Nonparametric statistical temporal downscaling of daily precipitation to hourly precipitation and implications for climate change scenarios

- Develop a stochastic temporal downscaling model from daily rainfall to hourly.
- Adopt Genetic Algorithm to provide new temporal patterns.
- Downscale the daily precipitation of observed and RCP 4.5 and 8.5 data to hourly.
- Compare the statistical characteristics between the observed and downscaled data.

SummaryHydro-meteorological time series on finer temporal scales, such as hourly, are essential for assessing the hydrological effects of land use or climate change on medium and small watersheds. However, these time series are, in general, available at no finer than daily time intervals. An alternative method of obtaining finer time series is temporal downscaling of daily time series to hourly time series. In the current study, a temporal downscaling model that combines a nonparametric stochastic simulation approach with a genetic algorithm is proposed. The proposed model was applied to Jinju station in South Korea for a historical time period to validate the model performance. The results revealed that the proposed model preserves the key statistics (i.e., the mean, standard deviation, skewness, lag-1 correlation, and maximum) of the historical hourly precipitation data. In addition, the occurrence and transition probabilities are well preserved in the downscaled hourly precipitation data. Furthermore, the RCP 4.5 and RCP 8.5 climate scenarios for the Jinju station were also analyzed, revealing that the mean and the wet-hour probability (P1) significantly increased and the standard deviation and maximum slightly increased in these scenarios. The magnitude of the increase was greater in RCP 8.5 than RCP 4.5. Extreme events of different durations were also tested. The downscaled hourly precipitation adequately reproduced the statistical behavior of the extremes of the historical hourly precipitation data for all durations considered. However, the inter-daily relation between the 1st hour of the present day and the last hour of the previous day was not preserved. Overall, the results demonstrated that the proposed temporal downscaling model is a good alternative method for downscaling simulated daily precipitation data from weather generators or RCM outputs.