Rapid changes in high‐latitude hydroclimate have important implications for human societies and environment. Previous studies of different regions have indicated better agreement between climate model results and observation data for the thermodynamic variable of surface air temperature ( T ) than for the water variables of precipitation ( P ), evapotranspiration ( ET ), and runoff ( R ). Here we compare climate model output with observations for 64 Nordic and Arctic hydrological basins of different sizes, and for the whole region combined. We find an unexpectedly high agreement between models and observations for R , about as high as the model‐observation agreement for T and distinctly higher than that for P or ET . Model‐observation agreement for R and T is also consistently higher on the whole‐region scale than individual basin scales. In contrast, model‐observation agreement for P and ET is overall lower, and for some error measures also lower for the whole region than for individual basins of various scales. Region‐specific soil freeze–thaw bias of climate models can at least partly explain the low model‐observation agreement for P and ET , while leaving modeled R relatively unaffected. Thereby, model projections for this region may be similarly reliable and directly useful for large‐scale average conditions of R as of T . Plain Language Abstract

Climate is changing more rapidly in the Arctic than in many other places on Earth. To understand and project how the Arctic climate is changing, we use computer models that simulate the climate system. Often, such models are reported to perform better for simulations of temperature than for simulations of water, such as rain, snow, or river flow. They are often also reported to be better at large scales than at small scales. In our research, we have studied how well climate model results agree with meteorological observations, by studying data for 64 different rivers in the Nordic and Arctic regions. Contrary to what we expected, we found that models were about as good in simulating river flow as they were in simulating temperature and clearly better than they were at simulating rain and snow. For simulations of river flow and temperature, models were better at larger scales than smaller scales. Our results mean that it may be possible to use these models for understanding and projecting river flows with a similar reliability as for temperature for this region, at least on large scales.