Downscaled rainfall projections in south Florida using self-organizing maps

Highlights: • Downscaling of rainfall totals in south Florida, U.S., reproduced observed values during the historical period (1976-2005). • Raw climate model outputs poorly replicated historical rainfall totals but statistical downscaling reproduced observed values. • Projections of future rainfall yielded drying conditions, magnitude depends on timeframe and future greenhouse gas emissions. • Anticipated frequency of wet days will likely decrease and the average length of dry spells increases. We make future projections of seasonal precipitation characteristics in southern Florida using a statistical downscaling approach based on Self Organized Maps. Our approach is applied separately to each three-month season: September–November; December–February; March–May; and June–August. We make use of 19 different simulations from the Coupled Model Inter-comparison Project, phase 5 (CMIP5) and generate an ensemble of 1500 independent daily precipitation surrogates for each model simulation, yielding a grand ensemble of 28,500 total realizations for each season. The center and moments (25%ile and 75%ile) of this distribution are used to characterize most likely scenarios and their associated uncertainties. This approach is applied to 30-year windows of daily mean precipitation for both the CMIP5 historical simulations (1976–2005) and the CMIP5 future (RCP 4.5) projections. For the latter case, we examine both the "near future" (2021–2050) and "far future" (2071–2100) periods for three scenarios (RCP2.6, RCP4.5, and RCP8.5).