Long-lead station-scale prediction of hydrological droughts in South Korea based on bivariate pattern-based downscaling

Abstract

Capturing climatic variations in boreal winter to spring (December–May) is essential for properly predicting droughts in South Korea. This study investigates the variability and predictability of the South Korean climate during this extended season, based on observations from 60 station locations and multi-model ensemble (MME) hindcast experiments (1983/1984–2005/2006) archived at the APEC Climate Center (APCC). Multivariate empirical orthogonal function (EOF) analysis results based on observations show that the first two leading modes of winter-to-spring precipitation and temperature variability, which together account for ~80 % of the total variance, are characterized by regional-scale anomalies covering the whole South Korean territory. These modes were also closely related to some of the recurrent large-scale circulation changes in the northern hemisphere during the same season. Consistent with the above, examination of the standardized precipitation evapotranspiration index (SPEI) indicates that drought conditions in South Korea tend to be accompanied by regional-to-continental-scale circulation anomalies over East Asia to the western north Pacific. Motivated by the aforementioned findings on the spatial–temporal coherence among station-scale precipitation and temperature anomalies, a new bivariate and pattern-based downscaling method was developed. The novelty of this method is that precipitation and temperature data were first filtered using multivariate EOFs to enhance their spatial–temporal coherence, before being linked to large-scale circulation variables using canonical correlation analysis (CCA). To test its applicability and to investigate its related potential predictability, a perfect empirical model was first constructed with observed datasets as predictors. Next, a model output statistics (MOS)-type hybrid dynamical–statistical model was developed, using products from nine one-tier climate models as inputs. It was found that, with model sea-level pressure (SLP) and 500 hPa geopotential height (Z500) as predictors, statistically downscaled MME (DMME) precipitation and temperature predictions were substantially improved compared to those based on raw MME outputs. Limitations and possible causes of error of such a dynamical–statistical model, in the current framework of dynamical seasonal climate predictions, were also discussed. Finally, the method was used to construct a dynamical–statistical system for 6 month-lead drought predictions for 60 stations in South Korea. DMME was found to give reasonably skillful long-lead forecasts of SPEI for winter to spring. Moreover, DMME-based products clearly outperform the raw MME predictions, especially during extreme wet years. Our results could lead to more reliable climatic extreme predictions for policymakers and stakeholders in the water management sector, and for better mitigation and climate adaptations.