OpenDA tutorial: Introduction to Data Assimilation

Data assimilation is about the combination of two sources of information, computational models and observations, in order to exploit both of their strengths and compensate for their weaknesses.

Computational models are available nowadays for a wide range of applications: weather prediction, environmental management, oil exploration, traffic management and so on. They use knowledge of different aspects of reality, e.g. physical laws, empirical relations, human behaviour, etc., in order to construct a sequence of computational steps, by which simulations of different aspects of reality can be made.

The strengths of computational models are the ability to describe/forecast future situations (also to explore what-if scenarios), and a large amount of spatial and temporal detail. For instance weather forecasts are run at ECMWF using a horizontal resolution of about 50 km for the entire earth and a time step of 12 minutes. This is achieved with the tremendous computing power of modern day computers, and with carefully designed numerical algorithms.

However, computations are worthless if the system is not initialized properly. "Garbage in, garbage out". Further the "state" of a computational model may deviate from reality more and more because of inaccuracies in the model, aspects that are not considered or not modelled well, inappropriate parameter settings and so on. Observations or measurements are generally considered to be more accurate than model results. They always concern the true state of the physical system under consideration. On the other hand, the number of observations is often limited in space and in time.

The idea of data assimilation is to combine model and observations, and exploit as much of the information contained in them.