Nonparametric temporal downscaling with event-based population generating algorithm for RCM daily precipitation to hourly: Model development and performance evaluation

Abstract

It is critical to downscale temporally coarse GCM or RCM outputs (e.g., monthly or daily) to fine time scales, such as sub-daily or hourly. Recently, a temporal downscaling model employing a nonparametric framework (NTD) with k-nearest resampling and a genetic algorithm has been developed to preserve key statistics as well as the diurnal cycle. However, this model's usage can be limited in estimating precipitation for design storms or floods because the key statistics of annual maximum precipitation (AMP), especially for longer hourly durations, present a systematic bias that cannot be preserved due to the discontinuity of multiday consecutive precipitation events in the downscaling procedure. In the current study, we develop an approach to downscale a consecutive daily precipitation at once focusing on the reproduction of AMP totals for different durations instead of day-by-day downscaling. The proposed model has been verified with the precipitation datasets for the 60 stations across South Korea over the period 1979-2005. Additionally, two validation studies were performed with the recent datasets of 2006-2014 and nearest neighbor stations. The verification and the two validation tests conclude that the population-based NTD (PNTD) model proposed in the current study is superior to the existing NTD model in preserving the key statistics of the observed AMP series and suitable for downscaling future climate scenarios. (C) 2017 Elsevier B.V. All rights reserved.