JRC (Arwyn Jones/JRC-Ispra, personal communication) and advanced statistical analysis and/or machine learning (e.g. Random Forest regression) using Google Earth Engine. The potential to use SAR satellite data from Sentinel-1 and/or SMAP will be explored to correct for surface roughness, crop residue (non-photosynthetic vegetation) and soil moisture – all impacting bare soil reflectance mapping from multispectral sensors.