Downscaling With Constructed Analogues: Daily Precipitation and Temperature Fields Over The United States

Abstract

Daily precipitation and average temperature patterns for the contiguous United States were downscaled from a 2.5 x 2.5 degree (coarse) resolution grid to a 1/8 x 1/8 degree (fine) resolution grid using a constructed‐analogues method. Choice of predictors, and the selection of subsets of most‐suitable historical dates to be included in the constructed analogues proved to be important determinants of the method’s skill, especially for precipitation. The downscaling method skillfully reproduces daily variations of precipitation and average temperature anomalies, as well as seasonal cycles, across the contiguous United States. The method tends to overestimate the number of wet days, producing a very light “drizzle” on many of the effectively dry days. There are also biases in the monthly climatologies of precipitation and average temperature in some regions, which tend to average out at annual timescales. Averaging daily downscaled patterns into monthly means yielded even more skillful results, capturing about 55 percent of the variations of monthly precipitation anomalies and about 80 percent of the variations of average temperature monthly anomalies across the contiguous United States. The choice of the domain of the predictor also influences the skill. For example, in California, the most skillful precipitation downscaling was obtained when the precipitation predictors covered the state, whereas average temperature downscaling was most skillful when average temperature predictors included continent‐wide patterns. Overall, the method showed encouraging results for downscaling daily precipitation and average temperature continentalwide patterns in North America—in particular, those of the western United States.