Projected Changes of Precipitation Characteristics Depend on Downscaling Method and Training Data: MACA versus LOCA Using the U.S. Northeast as an Example.

This study compares projected changes of precipitation characteristics in the U.S. Northeast in two analog-based climate downscaling products, Multivariate Adaptive Constructed Analogs (MACA) and Localized Constructed Analogs (LOCA). The level of similarity or differences between the two products varies with the type of precipitation metrics. For the total precipitation amount, the two products project significant annual increases that are similar in magnitude, spatial pattern, and seasonal distribution, with the largest increases in winter and spring. For the overall precipitation intensity or temporal aggregation of heavy precipitation (e.g., number of days with more than one inch of precipitation, the simple intensity index, and the fraction of annual precipitation accounted for by heavy events), both products project significant increases across the region with strong model consensus; the magnitude of absolute increases are similar between the two products, but the relative increases are larger in LOCA due to an underestimation of heavy precipitation in LOCA's training data. For precipitation extremes such as the annual maximum 1-day precipitation, both products project significant increases in the long-term mean, but the magnitude of both the absolute and relative changes are much smaller in LOCA than in MACA, indicating that the extreme precipitation differences in the training data are amplified in future projections as a result of the analog-based downscaling algorithms. The two products differ the most in the intensity and frequency of rare extremes (e.g., 1-in-20-years events) for which MACA projects significant increases while the LOCA-projected changes are inconclusive over much of the study area. [ABSTRACT FROM AUTHOR]