Climate Model Resolution

Model resolution is the spatial and temporal scale at which a climate model represents the atmosphere, ocean and land. Because model variables usually describe averages over grid cells and time steps, phenomena occurring at smaller scales are not represented explicitly and must be approximated through parameterisation.

Why local extremes are difficult to resolve

Localised, short-duration extremes often arise from processes operating below or near the model grid scale. Averaging across a large grid cell smooths sharp rainfall, temperature and wind peaks, while coarse time steps can miss their rapid development and decay.

• Small convective storms and cloud microphysics may be sub-grid processes, so their effects are estimated rather than calculated explicitly.
• Coarse grids simplify topography, coastlines and land cover, weakening local influences such as orographic uplift, sea breezes and urban heating.
• Extreme events depend on nonlinear interactions among moisture, instability and circulation; small errors in these conditions can produce large errors in event intensity or location.

Computational constraint

Increasing resolution sharply raises the number of grid cells and generally requires shorter time steps for numerical stability. This increases computational cost, data storage and processing requirements, particularly for long simulations and multi-member ensembles.

• Climate projections must cover decades and often use ensembles to sample internal variability and model uncertainty, limiting the resolution affordable for each run.
• Higher resolution improves the representation of many processes, but it does not automatically remove errors in physical parameterisations or initial and boundary conditions.

Ways to obtain local information

Dynamical downscaling nests a regional climate model within a global model, while statistical downscaling relates large-scale climate variables to observed local conditions. Convection-permitting models can represent some storms more explicitly, but local projections remain dependent on model physics, observations and the driving large-scale simulation.

• Downscaling can add spatial detail, but it cannot fully correct biases inherited from the parent model.
• Reliable assessment therefore combines higher-resolution modelling, ensembles, observations and explicit communication of uncertainty.