Combining process-based and deep learning models to understand the dual role of microbes as decomposers and stabilizers of soil carbon
BackgroundRecent process-based models of soil organic matter cycling have emphasized the dual role of soil microbes. On the one hand, increased microbial activity enhances the decomposition of soil organic matter (SOM) and hence decreases stocks. On the other hand, the death of microbes can lead to the formation of mineral-associated organic matter (MAOM), which is believed to be more stable. Understanding and modelling why and how microbes contribute to SOM decomposition versus SOM stabilization is therefore pivotal for projecting future soil carbon changes. There are, however, still large uncertainties since, for example, many model parameters associated with microbial growth, carbon use efficiency (CUE), turnover and stabilization cannot easily be measured in the field and are not available at global scales.
Machine learning methods can offer a potential solution. They have become a mainstay not only in Earth sciences in general but also in soil organic matter research. Projects like soilgrids.org have employed machine learning to estimate soil organic carbon stocks globally from datasets of soil profiles worldwide. However, the black-box nature of machine learning hinders understanding of why and how soil organic matter persists in soils. Combining process-based and deep learning models into so-called hybrid models can offer the best from both worlds (Reichstein et al., 2019). We can incorporate the already existing knowledge into process-based models while retaining the data-adaptiveness of deep learning that allows the incorporation of new large-scale datasets.
Goals and tasksThe prospective PhD student will contribute to implementing a hybrid version of process-based soil carbon models developed at the institute (Ahrens et al., 2020, Wutzler et al., 2017). The hybrid model (Figure) will then be trained on new global datasets on the partitioning between mineral-associated and particulate organic carbon, their radiocarbon contents, and microbial properties. The PhD project will tackle one or more of the following questions:
Your profileWe are looking for a highly motivated PhD student who sees scientific progress as a team effort. The successful candidate should have a broad interest in soil processes and should strive to deepen their understanding of soil organic carbon decomposition and formation. The ideal candidate has already a strong background in soil biogeochemistry, numerical modelling, or machine learning. With a willingness to learn and curiosity for new fields, one of the three backgrounds can be enough for a good candidate. The successful candidate holds a master in geoecology, biogeosciences, environmental sciences, biology, mathematics, forestry, physical geography, or something related. Finally, very good oral and written communication skills in English are required. The Max Planck Society seeks to increase the number of women in those areas where they are underrepresented and therefore explicitly encourages women to apply. The Max Planck Society is committed to increasing the number of individuals with disabilities in its workforce and therefore encourages applications from such qualified individuals.
Working environmentThe Max Planck Institute for Biogeochemistry in Jena offers an exceptional dynamic, creative, international, and multidisciplinary working environment. The successful applicant will join the Soil Biogeochemistry group, which is encompassing experimental and theoretical work on the persistence and sensitivity of organic carbon in soils, and interactions between biogeochemical cycles of carbon, nutrients, and water at all spatial scales. The project will be conducted in close collaboration with the MPI-USMILE group on Understanding and Modelling the Earth System with Machine Learning.
For further information, please contact Bernhard Ahrens (email@example.com), Thomas Wutzler (firstname.lastname@example.org), Marion Schrumpf (email@example.com), and Markus Reichstein (firstname.lastname@example.org).
ReferencesAhrens B, Guggenberger G, Rethemeyer J et al. (2020) Combination of energy limitation and sorption capacity explains 14C depth gradients. Soil Biology and Biochemistry, 148, 107912.
Reichstein M, Camps-Valls G, Stevens B, Jung M, Denzler J, Carvalhais N, Prabhat (2019) Deep learning and process understanding for data-driven Earth system science. Nature, 566, 195-204.
Wutzler T, Zaehle S, Schrumpf M, Ahrens B, Reichstein M (2017) Adaptation of microbial resource allocation affects modelled long term soil organic matter and nutrient cycling. Soil Biology and Biochemistry, 115, 322-336.
Hybrid modelling framework that links machine learning (Neural Network in figure) and a process-based soil organic carbon model through learning of hybrid mechanistic parameters.
>> more information about the IMPRS-gBGC + application