In the last 1000 years there have been a number of climate events that have interested scientists. This project attempted to explore these events and to match them against other records to help better inform the climate modelling process.


We’d like to understand the climate changes since 800 AD (i.e., for just over a millennium). In addition to the post-industrial era, this period includes the so-called Medieval Warm Period (~900-1300 AD) and Little Ice Age (~1300-1900AD). The anomalously warm and cold periods are probably caused by the variation of volcanic & solar activities, land use changes and perhaps the change of oceanic circulation pattern, but the contribution of each component is not well understood.

In addition, to satisfy scientific and historical curiosity, this experiment is driven by the urgent need to refine the climate predictions. The climate models have so far been evolved to simulate the recent climate. The last “millennium” can provide an extra constraint to refine the models further.

However, both the driving force and the climate reconstructions over the pre-industrial era are based on the analysis of the natural archives of climate sensitive quantities, such as the growth of trees and seashells, and the changes of chemical, biological, and isotopic compositions in lake sediments and ice core samples. These “proxy” data are truly useful only when we fully explore the enormous range of the combined uncertainty.


To overcome this difficulty, we use a grand ensemble of a faster variant of the well-established UKMO climate model, called FAMOUS. It runs about 10 times faster than the HadCM3L, which was used for the BBC experiment, because of reduced atmospheric resolution and increased ocean time-step. Therefore, the full 1600 year simulation from initial adjustment to near future projection is only as expensive as the 160 year-long BBC experiment.

Atmosphere: 3D GCM (5.0 x 7.5 degrees resolution, 11 levels).
Ocean: 3D GCM (2.5 x 3.75 degrees resolution, 20 levels).
Sea Ice: simple thermodynamic model with drift.
Land: vegetation and hydrological cycle.

In addition to perturbations for internal physics parameters of the model and initial condition, this experiment requires a large number of forcing perturbations to deal with the large uncertainty in the historical forcings.

To reduce the CPU time further, parallelization by domain decomposition was tested. For the initial release, however, we decided not to distribute the parallel version because we need more preparation time for packaging.

Warning! Warning! Warning!

The Millennium/Famous experiment uses a radically new way to run perturbed atmosphere-ocean coupled models without dynamic flux adjustments. Therefore, we expect many models to run away quickly from realistic climatic conditions, especially when we perturb the internal parameters of the model. The main test of such models is performed for the “spin-up” period labelled as 599-799 AD. For this period, we don’t use time-evolving external forcing data. We simply keep them constant to see how each model adjusts to its own new equilibrium. However, since some of our perturbations are fairly strong, many models will simply run away!

Also, we use several extreme forcing scenarios. For example, the largest volcanic eruptions in the last Millennium are estimated to be several times larger than the Pinatubo eruption in 1991. Moreover, to cover the full range of uncertainty in the historical volcanic eruption data, we even try the case with 3 times the best estimate of volcanic forcing.

Once we find a variety of not-outrageous models, we will select these to run more realistic simulations from 800 AD towards the near future. Such runs still include the setup with the extreme scenarios, and some of the models that survived the spin-up period will probably go wild. We also noticed that the current Famous model is perhaps somewhat less stable than other models we have been using. In any event, we compare model outputs with observational data and evaluate the goodness of each model run. The goodness measure will be used to illustrate the range of plausible futures.


The Millennium experiment has been primarily funded by European Commission through Framework 6 project European Climate of the last Millennium. While the modelling experiment is performed, the Millennium partners have been working to quantify and reduce uncertainty in the climate “proxy” information by collecting and analysing more climate records. For technical refinements of the model, the Oxford team has been working with UK Met Office and Sony Computer Science Laboratory Paris. The EC funding formally finished at the end of December 2010, but Stockholm University paleoclimate group continues to offer a full-time contribution for the analysis of the model outputs until summer 2011. Also involved in further analysis is Quaternary Research Cluster in Newcastle University, which now includes a former member of the core team.


This experiment was designed in AOPP, Oxford through discussion in the paleo-climate community in Europe and beyond. The Millennium/Famous model was released on March 24, 2010, for intensive testing and prerequisite runs. The Millennium/Famous experiment was released on June 10, 2010 with grand ensemble of 200 year work units, thanks to continuous and patient support by beta testers. In March 2011, the first phase of the experiment ended, with successful completion of over 40,000,000 years of model simulations in total.


  • Start: 2010
  • Model used: FAMOUS
  • Lead scientist: Hiro Yamazaki
  • Finish: 2011