CAPITULO I 5 MARCO TEÓRICO
5.1 La actividad artística
5.1.8 Evaluación del aprendizaje.
The LRF Group (Bernd Becker, Anca Brookshaw, Andrew Colman, Mike Davey, Margaret Gordon, Richard Graham, Matt Huddleston, Sarah Ineson, Bruce Ingleby, Malcolm MacVean (at ECMWF) and Peter McLean)
Met Office, Fitzroy Road, Exeter EX1 3PB, UK
The Met Office has been active in long range forecasting for many years, with a particular interest in the Atlantic sector: the longest series of seasonal rainfall forecasts that we still issue issued are those for the Sahel (since 1986) and north east Brazil (since 1987). Originally statistical methods (principally linear regression and discriminant analysis) were used to make the long-range forecasts, by relating precursor SST anomaly patterns to seasonal regional climate variations. From the outset these forecasts incorporated probability information about several forecast categories. The continued evolution of dynamical methods, based on general circulation models, led to the introduction of ensemble forecast systems and a wider range of LRF products. This article provides an overview of recent and current LRF activities, emphasising the seasonal aspects. Forecast products and further information can be found at
http://www.metoffice.com/research/seasonal and http://www.metoffice.com/monoutlook .
GCM ensemble forecast systems and products
The Met Office global seasonal forecasting system (known as GloSea) is based on a coupled GCM that is a variation of the Hadley Centre HadCM3 climate model. The GloSea CGCM has basically the same atmospheric component (HadAM3) as HadCM3, but the ocean component was enhanced to have higher horizontal (1/3 degree meridional grid near the equator) and vertical (40 levels) resolution, and a surface tiling scheme was included among other modifications. With the aim of an operational multi-model, the GloSea system infrastructure was designed in parallel with ECMWF system2, and the CGCM is run on the ECMWF computing facility. Ocean initial conditions in the GloSea system are produced at the Met Office and transferred as required to ECMWF. A 5-ensemble of ocean analyses is generated by using sampled perturbations to the surface wind stress. (Recently this component has been modified by reducing the perturbations and tightening constraints to sub-surface temperature observations, in order to reduce the analysis ensemble spread which was judged to be too large and which influenced the ensemble spread in the first few months of the forecasts.)
A 40-member forecast ensemble is run in the middle of each month, with a start date of the first day of the month, and a range of six months. Precipitation and surface temperature data from the ensemble are calibrated by calculating anomalies relative to the model forecast climatology for the same start time-of-year and forecast range, based on a 15-year 15-member hindcast set. Two- and three- category 3-month-average global gridded forecast probability products are derived from the calibrated ensemble for the Met Office forecast website. The hindcast set is also used to provide skill information about each of the products. With regard to the tropical Pacific, forecast plumes (trajectories for each of the ensemble members) are produced for the standard Nino3, Nino3.4, Nino4 regions. Multi-model products are still under development and not yet openly available, but the intention is to introduce these in the near future.
Using the GloSea system, 9-member hindcasts over a 43 year period were produced quarterly as part of the EC DEMETER project, along with other European partners. Extensive information about the performance of the DEMETER models, individually and in multi-model form, is
available on http://www.ecmwf.int/research/demeter . An overview can be found in Palmer et al.
(2004).
Comparison of 2-tier and coupled ocean-atmosphere systems
Prior to implementation of the coupled GCM system in 2003, a 2-tier system with essentially the same atmospheric model (HadAM3) with statistically predicted SST was the main dynamical LRF model. As extensive hindcasts with the 2-tier system were also produced in DEMETER, we
have been able to compare the performance of the coupled and uncoupled systems in detail (Graham et al., 2004). The coupled system has the advantage of ocean-atmosphere interactions, but the disadvantage of larger drifts and biases, particularly in ocean regions. Overall the forecast performance of the coupled system is better, largely as a result of improved representation of ENSO events and their teleconnections.
