Data assimilation over the last millennium using a large ensemble of simulations Hugues Goosse Université catholique de Louvain, Belgium M.E. Mann, H. Renssen, A. Timmermann National Gallery, London
Description of LOVECLIM LOVECLIM (3D) AGISM (ice sheets) ECBilt (atmosphere) CLIO (sea ice-ocean) VECODE (terr. biosphere) LOCH (oceanic carbon cycle) ECBilt (Opsteegh et al., 1998) Quasi-geostrophic atmospheric model (prescribed cloudiness; T21, L3). CLIO (Goosse and Fichefet, 1999) Ocean general circulation model coupled to a thermodynamic-dynamic sea ice model (3 x 3, L20). VECODE (Brovkin et al., 2002) Reduced-form model of the vegetation dynamics and of the terrestrial carbon cycle (same resolution as ECBilt). LOCH (Mouchet and François, 1996) Comprehensive oceanic carbon cycle model (same resolution as CLIO). AGISM (Huybrechts, 2002) Thermomechanical model of the ice sheet flow + visco-elastic bedrock model + model of the mass balance at the ice-atmosphere and ice-ocean interfaces (10 km x 10 km, L31).
Ensemble of simulations over the last 1000 years Northern Hemisphere annual mean temperature red: ensemble mean grey:2 standard deviation of the ensemble Temperature (ºC) Summer temperature in Fennoscandia red: ensemble mean grey:2 standard deviation of the ensemble Temperature (ºC) The time series are grouped in 25-year averages Time (yr AD)
Data assimilation Goal: to combine directly model results and proxy records in order to have a reconstruction of past climate changes that is consistent with proxy data, model physics and the forcing. A few techniques have been proposed: Jones and Widmann (2003) and van der Schrier and Barkmeijer (2005) constrain model results to remain close to a reconstruction of the observed atmopsheric circulation. Assimilation of a pattern of winter-mean sea level pressure (mb) corresponding to the period 1790-1820 (van der Schrier and Barkmeijer 2005)
Using paleoclimate proxy-data to select the best realisation in an ensemble Simulation of the climate of the last 1000 years : selecting among a relatively large ensemble of simulations the one that is the closest to the observed climate. The experiment selected is the one that minimise a cost function CF for a particular period : Where n is the number of reconstructions used in the model/data comparison. Fobs is the reconstruction of a variable F, while Fmod is the simulated value of the corresponding variable. wi is a weight factor. Goosse et al. 2006
Using paleoclimate proxy-data to select the best realisation in an ensemble Example using 5 ensemble members and two constraints (observations, in red). The best member selected is displayed in bold while the other ones are dashed Temperature Temperature Time Time Goosse et al. 2006
Simulation over the last 150 years constrained by HadCRUT3 dataset The region northward of 30°N has been divided in 7 boxes. 96 simulations are performed in the ensemble Annual mean temperature northward of 30°N Temperature Anomaly Reconstruction Uncertainty range (2s) HadCRUT3 Time Musée du Louvre, Paris Goosse et al., in preparation
Simulation over the last 150 years constrained by HadCRUT3 dataset Annual mean temperature in the Arctic Temperature Anomaly For all the regions, the reconstruction is close to the observations and the uncertainty range is small. Reconstructions Uncertainty range Annual mean temperature in Europe HadCRUT3 Temperature Anomaly Time Musée du Louvre, Paris Goosse et al., in preparation
Reconstructing temperature using 22 long proxy records The model is constrained to follow 22 long proxy records, located on continents, mainly northward of 30°N Location of the proxies on model grid Mann et al., in preparation
Reconstructing temperature using 22 long proxy records The model is constrained to follow 22 long proxy records. 96 simulations are used in the ensemble Annual mean temperature northward of 30N Model-based reconstructions Reconstruction + and – 2 s HADCRUT3 Temperature anomaly Time The red and orange lines represent two reconstructions with different initial conditions. Goosse et al., in preparation
Reconstructing temperature using 22 long proxy records Annual mean temperature in the Arctic The reconstruction is still close to the observations but the uncertainty range is larger. Model-based reconstructions Reconstruction + and – 2 s HADCRUT3 Temperature anomaly Annual mean temperature in Europe The results are better at hemispheric scale than at regional scale. Time The red and orange lines represent two reconstructions with different initial conditions.
Reconstructing temperature over the last millennium: local agreement with the proxies. Annual mean temperature in Northern Europe Temperature anomaly The agreement between the proxies and the reconstruction is good in the extratropics. Mean of the available proxy records Model-based reconstruction Annual mean temperature in Canada-Greenland Temperature anomaly Time
Reconstructing temperature over the last millennium The model is constrained to follow 22 long proxy records. 96 simulations are used in the ensemble. Annual mean temperature northward of 30N Model-based reconstructions Reconstruction + and – 2 s Mann and Jones, 2003 Temperature anomaly Time As Mann and Jones (2003) provide a reconstruction for the Northern hemisphere, it has been rescaled here.
Reconstructing temperature over the last millennium Annual mean temperature in Europe Reconstruction + and – 2 s Temperature anomaly Model-based reconstructions Luterbacher et al. (2004) Overpeck et al. (1997) Annual mean temperature in the Arctic Temperature anomaly Time A 21-year running mean has been applied to the time series while the error bars are computed from the ‘decadally’ smoothed time series..
Reconstructing temperature in the Arctic Annual mean temperature anomaly in the Arctic during two warm periods 1750-1800 1500-1550 Are those patterns realistic ? What are the mechanisms behind those patterns ? The anomalies are computed relative to the period 1000-1850.
Conclusions Galleria degli Uffizi, Firenze
Conclusions • Very good reconstructions are obtained at hemispheric and regional scales when a large number of good temperature records are available (e.g. HADCRUT3). • When using proxy data to constrain model evolution, the reconstruction follow the proxy records at regional scale. • When using 22 proxies, the skill of the reconstruction is good at hemispheric scale. For regional scales, more proxies are probably needed. • Next step: tests with more proxies (last 600 years) in order to analyse regional patterns.