Decadal Prediction with GEOS-5
The climate extremes experienced within the last couple of decades — such as the current drought over much of the central U.S., the prolonged drought in the American Southwest, the 2003 European heat wave, the 2010 Russian heat wave, and the flooding in Pakistan — and the need to develop adaptation strategies to climate change have served to increase interest in the prediction of climate change over the next 10–30 years, otherwise referred to as the "decadal" time scale (e.g., Meehl et al. 2009).
The interest in this time scale is also partly driven by the science, since it is a natural extension of both the work done by the Seasonal-to Interannual (SI) prediction community (extending to longer time scales) and that done by the climate change community (extending to shorter timescales). It is now generally accepted that the decadal problem requires information on both initial and boundary conditions. This contrasts with the SI prediction problem where the initial condition is a key, and the centennial time-scale problem where changing boundary conditions dominate the projections. The predictability of climate signals on these timescales is not yet understood. To help address decadal predictability and prediction, the Coupled Model Intercomparison Project, Phase 5 (CMIP5) includes the protocol for experiments on these timescales as well as for long-term change.
We have used the GEOS-5 AOGCM to contribute decadal forecasts to CMIP5 (Ham et al., 2012). The three-member ensemble decadal predictions, initialized in December of each year from 1960 to 2005, have now been published through the Earth System Grid node in the NCCS. The forecast results have been evaluated against the ocean reanalysis used to initialize the forecasts and also against uninitialized forecasts from a free-running climate simulation in the latter half of the 20th century (C20C).
The GEOS-5 AOGCM configuration is:
AGCM: 2.5° longitude x 2° latitude grid x 72 vertical levels up to 0.01hPa.
OGCM: 1°, with a meridional equatorial refinement to 1/3°, and 50 vertical levels.
The model includes a river runoff routing scheme, and an aerosol model based on the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model using emissions prescribed by the CMIP5 protocol. Only those volcanic aerosols from continually outgassing volcanoes have been included.
The initial conditions for the decadal forecasts/hindcasts are obtained from the GMAO's ocean reanalysis (ORA), which was generated using a version of the GEOS ocean and sea-ice data assimilation system that is a multi-variate Ensemble Optimal Interpolation (EnOI) analysis scheme , while the atmosphere is constrained by MERRA from 1979 to 2005 and by a related atmospheric analysis prior to 1979.
The results show that initialized and uninitialized GEOS-5 forecasts have some skill in forecasting SSTs and upper ocean temperature, suggesting the importance of both external forcing and initialization in decadal forecasts. Forecasts initialized with observations do show improved skill over many areas of the globe when compared to uninitialized forecasts, except in parts of the Pacific and Indian Oceans. The increase in skill is maintained for almost a decade over the subtropical and mid-latitude Atlantic. Although the increase in skill decreases as forecast lead times increase (see Figure 1), some benefit from initialization remains for up to a decade. On the other hand, the initialization reduces the skill in predicting the warming trend over some regions outside the Atlantic, presumably from dynamical adjustments to imbalances in the initial conditions.
Figure 1: The top left panel shows the anomaly correlation of the 3-year moving averaged SST from the C20C simulation with the GMAO ocean reanalysis. The panels below show the difference of the C20C correlation from that of the initialized predictions at increasing lead times. The upper right hand panel shows the Mean Square Error (MSE) of the uninitialized C20C simulation. The panels below show the Mean Squared Skill Score (MSSS) at increasing lead times. Orange and red colors denote favorable model performance.
While the skill measured by Mean Squared Skill Score (MSSS) shows 50% improvement up to a 10-year lead forecast over the subtropical and mid-latitude Atlantic, prediction skill is relatively low in the subpolar gyre, due in part to the fact that the spatial pattern of the dominant simulated decadal mode in upper ocean heat content over this region appears to be unrealistic. An analysis of the large-scale temperature budget shows that this is the result of a model bias, implying that realistic simulation of the climatological fields is crucial for skillful decadal forecasts.
The degradation in forecast skill with lead time in this region is clearer in the anomaly correlation of the heat content in the upper 500m of the ocean when the trend related to global warming is removed (see Figure 2). The degradation with forecast lead can be related to biases in the transports in the subpolar region. Thus, model biases need to be reduced to improve the predictability of climate variations in this region.
Figure 2: The correlation skill of the de-trended annual mean HC500 anomaly in the C20C simulation, and increasing leads for the initialized forecasts up to a 5-year lead forecast.
Many studies have pointed out the role of the Atlantic Meridional Overturning Circulation (AMOC) as a source of decadal predictability in the Atlantic. Figure 3 shows the time series of the anomalous AMOC index (defined here as the maximum of the zonally-integrated annual mean overturning streamfunction averaged over 43–45°N) using the ORA and prediction output.
Figure 3. The time series of Atlantic Meridional Overturning Circulation (AMOC) index in the GMAO ORA (black) and decadal forecast. The AMOC index is defined as the maximum of the zonally-integrated annual mean overturning streamfunction averaged over 43-45°N.
In the ORA, there is a decreasing trend in the AMOC at this latitude. The decadal variations to the mid-1990s mimic closely those in other studies (see Ham et al. (2012) for details). The overall variation, with a minimum in the mid-1970s and a decrease from the mid-1990s, is captured at all lead times up to 5 years, though the details of the shorter timescale variations change with lead time. For example, the small peak in the early 1970s is missed in 1-to-3-year lead forecasts, but is captured in the 4- and 5-year lead, potentially because of better initialization. Similarly, the peak in the mid-1990s is not captured well, with only a small increase in forecast transport at 1- and 2-year lead times, but a larger increase for forecasts at 3- to-5-year lead times, albeit at a later time and having first undergone a decrease in transport from 1990 values. The analyzed very low transport in 2006 is not captured in any of the forecasts, although there is a small decrease in transport in the forecasts with a 5-year lead time. The correlation of the forecast time series with the ORA transport is over 0.6 for leads up to 5 years. In comparison, the correlation skill of a persistent forecast is above 0.6 only up to 2-year lead times, implying that the predictability of AMOC is systematically higher in the dynamical forecast system.
Ham et al. (2012) undertakes further analysis of the forecasts with the global warming influence removed, particularly the details of the upper ocean heat content in the subpolar North Atlantic. The analysis shows that initializing forecasts with observations generally has a positive impact on forecast skill, but any such improvements can be limited by model biases.
The forecasts are available at the PCMDI's CMIP5 Earth System Grid Gateway. Due to issues with initializing the decadal suite using several ocean analysis streams, we have not published the forecasts from 1981 to 1984.
References:
Ham, Y.-G., M.M. Rienecker, M.J. Suarez, Y. Vikhliaev, B. Zhao, J. Marshak, G. Vernieres, and S.D. Schubert, 2012: Decadal prediction skill in the GEOS-5 forecast system. Climate Dynamics (submitted).
Meehl, G.A. et al, 2009: Decadal prediction: can it be skillful? Bull. Amer. Meteor. Soc., 90, 1467–1485.doi:10.1175/2009BAMS2607.1
