Climate model downscaling

Climate model downscaling converts the coarse output of a global climate model, typically grid cells tens to hundreds of kilometers across, into finer-resolution projections useful for a specific region, watershed or site, where local terrain, coastlines and land cover matter for decisions like infrastructure design or flood mapping. Global models simply cannot resolve features smaller than their grid spacing.

Two main approaches exist: dynamical downscaling, which nests a higher-resolution regional climate model, using the same kind of physics as numerical weather prediction, inside the global model's boundary conditions, and statistical downscaling, which learns a statistical relationship between historical coarse-model and fine-scale observed or reanalysis data variables and applies that relationship to future projections. Dynamical downscaling is more physically grounded but computationally expensive; statistical downscaling is fast but assumes the historical relationship between local and large-scale climate holds unchanged in a warmer future.

Downscaled projections, typically run across multiple shared socioeconomic pathways emissions scenarios and multiple models to sample uncertainty, feed physical climate risk assessments, infrastructure design standards and local adaptation planning. The main pitfalls are presenting a single downscaled run as a definitive local forecast rather than one plausible outcome among an ensemble-style range, and applying statistical downscaling methods trained on a climate regime very different from the projected future without acknowledging that assumption.