Ensemble-based Methods for Environmental Monitoring and Prediction (EmblA)

The Nordic Centre of Excellence for ensemble-based methods for environmental monitoring and prediction (NCoE EmblA) is working to improve computer models and methods to better assess uncertainties in future climate and environmental conditions. The centre is interdisciplinary and the results will be applicable in numerous areas where forecasting is needed.

Towards greater precision in predicting weather, climate and environmental conditions

At the Nansen Environmental and Remote Sensing Center in Bergen,  Professor Geir Evensen and Project Deputy Leader Laurent Bertino head the EmblA research project (Ensemble-based Methods for Environmental Monitoring and Prediction), a NordForsk Centre of Excellence.

The five-year project is funded under NordForsk’s Nordic eScience Globalisation Initiative (NeGI). The interdisciplinary centre has partners in Norway (Nansen Environmental and Remote Sensing Center), Finland (Finnish Meteorological Institute), Denmark (DHI) and France (MINES ParisTech), and incorporates competencies in the fields of mathematics, geostatistics, physics and eScience. Altogether, ten researchers work together at NCoE EmblA.

The focus is on developing mathematical methods that can enhance the assimilation of observational data related to weather, climate and the environment. The research project combines numerical, computer-based models with empirical observations.

Geir Evensen







Professor Geir Evensen, Nansen Environmental and Remote Sensing Center

Uncertainty in predicting climate conditions will decrease

The objective of EmblA’s work is to improve the reliability of predictions of future weather and climate conditions – i.e., more precisely, to minimise uncertainty in forecasting weather, among other things.

“When a storm is forecast it is very important to have access to the most extreme scenario, for the storm can be dangerous if it reaches a certain intensity. EmblA is working to develop a system that gives us insight into the level of uncertainty in a prediction, which makes it possible to assess worst-case scenarios,” Professor Evensen explains.

EmblA combines observational data with computer modelling to improve the reliability of forecasts and, in particular, to measure more precisely the uncertainties of a prediction. What will happen in the future and what is the range of uncertainty?

It is valuable to be able to calculate both more precise averages and to foresee the amplitude of extreme conditions. This applies not only to weather forecasts for tomorrow, but also to guiding shipping traffic through sea ice, to evaluating whether air traffic should be shut down due to ash clouds from volcanic eruption and where offshore drilling rigs should be situated and how strong they need to be. The centre’s activities can have many areas of application since the method in focus is based on mathematical grounds.

“EmblA’s efforts can, for example, be used to determine if drilling an oil well will be lucrative. When planning the dimensions of an oil well, different scenarios are devised for how much it costs to make and for how much it can produce. These scenarios contain a large degree of uncertainty. If the well is too advanced it will be too expensive to operate. If it is too simple it won’t produce enough which will also make it too costly. The ability to assess uncertainties better will help to save a lot of resources,” Dr Evensen states.

What will the weather be like in Bergen tomorrow?

A wide array of models for predicting weather conditions exist today. The challenge for EmblA was to develop computer calculations by integrating the deviations between prediction and reality to minimise uncertainty. Geir Evensen explains:

“Imagine that you have produced a weather forecast for the following day but when the next day arrives your prediction turns out to be inaccurate. There is a discrepancy to account for. If the forecast and the observation do not match up, the information can be used to improve the model being developed. This is where ‘ensemble-based methods’ come in. We record the difference between prediction and observation and then try to update the model to generate more reliable prognoses. The results we obtain have a quantifiable range of uncertainty. That is, we know how uncertain the results are. So we would know that tomorrow’s temperature in Bergen will be between 21 and 23 degrees Celsius rather than knowing that it will be close to 22 degrees.”

Laurent Bertino adds:

“Our work can be compared with looking at a map on a table. It is often taken as reality, but it isn’t; it’s just a map. Maps contain uncertainties, but it’s hard to see this when you only look at one. If you put more maps together you start to see things differently.”

EmblA works with models that describe physical systems such as ocean circulation, the spreading of volcanic ash clouds, reservoir description or a groundwater system. The basis is a mathematical model into which data is entered and which generates output for analysis.

“In traditional modelling you select a single value for input and this gives you a result. If a value is chosen from here,” says Dr Evensen, pointing to a graph, “the result generated may not really be representative. It is not a good idea to make major decisions based on a single result, so you need to look at the effect of the overall uncertainty and the range of possible outcomes that follow. This makes it possible to state what has the greatest likelihood of happening. The results have a minimum and a maximum point with the uncertainty extending between them. It is a stochastic process instead of the deterministic one which is used in many disciplines,” Geir Evensen states.

“We typically deal with nonlinear models and we work with multiple dimensions – that is, there are millions of unknown factors. Consequently, we can seldom use just personal computers because we need large machines with high processing power,” Dr Bertino says.

In EmblA we want to learn from outside our own discipline

NCoE EmblA was launched at the beginning of 2014 and has found its stride.

Laurent Bertino:

“I feel we are well underway. The project methodology is in place so that all partners are using the same methods, and we exchange personnel to learn from one another and to make sure that we are working towards the same goals. EmblA is a highly interdisciplinary project. I know only a little about what they know in Finland and vice-versa. What we have in common are the methods.”

What has posed the greatest difficulty thus far? Geir Evensen answers without hesitation: “Our most difficult task has been finding people. The project requires personnel with highly specialised expertise and there are not many candidates out there. The low number of Nordic candidates that apply for our doctoral fellowships has surprised us. Fortunately, we have applicants from all over the world and only one position remains to be filled. It can be challenging to evaluate international candidates because we aren’t familiar with the universities where they were educated, so the process can become quite comprehensive.

The best? Laurent Bertino has no doubts either: “It’s very interesting to see that some of the methods we have been working on for several years can now be adopted and used in other subject areas. This makes our work very exciting.  The fact, for example, that we can engage in scientific discussions with a meteorologist in Finland and an oceanographer in Norway is something that may otherwise have been difficult to arrange.”

EmblA’s activities comprise both basic and applied research at the same time. The researchers are working with complicated mathematical problems, yet the resulting models and methods can relatively quickly be applied by other researchers, also for commercial purposes.

“Our work is part of an expert system where each of us needs to learn something from other disciplines. We don’t know exactly who will be using our methods, but they are openly accessible on the Internet. The only form of protection is knowledge, because there are few people capable of using it. Everything is published and the code is accessible as well, so that others can download it and try it out,” says Geir Evensen.

Text: Linn Hoff Jensen

Facts about the project

Project number: 56801

Project leader

Geir Evensen, professor II of applied mathematics at the Nansen Environmental and Remote Sensing Center in Bergen

5 år

Science Advisory Board

Prof. Robert Gurney, Uni. of Reading, Dept. of Meteorology, UK

Dr. Sylvie Joussaume, National Institute of Sciences of the Uni. at CNRS, France

Dr. Serge Guillas, Uni. College London, Dept. of Statistical Science, UK

Dr. David Wallom, Oxford eResearch Centre, UK

Prof. Peter Jan van Leeuwen, Uni. of Reading, Dept. of Meteorology, UK