Towards a robust framework for data assimilation and uncertainty quantification in environmental forecasting

Environmental forecasting involves the use of increasingly complex models to predict the response of ecosystems to external drivers or anthropogenic pressures. An ongoing challenge in this context is to quantify and reduce uncertainties in model inputs, parameters and structure, and to understand their impact on model predictions. In this talk, I will review our current methodological framework for assessing and reducing such model uncertainties, present their application in several case studies, and discuss current limitations and issues in the field. I will argue that structural uncertainty remains one of the most challenging problems in environmental forecasting. Hybrid approaches that combine machine learning or AI with process-based modelling approaches are among the most promising candidates to address this issue. I will conclude that the consistent consideration of all uncertainties is essential for the correct interpretation of model predictions and thus forms the basis for a responsible use of models in ecological forecasting.