We collected data from the FLUXNET research network of ecosystem flux measurements, paired CO2 exchange flux measurements with remotely sensed vegetation greenness data around respective sites, and soil moisture estimates from a set of sources and models. We then developed a machine learning approach (based on Artificial Neural Networks), designed to detect and quantify soil moisture effects on the efficiency by which vegetation converts CO2 and absorbed light energy into photosynthates (referred to as light use efficiency). This allowed us to isolate the soil moisture effect on photosynthesis from other drivers, including the (partly co-varying) vapour pressure deficit. By analysing patterns in the data from flux measurements and satellites, we found a clear additional effect by soil moisture that substantially reduces photosynthesis during “soil moisture droughts” - an effect that cannot be estimated when solely relying on information on vapour pressure deficit. We quantified this effect for a large number of sites across the globe and characterised sites with respect to their sensitivity to low soil moisture. We found that there is a tendency of sites in relatively wet climates to exhibit a low sensitivity to a drying of the topsoil (max. top 1 m), possibly due to access to water in deeper soil layers. This research was published in New Phytologist (Stocker et al. 2018).
In a next step, we evaluated a set of widely-used remote sensing-based models for estimating vegetation productivity across the globe and over time. The question we asked was whether these models can accurately simulate photosynthesis during soil moisture drought periods, which we have managed to identify precisely in our previous study (Stocker et al. 2018). This was an obvious question to ask because these remote sensing-based models do not account for soil moisture directly, but use information on atmospheric humidity as a proxy for dryness in general. We found that all the models we investigated had a very clear relationship with the timing and magnitude of the apparent soil moisture stress we identified earlier. This indicates that information on soil moisture should be included explicitly in estimates of vegetation productivity - not just atmospheric humidity and remotely sensed vegetation greenness. This research was published in Nature Geoscience (Stocker et al., 2019).
Apart from data analysis, this project yielded both new methods for data analysis, and new model development for global vegetation productivity simulations using a new modular modelling framework (SOFUN):
- Benjamin Stocker. (2018, April 6). sofun: v1.1.0 (Version v1.1.0). Zenodo.
http://doi.org/10.5281/zenodo.1213758(opens in new window)All code and model outputs are made freely accessible through github (
https://github.com/stineb/soilm_global(opens in new window) https://github.com/stineb/nn_fluxnet2015
https://github.com/stineb/fvar(opens in new window)) and on the open access permanent data repository Zenodo:
- Stocker, Benjamin. (2018). fLUE [Data set]. New Phytologist. Zenodo.
http://doi.org/10.5281/zenodo.1158524(opens in new window)- Benjamin Stocker. (2018, February 20). nn_fluxnet2015: Final release (Version submission_3). Zenodo.
http://doi.org/10.5281/zenodo.1181691(opens in new window)- Benjamin Stocker. (2018, April 6). sofun: v1.1.0 (Version v1.1.0). Zenodo.
http://doi.org/10.5281/zenodo.1213758(opens in new window)- Stocker, Benjamin. (2018). GPP: Site-scale and global model outputs from P-model used for Stocker et al. (2019) Nature Geosci. (Version v1.0) [Data set]. Zenodo.
http://doi.org/10.5281/zenodo.1423484(opens in new window)- Benjamin Stocker. (2019, January 18). soilm_global: Code for Stocker et al. (2019) Nature Geosci. (Version v1.0). Zenodo.
http://doi.org/10.5281/zenodo.2543324(opens in new window)