A Gaussian model lowered dry-region count error by 6.6% against independence across nine US regions in 1961-2025

We tested the distribution of the number of dry regions across nine NOAA US climate regions. Each test year in 1961-2025 used only earlier years from 1895 onward. Dry means annual precipitation at or below the regional training-set 20th percentile. A Gaussian copula lowered mean ranked probability score from 0.09596 under independence to 0.08962, a decrease of 6.6%. The mean difference was -0.00635, with a conditional 95% five-year block interval of -0.01169 to -0.00117. Its difference against earlier whole-year patterns was -0.00074, with an interval of -0.00194 to 0.00044. Only two test years had at least five dry regions, compared with 5.06 expected under the Gaussian model. A conditional independent-year count check gave a probability of 0.110 for at most two events. Thus, better count-distribution performance does not establish correct extreme-event probabilities. A changing marginal dry rate, few extremes, source revision and conditional intervals limit the result. The package includes annual evidence, all probabilities, code, source hashes, a figure and an arithmetic audit.

precipitation · spatial drought · Gaussian copula · probability score · climate regions · out-of-sample evaluation