A local-regional scaling-invariant Bayesian GEV model for estimating rainfall IDF curves in a future climate

• Published in: Journal of hydrology (Amsterdam) Vol. 566; pp. 73 - 88
• Main Authors: Lima, Carlos H.R., Kwon, Hyun-Han, Kim, Yong-Tak
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2018
• Subjects: Bayesian model; Climate change; Intensity-Duration-Frequency curves; Rainfall; Regional estimates; Rainfall; Regional estimates; Climate change; Bayesian model; Intensity-Duration-Frequency curves
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2018.08.075

Future changes in rainfall patterns induced by climate changes will affect society and ecosystems, and quantifying these changes is of utmost importance for the management of hydroclimate risk. In particular, the estimation of intensity-duration-frequency (IDF) curves for rainfall data is a routine procedure in urban hydrology and hydraulic studies and should be revisited to reflect future changes in rainfall variability. In this work, we propose a novel methodology based on the scaling-invariant property of rainfall duration versus intensity to estimate parameters of a generalized extreme value (GEV) distribution at sub-daily scales. A Bayesian inference framework is developed so that uncertainties are reduced and can be easily propagated to IDF curves. The proposed model can be employed to: (i) improve local (at-site) GEV estimates for sites with limited rainfall records; (ii) estimate GEV parameters at sub-daily scales and construct IDF curves for sites where only daily rainfall records are available (partially gauged sites); (iii) construct regional IDF curves for homogeneous hydrologic regions; and (iv) update local and regional IDF curves from simulations of future daily rainfall. The model is tested using historical rainfall data from 18 gauges located in the Han River Watershed in South Korea, and projected climate change scenarios RCP 6 and RCP 8.5 from the Met Office Hadley Centre HadGEM2-AO model. When considering historical data, the results show that the model satisfactorily estimate IDF curves for both gauged and partially gauged sites. In future scenarios, the model reveals a substantial increase in rainfall events of rare intensity (large return periods), mostly due to changes in the rainfall variability rather than changes in the average rainfall. Particularly, for a 100-year return period event, we expect an increase of about 23% in scenario RCP 6 and about 30% under scenario RCP 8.5 when projected using regional IDF curves. To the best of our knowledge, this is the first statistical approach in the literature to assess future changes in regional IDF curves, which in our opinion is more suitable than evaluating local estimates only.