Geração estocástica de precipitação subdiária e simulação de eventos extremos

Abstract

Rainfall time series with high temporal resolution are often required for estimating storm events for the design of urban drainage systems, for performing rainfall-runoff simulation in small catchments and for modeling flash-floods. Nonetheless, large and continuous sub-daily rainfall samples are, most often than not, unavailable. In order to obtain time series with the mentioned premises, the stochastic rainfall generators are widely used. Those generators aim to reproduce the statistical proprieties of the observed rainfall series, besides quantify the uncertainties and assess the risks involved in the extreme rainfall quantiles estimative, for an appropriate group of durations. In this research, a sub-daily stochastic rainfall generator was developed. This generator is able to appropriate simulate both extreme and regular events. To achieve this, a daily and sub-daily stochastic rainfall generators were coupled. The use of a daily model allowed to introduce variability to the series to be disaggregated, obtaining more reliable estimates for the low exceedance probability rainfall events. After generating the daily series, a non-parametric approach of resampling was used herewith a regionalised similarity approach, where the “fragments” of subdaily rainfall are randomly sampled of sub-daily record gauges at nearby stations, conditioned on the daily precipitation amount at the location of interest. The proposed disaggregation method was applied to a set of 40 rainfall gauging stations. Disaggregation of daily rainfall was performed for the durations of 60, 180, 360 and 720 minutes. Results indicated an appropriate performance for the daily rainfall generation as well as for its disaggregation, reasonably reproducing monthly, daily and sub-daily summary statistics, for the evaluated durations. In addition, the annual block-maxima behavior, even for low exceedance probabilities, was relatively well described, properly summarizing the expected variability in the quantiles. Overall, the proposed approach proved a sound alternative for simulating continuous sub-daily rainfall amounts from coarse-resolution records.