Chapter 4
Introduction to Numerical Weather
Prediction Data Assimilation
Phillipe Courtier
ECMWF, Shinfield Park
Reading, Berkshire RG2 9AX
UK
4.1 The Problem
In numerical weather prediction data assimilation consists of the process which estimates the
initial conditions of the forecast using all the available information. A description of the current
observing system can be found in McGrath (1993); around 105 elementary pieces of information
are currently used over 24 hours by the operational data assimilation system.
The current operational ECMWF model (Simmons, 1991) has a horizontal resolution of 90 km
and covers the whole globe. On the vertical the atmosphere is sampled with 31 levels from the
surface up to 10 hPa. The number of degrees of freedom of the model is then of the order of
10 7 .
Over 24 hours, the estimation problem is clearly underdetermined. The time dimension is
thus a critical element of any data assimilation system: it is essential to carry forward in time
information from past observations using the forecast model since it is the best information
propagator available.
We have identified the main difficulties of data assimilation:
- large dimension problem
- time dimension critical but non linear dynamic
- observations of variable nature and quality
The purpose of this paper is to provide the theoretical basis of the algorithms llsed in operational
meteorology together with some recent developments. Most of the material presented here is
well documented in the meteorological literature (Lorenc, 1986; Ghil and Manalotte-Rizzoli,
1991; Daley, 1991) or in other fields under the generic name of inverse problems (Tarantola,
1987).
NATO ASI Series. Vol. I 45
Radiation and Water in the Climate System:
Remote Measurements
Edited by Ehrhard Raschke
Q Springer- Verlag Berlin Heidelberg 1996
Introduction to Numerical Weather
Prediction Data Assimilation
Phillipe Courtier
ECMWF, Shinfield Park
Reading, Berkshire RG2 9AX
UK
4.1 The Problem
In numerical weather prediction data assimilation consists of the process which estimates the
initial conditions of the forecast using all the available information. A description of the current
observing system can be found in McGrath (1993); around 105 elementary pieces of information
are currently used over 24 hours by the operational data assimilation system.
The current operational ECMWF model (Simmons, 1991) has a horizontal resolution of 90 km
and covers the whole globe. On the vertical the atmosphere is sampled with 31 levels from the
surface up to 10 hPa. The number of degrees of freedom of the model is then of the order of
10 7 .
Over 24 hours, the estimation problem is clearly underdetermined. The time dimension is
thus a critical element of any data assimilation system: it is essential to carry forward in time
information from past observations using the forecast model since it is the best information
propagator available.
We have identified the main difficulties of data assimilation:
- large dimension problem
- time dimension critical but non linear dynamic
- observations of variable nature and quality
The purpose of this paper is to provide the theoretical basis of the algorithms llsed in operational
meteorology together with some recent developments. Most of the material presented here is
well documented in the meteorological literature (Lorenc, 1986; Ghil and Manalotte-Rizzoli,
1991; Daley, 1991) or in other fields under the generic name of inverse problems (Tarantola,
1987).
NATO ASI Series. Vol. I 45
Radiation and Water in the Climate System:
Remote Measurements
Edited by Ehrhard Raschke
Q Springer- Verlag Berlin Heidelberg 1996