Figure 1
From: Physics-guided probabilistic modeling of extreme precipitation under climate change
The performance of the Bayesian model is compared to using the raw ensemble in terms of out-of-sample accuracy and predictive coverage across 18 watersheds that comprise the continental U.S. Coloring represents accuracy of the posterior relative to using an ensemble average approach, measured as \(\frac{RMS{E}_{p}}{RMS{E}_{e}}\). Accuracy is higher in 15 out of 18 watersheds. In 15 of 18 watersheds, using a 99% credible interval, posterior coverage is larger than or equal to than ensemble coverage in all watersheds, where coverage ranges from 0 to 1 (not depicted). The three regions where posterior coverage is smaller than that of the original ensemble are the Tennessee, Pacific Northwest, and California watersheds. Upper coverage is equivalent or improved in only 3 out of 18 watersheds using the same 99% credible interval, including Lower Mississippi, Texas-Gulf, and Upper Mississippi. Watersheds are labeled by name and their respective \(\frac{RMS{E}_{p}}{RMS{E}_{e}}\) values. (We use R package that wraps around the open source Javascript project Leaflet: https://rstudio.github.io/leaflet/ to create the figure).
