Uncertainty in Projected Precipitation Change and Its Dependency on Aggregation to the River‐Basin Scale

Abstract Model projections of changes in precipitation express large uncertainties, partly owing to climate noise from internal variability dominating the greenhouse gas forced signal. In this paper we intend to give guidance on processing uncertainty information depending on application‐specific needs. In a high‐resolution initial‐condition regional climate model ensemble, we investigate the signal‐to‐noise problem for precipitation changes in major river basins in Western Europe, focusing on how it depends on different methods of spatial aggregation. This is motivated by the common need for spatially aggregated climate change information. We find that the forced climate signal is only moderately dependent on spatial aggregation. However, uncertainties (and likewise the signal‐to‐noise ratio) strongly depend on the aggregation method. Computing uncertainties at the grid scale—as typically provided in maps in climate change studies—and averaging those to larger areas, leads to large uncertainty estimates and corresponding low signal‐to‐noise ratios. This method does not account for spatiotemporal dependencies in the underlying data, and is representative for the uncertainties at a specific location. Computing changes at the grid scale, spatially averaging those per ensemble member, and computing the uncertainty in these spatial averages results in much smaller uncertainty estimates, up to a factor of 2 for 1‐day precipitation maxima in summer. This information is useful for stakeholders with an aggregated uncertainty perspective, for example insurance. Spatial aggregation of precipitation itself and computing uncertainties in basin‐scale precipitation changes—relevant for example river flooding—also yields smaller uncertainties, up to a factor of 1.5 for 1‐day summer maxima.