The effect of geographic sampling on extreme precipitation: from models to observations and back again
share
In light of the significant uncertainties present in global climate models' characterization of precipitation extremes, it is important to properly use observational data sets to determine whether a particular climate model is suitable for simulating extremes. In this paper, we identify two problems with traditional approaches for comparing global climate models and observational data products with respect to extremes: first, daily gridded products are a suboptimal data source to use for this comparison, and second, failing to account for the geographic locations of weather station data can paint a misleading picture with respect to model performance. To demonstrate these problems, we utilize in situ measurements of daily precipitation along with a spatial statistical extreme value analysis to evaluate and compare model performance with respect to extreme climatology. As an illustration, we use model output from five early submissions to the HighResMIP subproject of the CMIP6 experiment (Haarsma et al., 2016), comparing integrated metrics of an "extreme" bias and Taylor diagrams. Our main point is that the choice of methodology for model comparison of extremes (with respect to choice of observational data and accounting for geographic sampling) is relatively unimportant for well-sampled areas but very important for sparsely sampled areas. While we focus on the contiguous United States in this paper, our results have important implications for other global land regions where the sampling problem is more severe.
READ FULL TEXT