Overview

The Biogeochemical Model-Data Integration group aims to better represent global climate-soil-vegetation interactions, to recognize the current state of the global terrestrial biosphere as well as predict the Ecosystems' behaviour under different past and future environmental conditions by developing new modelling and model-data integration approaches.

The current debate about “global change” mostly emphasises the greenhouse effect, the associated warming of the atmosphere and the feedback through the carbon cycle. However, the Earth is much more complex. For a comprehensive understanding of the Earth system the interaction of the carbon cycle with water and nutrient cycles and the role of vegetation-soil feedbacks have to be addressed much more thoroughly. Hence the Biogeochemical Model-Data Integration Group (BGC-MDI) is dedicated to develop new methods and models capable of better diagnosing the state and dynamics of the terrestrial biosphere. This should allow to better reconstruct, as well as predict, the Ecosystems’ behaviour under different past and future environmental conditions.

Hence, our activities relate most strongly to the institute’s research questions:

“What are the key processes and factors controlling biogeochemical fluxes?” and “How can we extract information from multiple data streams about (non-linear) biogeochemical processes?”.

Within this overall context, we are addressing two key questions with various sub-questions:

• What is the role of the soil in the Earth System? How can its representation in global ecosystem models be improved?
  • How important is the development of vertical soil profiles and organic horizons for biogeochemical cycles?
  • What is the role of soil biota, such as fungi, bacteria, roots, and macrofauna, in the carbon cycle?
  • How strongly does the carbon cycle feedback on the water cycle in the soil?
  • How will the soil organic matter (SOM) dynamics change under changing climate and vegetation cover in different soil types and climate zones?
• How can the contemporary network of ecosystem observations (fluxes and stocks), in combination with manipulation experiments and spatial data (e.g. remote sensing, streamflow data), facilitate a deeper understanding of major biogeochemical cycles and biosphere-atmosphere interactions?
  • Which external and biological factors influence spatial and temporal variability of carbon, water, and energy cycles at different scales? What are the mechanisms? Do ecosystems respond to climate variability in a dampening or amplifying way?
  • Can we derive functional ecosystem ‘properties’ from observations? Can these properties be linked to plant traits and be used for an improved functional classification of global ecosystems? Can we identify tradeoffs between ecosystem properties (e.g. water-, radiation- and nitrogen-use efficiencies)?
  • Can we infer fully empirical estimates of the global carbon and water cycles by integrating point observations (e.g. eddy covariance fluxes) with spatially gridded (e.g. remote sensing) or integrated (e.g. streamflow, atmospheric CO2 concentrations) data via novel machine-learning approaches?
  • How can we best use these data streams to evaluate and constrain Earth System models?

Methodological approach

The approach towards achieving improved diagnostics and prediction of biogeochemical cycles builds upon the notions that

• no single observation source is able to fully constrain ecosystem models. Instead, different data streams contain complementary information, and are used together only to embrace the range of temporal and spatial scales addressed (multiple constraint approach).
• both process- and data-oriented modelling approaches have complementary strengths and should be used in synergy. For example, data mining and time-series analysis techniques may be used to characterize model-data residuals. Thus, they can help to identify missing processes and time-space scales in the current process models. In addition, data oriented methods aim at recognizing general patterns (e.g. spatial gradients) in the data, which should be reproduced by process-oriented ecosystem models. On the other hand, process-modelling fosters insight into the underlying system structure and processes. It also has the capacity to extrapolate, based on first principles. Modelling experiments over huge spatiotemporal domains can be used to directly study feedback mechanisms between the atmosphere and the land surface.

Our model-data integration approach differs from typical data-assimilation schemes (e.g. in Numerical Weather Prediction) in the sense that we view parameter uncertainty only as part of the problem. We follow a model-data integration cycle, where also model structure and data validity are tested, and where ecological parameter interpretation plays an important role.