EmblA: Ensemble-based data assimilation for environmental monitoring and prediction
EmblA is the Nordic Centre of Excellence for ensemble-based data assimilation (DA). It supports Nordic users of DA methods with instruction and training, state-of-the-art open-source codes, and works for R&D in several applications.
Objectives
A major challenge today is to create precise, computer-based climate-change models, which are essential for making reliable predictions of future climate and environmental conditions.
Systems for forecasting e.g. the weather are based on two sources: knowledge about the physics of the flow (which is converted into a numerical model that computers can run) and observations from weather stations and satellites. Neither of these sources alone can provide sufficient information for making reliable forecasts. The objective of this new NCoE is to develop methods that efficiently combine observational data in order to apply this data in practice to predict future conditions and assess uncertainty in forecasts. Scientifically the NCoE domain is in the interface between mathematics, geostatistics, geophysics, physics and eScience.
Project Summary
Our environment is undergoing major anthropogenic and natural changes that we need to understand, attribute and predict. To be able to use complex 3-dimensional models of our environment to generate reliable predic- tions with documented accuracy is therefore a major challenge for e-Sciences.
Fortunately, the steady increasing availability of e-Infrastructure supports the developments and operational usage of advanced numerical models. The last decade has seen a spectacular increase of the number of parallel processing units, while the processors peak performance only increased marginally. This tendency has strongly guided the model development strategies and will continue to do so.
In parallel with the increasing availability of distributed computing capacity there has been a transition in math- ematical modeling from using a single-realization deterministic model to multi-realization stochastic models. A single model solution only describes one possible realization of a physical system and provides no informa- tion about the prediction uncertainty. By using multiple realizations, it is possible to represent and predict a physical system with its uncertainty. More importantly, when multiple model realizations are used, ensemble- based data-assimilation methods, e.g. Ensemble Kalman Filter and Ensemble Smoother, can be used to condi- tion the model solution on available observations. Recent iterative versions of these methods show increasing similarities with variational techniques.
Ensemble methods are currently used for model conditioning in several operational environments (see next section) and the purpose of the NCoE on ”Ensemble-based data assimilation for environmental monitoring and prediction,” is to further develop ensemble methods for data assimilation in new applications of environ- mental research across the Nordic countries. Thus, the center will both work on methodological research and development as well as contribute to the widespread use of ensemble methods in new operational and research environments within the Nordic countries.
Acknowledgment
On publications, please acknowledge the EmblA Nordic Center of Excellence funded by NordForsk contract number 56801
