AbstractLong-term climate data play a vital role in climate change analysis, but in situ observations are spatio-temporally limited in many regions throughout the world – including South Korea. A lack of data may lead to considerable uncertainties in hydrological applications and make it difficult to detect nonstationarity that can significantly affect future risk estimation. This challenge is especially the case in rainfall frequency analysis. This thesis aims to analyse the climate change impact, especially for rainfall intensity change, by adopting globally available century-long reanalysis data. Although a few century-long reanalyses have been developed, there is still a knowledge gap on the suitability of such data for regional-scale studies. Due to the systematic errors in the data, bias correction should be considered before any applications. Accordingly, this thesis mainly consists of three parts: 1) the assessment of the reanalysis data in regional-scale analyses, 2) the bias correction of the century-term reanalysis data to reduce design rainfall uncertainty, and 3) the reanalysis-data-based nonstationary rainfall frequency analysis.
The first part evaluates the long-term reanalyses in South Korea, which is the selected study region in this thesis. Multi-decadal reanalyses (ERA-20cm, ERA-20c, ERA-40, and 20CR) for monthly mean precipitation and temperature were first assessed in South Korea compared with the global gridded observations (CRUv3.23 and GPCCv7). This analysis showed that reanalysis data could statistically reproduce observations well, but all products should be locally adjusted before their hydrological applications. A two-step approach was followed for the bias correction of the century-long reanalysis data: 1) a quantile mapping (QM) method using a composite gamma-Pareto distribution for the reference period (1973–2010) and 2) a trend-preserving QM method (i.e. quantile delta mapping method) for the whole 20th century. The evaluation suggested that the proposed bias correction scheme was useful for a regional-scale modelled data with a limited network of rain gauges. Meanwhile, the century-long data contributed to the reduction of design rainfall uncertainty. The final part presents a reanalysis-product-based nonstationary analysis. This new approach suggested that the stationary approach could underestimate future risk in many areas.
The findings in this thesis show that despite the biases, reanalysis data can provide valuable information that helps researchers to understand an area’s climate change impact using limited observations.
|Date of Award||23 Jan 2019|
|Supervisor||Dawei Han (Supervisor)|