Questions: Regional Climate Downscaling and Projections

Question 1

A regional climate model (RCM) is driven by a GCM that has a systematic bias — it places the mid-latitude jet stream 200 km too far north. What does the downscaled output show for the affected region?

• A. The RCM corrects the jet stream bias because its higher resolution allows it to resolve local atmospheric dynamics more accurately
• B. The RCM inherits the jet stream bias from the GCM because it uses GCM output as boundary conditions; fine-scale resolution cannot fix a large-scale circulation error
• C. Bias correction applied to the RCM output automatically removes the jet stream displacement
• D. The RCM replaces GCM circulation patterns with observed reanalysis data at the domain boundaries

Answer: (B)

Dynamical downscaling embeds a high-resolution RCM inside the GCM — the GCM provides the boundary conditions (temperature, winds, humidity at the edges of the RCM domain). If those boundaries encode an incorrect jet stream position, the RCM's interior solution is constrained to that error. Fine-scale physics within the domain cannot override the large-scale circulation imposed at the boundaries. This is why downscaled projections always inherit the biases of their parent GCM — it is a fundamental architectural limitation, not a fixable technical problem.

Question 2

What is the 'stationarity assumption' in statistical downscaling, and why is it a concern for climate projections?

• A. The assumption that GCM grid spacing remains fixed throughout the projection period
• B. The assumption that the historical relationship between large-scale circulation and local weather will hold in future climates, which may fail as the climate shifts into states without historical precedent
• C. The assumption that dynamical and statistical downscaling methods produce equivalent results when applied to the same GCM
• D. The assumption that regional temperatures remain stationary (constant mean) during the reference period used for calibration

Answer: (B)

Statistical downscaling calibrates an empirical model — e.g., 'when the 500 hPa geopotential height pattern looks like X, local rainfall is Y' — using historical observations. Stationarity is the assumption that this empirical relationship will still hold in the future. But future climates may include temperature regimes, atmospheric moisture levels, or circulation patterns with no historical analog. When that happens, the statistical model is being extrapolated beyond its training domain, and its predictions may be unreliable. Dynamical downscaling doesn't make this assumption because it uses physical equations, though it has its own structural assumptions.

Question 3

A high-resolution regional climate model can correct systematic errors in the driving GCM's large-scale atmospheric circulation patterns.

• T. True
• F. False

Answer: False

This is a critical misconception about the limits of downscaling. The RCM's lateral boundary conditions are supplied by the GCM — the large-scale circulation at the domain boundaries is prescribed, not simulated by the RCM. If the GCM places the jet stream in the wrong position or misrepresents the strength of blocking events, the RCM's interior solution is anchored to those errors. Adding more resolution within the domain adds local detail (orographic effects, coastal gradients) but cannot override the inherited large-scale circulation. The phrase is: 'garbage in, garbage out' for large-scale drivers.

Question 4

Using an ensemble of multiple GCMs and downscaling methods is intended to eliminate uncertainty in regional climate projections.

• T. True
• F. False

Answer: False

Ensemble approaches characterize and bracket uncertainty — they do not eliminate it. Uncertainty in regional projections comes from multiple sources: the emission scenario, structural differences between GCMs, the choice of downscaling method, and bias correction decisions. Running an ensemble of methods reveals the spread of plausible outcomes, giving decision-makers a sense of the range rather than a false single number. A wide ensemble spread means the uncertainty is genuinely large. Calling this 'elimination' of uncertainty would misrepresent what ensembles provide and could lead to overconfident decisions.

Question 5

Explain why downscaled climate projections necessarily inherit the biases of the driving GCM, and what this implies for how regional projections should be interpreted.

Model answer: Downscaling — whether dynamical or statistical — refines GCM output but cannot generate information the GCM doesn't contain. Dynamical downscaling uses GCM output as boundary conditions, so large-scale circulation errors are imposed on the regional model from outside. Statistical downscaling builds empirical relationships with GCM predictors, so any systematic GCM bias in those predictors propagates into local projections. Bias correction can partially address this, but it adds its own assumptions. The implication is that regional projections should always be presented as conditional on the parent GCM — reported as a range across multiple GCMs rather than as a single authoritative number, and interpreted as indicating the direction and plausible magnitude of change rather than a precise forecast.

This inheritance of bias is not a failure of downscaling but a fundamental property of its architecture. Downscaling adds resolution and local physical detail; it does not independently constrain the large-scale climate. Users of downscaled products (water managers, agricultural planners) need to understand this so they make decisions robust to the GCM spread, not decisions that would be invalidated if a different parent GCM had been chosen.