Ensemble forecasting and data assimilation: two problems with the same solution? Chapter uri icon


  • © Cambridge University Press 2006 and 2010. Introduction Until 1991, operational numerical weather prediction (NWP) centres used to run a single computer forecast started from initial conditions given by the analysis, which is the best available estimate of the state of the atmosphere at the initial time. In December 1992, both the US National Centers for Environmental Prediction (NCEP) and ECMWF started running ensembles of forecasts from slightly perturbed initial conditions (Molteni and Palmer, 1993; Toth and Kalnay, 1993; Tracton and Kalnay, 1993; Toth and Kalnay, 1997; Buizza et al., 1998; Buizza, this volume). Ensemble forecasting provides human forecasters with a range of possible solutions, whose average is generally more accurate than the single deterministic forecast (e.g. Figures 7.3 and 7.4), and whose spread gives information about the forecast errors. It also provides a quantitative basis for probabilistic forecasting. Schematic Figure 7.1 shows the essential components of an ensemble: a control forecast started from the analysis, two additional forecasts started from two perturbations to the analysis (in this example the same perturbation is added and subtracted from the analysis so that the ensemble mean perturbation is zero), the ensemble average, and the ‘truth’, or forecast verification, which becomes available later. The first schematic shows an example of a ‘good ensemble’ in which ‘truth’ looks like a member of the ensemble. In this case, the ensemble average is closer to the truth than the control due to non-linear filtering of errors, and the ensemble spread is related to the forecast error.

author list (cited authors)

  • Kalnay, E., Hunt, B., Ott, E., & Szunyogh, I.

citation count

  • 14

Book Title

  • Predictability of Weather and Climate

publication date

  • January 2006