Notable improvements in the tropical Atlantic and Indian ocean sectors are associated with lagged ENSO effects. In the equatorial Atlantic coupled model SST forecast skill is highest in NDJ and FMA, when correlation with observations are greater than 0.4 over most of the area with patches greater than 0.6 . However, in MJJ correlations are near zero in the equatorial Atlantic. In the north tropical Atlantic, predictability is best in NDJ and MJJ. In the extratropical North Atlantic CGCM performance is best in the North-East sector, where in NDJ we find correlations greater than 0.6 in the central north Atlantic, in a region historically linked with European climate variability in empirical studies. Relative to the 2-tier system, the CGCM has substantially improved reliability for spring season 2m temperatures over Europe. Case studies suggest that this improvement occurs because the CGCM can evolve SST anomalies realistically in some situations, such as 1989 when the NAO was strongly positive (Graham et al. 2004), but further investigation is needed before drawing firm conclusions.
For specific tropical Atlantic region rainy seasons, the CGCM has significant skill for the NE Brazil and Guinea areas at both 1-month and 2-month leads, with predictions for the latter region showing substantial benefits over the 2-tier system. However, scores for the Sahel region at 1-month and 2-month leads appear close to the no-skill levels with both systems. Somewhat better skill for the Sahel appears available at 3-month lead with the GloSea model. For these regions, for which specific statistically-based forecasts have been issued since the 1980s, dynamical forecast information is combined with the statistical predictions to improve the forecast quality. (See the forecast website for details.)
The global-mean annual-mean temperature forecast
Led by Chris Folland, in recent years the Met Office Climate Analysis group has issued a statistical forecast in December of global mean temperature for the year ahead, using predictors such as trends and foreseen atmospheric composition and ENSO state. Trials of dynamical year-ahead forecasts were made in 2003 with the GloSea ensemble system. Over a test period of 15 years the performance of the dynamical model was closely comparable to that of the statistical approach, and in December 2003 a real-time year-ahead forecast was produced.
The European summer forecast
Following evidence for a connection between Jan-Feb north Atlantic SST anomalies and European summer seasonal climate, gridded tercile probability forecasts of July-Aug temperatures in Europe have been issued since 1999, generated using simple statistical methods (see the forecast website: notably, the statistical forecast for summer 2003 strongly favoured the upper tercile). However, the mechanism for this connection is not yet well understood, and merits further investigation using dynamical systems.
The ENACT project: ocean data and analyses
The EC ENACT project ( http://www.lodyc/jussieu.fr/ENACT/ ) is now well underway, with the
aim of producing global ocean analyses from various ocean models and data assimilation systems, in the context of improving seasonal forecasts through improvements in ocean initial conditions. The ocean analyses have two main streams: 1987 to 2001 (when satellite altimetry data are available), and 1962 to 2001. An important preliminary outcome of this project is that several relevant datasets have been prepared with regard to ocean in situ data, altimeter-based sea level data, and surface flux data adapted from ERA40. The Met Office has been responsible for preparing the in situ dataset, covering the period 1962 to 2001. Temperature and salinity observations from several sources (including WOD01) have been collected, uniformly quality controlled, and written in a standard NetCDF format. It is expected that this dataset will be made generally available for research purposes.
References
Graham, R., Gordon, M., McLean, P., Ineson, S., Huddleston, M., Davey, M., Brookshaw, A. and Barnes, R., 2004: A performance comparison of coupled and uncoupled versions of the Met Office seasonal prediction General Circulation Model. Tellus. Submitted.
Palmer, T. N., Alessandri, A., Andersen, U., Cantelaube, P., Davey, M., Delecluse, P., Deque, M., Diez, E., Doblas-Reyes, F., Feddersen, H., Graham, R., Gualdi, S., Gueremy, J.-F., Hagedorn, R., Hoshen, M., Keenlyside, N., Latif, M., Lazar, A., Maisonnave, E., Marletto, V., Morse, A., Orfila, B., Rohel, P., Terres, J.-M., and Thomson, M., 2004: Development of a European multi-model ensemble system for seasonal to interannual prediction (DEMETER). Bull. Amer. Met. Soc. To appear.