1998

The NCAR Climate System Model, Version One (CSM-1)

B. A. Boville & P. R. Gent

Observation and Context

Coupled atmosphere-ocean general circulation models are essential for studying natural climate variability and long-term environmental changes. Historically, these coupled systems have suffered from significant “climate drift,” where systematic modeling errors cause simulated surface temperatures to unrealistically warm or cool over time. To hide these errors, older modeling systems commonly applied artificial “flux corrections” — constant additive numbers that mechanically adjust heat, momentum, and freshwater balances at the ocean interface. However, these corrections are non-physical, usually cause global energy imbalances, and introduce artificial heat transports that do not mimic the real Earth. To establish true scientific validity, climate models need to maintain stable surface conditions based strictly on real interfacial physics.

Hypothesis

If a comprehensive climate system model is built using state-of-the-art parameterizations that align the heat transports of its independent atmospheric and oceanic components, and if it is initialized using a multi-stage sequential spin-up procedure to align initial states, then the fully coupled model can be successfully integrated for centuries without the use of artificial surface flux corrections while maintaining stable surface temperatures.

Experiment and Methodology

The researchers engineered the NCAR Climate System Model, version one (CSM-1), linking four independent components through a centralized driver called the “flux coupler”:

  • Atmosphere: The Community Climate Model, version three (CCM3), configured at a spectral T42 resolution with 18 vertical layers and updated convection and boundary layer math.
  • Ocean: The NCAR CSM Ocean Model (NCOM), utilizing a variable global grid with 45 depth levels, specialized K-profile vertical mixing, and the Gent-McWilliams eddy parameterization.
  • Land: A land surface biophysics model (LSM1) tracking soil hydrology and vegetation types across four sub-grid subdivisions.
  • Sea-Ice: A dynamic-thermodynamic sea-ice model utilizing a cavitating fluid rheology and a three-layer thermodynamic solver.

Crucially, no flux corrections in momentum, heat, or freshwater were permitted. Interfacial values were calculated within the flux coupler, and a simple precipitation scaling scheme was used strictly to globalize freshwater conservation in the temporary absence of a river routing model.

To launch the experiment without destabilizing shocks, the team designed a detailed sequence of spin-up phases:

  1. Independent multi-year runs of CCM3 and NCOM were completed using separate observational datasets to reach baseline states.
  2. NCOM and the sea-ice model were run together for 25 years driven by archived CCM3 atmospheric data, operating in an accelerated tracer mode that provided an effective deep-ocean spin-up time of 250 years.
  3. The model’s ice-albedo feedback was activated for an additional 25 accelerated years (bringing the effective deep-sea tracer spin-up to 500 years).
  4. The acceleration was deactivated to let the ocean adjust in synchronous mode under standard daily fluxes for 10 years.
  5. Finally, the completely coupled CSM-1 system was launched into a continuous, uncorrected 300-year control simulation.

Results and Data

  • Surface Temperature Stability: The global annual mean surface temperature underwent a small initial adjustment of about 0.7 Kelvin during the first decade but remained exceptionally stable afterward. Least-squares tracking for years 11–299 showed a virtually non-existent regional or global surface temperature trend of just 0.03 Kelvin per century.
  • Ocean Heat Transport Matching: A primary driver of this stability was that the uncoupled heat transport produced by NCOM closely matched the heat transport implied by the uncoupled CCM3 model. When coupled, the final simulated ocean heat transport matched atmospheric data almost perfectly.
  • Component Performance: Sea-surface temperatures stayed within 1 Kelvin of real-world climatology over most of the globe, though marine stratus regions showed a 2–3 Kelvin warm bias due to atmospheric cloud shortcomings. Arctic sea ice was roughly 15% too extensive and accumulated excessive thicknesses locally because river freshwater inflows were missing.
  • Deep Ocean Drift: Although surface fields remained perfectly stable, the deep ocean experienced strong ongoing trends because it possesses the longest physical timescales. Total ocean heat content dropped roughly 15% before stabilizing, while deep-sea salinity at a depth of 4 kilometers rose monotonically throughout the run from 34.70 to nearly 35.00 parts per thousand, creating an upper ocean that was too fresh and a deep ocean that was too saline compared to reality.

Conclusion and Climate Impact

The hypothesis was supported. The experiment successfully demonstrated that a global climate model can eliminate centuries-long surface temperature drift without depending on artificial, non-physical flux corrections. This stability proves that when component models possess mathematically compatible heat transports and are initialized using precise spin-up steps, the surface climate naturally settles into a stable physical equilibrium.

The climate impact of CSM-1 is highly significant. By proving that uncorrected coupled models can achieve stable global surface simulations, this research removes a massive source of non-physical manipulation from global climate modeling. However, the remaining salinity and temperature drifts in the deep ocean confirm that securing true centennial surface stability is merely a stepping stone. To simulate stable climates on millennial timescales, models must integrate missing components like true river runoff networks and more advanced sea-ice dynamics to completely stop deep-ocean deterioration.

Citation

Boville, B. A., and Gent, P. R. (1998). The NCAR Climate System Model, Version One. Journal of Climate, 11(6), 1115-1134.