Seminário PGCST: Prof Mathew Williams, Universidade de Edinburgh, UK
Data: 27/11/2019 – quarta-feira
Local: Auditório do Lambda – Prédio Lambda – INPE/SJCampos
Título: “Opportunities and challenges for diagnosing terrestrial C cycling”
Society increasingly relies on complex models of the earth system to support decision making. The IPCC process is a clear example related to climate security, but there are other examples of decision support systems related to food and water security, sea level rise, afforestation and so on. Models have been tremendously useful in scaling fundamental knowledge across space and time. However, a major weakness has been in evaluating model uncertainty and communicating this to decision makers. Uncertainty arises from errors in model structure, forcing and parameters. Here, taking the example of the terrestrial carbon cycle, I discuss how new data products can be fused with simple models to address model errors and to provide forecast uncertainty estimation. This model-data fusion (MDF) eliminates many subjective components of model parameterisation, allowing the interaction between data, forcing and model structure to reveal spatial patterns in underlying processes. In allowing parameter maps to emerge from the model-data fusion, novel ecological information can be exploited. By using a simple model the issue of tuning is completely explicit in MDF, compared to typical complex models. Model-data fusion allows direct quantification of the information content of new global data sets, e.g. for biomass, soil moisture, and how this information propagates through the model (e.g. the role of fire and/or soil moisture in the C cycle). MDF provides an analysis of the current state of the terrestrial C cycle and its processes. As such, it can support robust evaluation of prognostic models of the C cycle, for instance via reliability ensemble averaging. There are enormous opportunities for MDF approaches to evaluate alternate model structures, to maximise the value of new earth observations, and to link to ecological datasets for better constraint. Ensemble outputs provide a means to inform policy makers about the uncertainties in forecasts.