Regional intensity-duration-frequency curves derived from ensemble empirical mode decomposition and scaling property

Abstract

Because of the large spatial variability of most precipitation data, site-specific Intensity Duration Frequency (IDF) curves generally cannot be reliably transferred to ungauged sites, even those located nearby. Further, most IDF curves of Canada are traditionally fitted to the Extreme Value type I (EVI) probability distribution (PD) with parameters derived by the Method of Moment (MOM), which as expected, are not as accurate as General Extreme Value (GEV) PD with parameters derived by the probability weighted moment (PWM) method. We propose deriving regional IDF curves based on the scaling property of precipitation data derived via the ensemble empirical mode decomposition (EEMD). Selected stations of annual maximum precipitation were first decomposed by EEMD to intrinsic mode functions (IMFs). Next, the scaling property of IMFs was examined and representative scale exponents were extracted. Results show that quantile estimates derived from GEV-PWM are more accurate than those derived from EVI-MOM, whose underestimation of rainfall intensity becomes obvious when the return period is over 25-yr, especially for storms of duration less than an hour. Therefore, quantile estimates of the GEV-PWM were selected to derive regional IDF curves. Generally three of the four IMFs of the precipitation data showed simple scaling property, which were used to derive regional IDF curves. These regional IDF curves derived from the scaling IDF and EEMD approach predicted accurate storm intensities for rain gauging sites at both the calibration and validation stages, though for storm of high return periods, e.g., about 100-year or higher, the predicted storm intensities are more subjected to uncertainties.