As advances have been made in climate models and process-based land-surface schemes, the requirements of long-term high-resolution climate data for efficient modelling and performance assessment have also increased. Recently, long-term gridded climate data have been actively employed to produce high-resolution climate projections by using statistical downscaling methods because the spatial resolutions of statistical downscaling methods correspond to those of historical data. This study presents an improved geographic information system (GIS)-based regression model (IGISRM) that incorporates an adaptive effective radius algorithm into the structure of a previous regression model of Kongju National University (KNU/RM). Long-term gridded climate data from 1973 to 2010 are generated, and the performances of IGISRM at various spatial resolutions are evaluated to identify the most suitable grid spacing with respect to accuracy and computational efficiency. In addition, inter-comparison of the performances of IGISRM and KNU/RM is conducted with multiple performance measures and extreme indices.

A number of statistical downscaling methodologies have been introduced to bridge the gap in scale between outputs of climate models and the climate information needed to assess potential impacts at local and regional scales. Four statistical downscaling methods—Bias-Correction/Spatial Disaggregation (BCSD), Bias-Correction/ Constructed Analog (BCCA), Multivariate Adaptive Constructed Analogs (MACA), and Bias-Correction/Climate Imprint (BCCI)—are applied to downscale the latest Climate Forecast System Reanalysis (CFSR) data to stations for precipitation (PRCP), maximum temperature (TMAX), and minimum temperature (TMIN) over South Korea. All methods are calibrated with observational station data for 19 years from 1973 to 1991 and are tested for the recent 19 years from 1992 to 2010. A comprehensive suite of performance metrics is presented to inter-compare methods with respect to reproducing the sequencing of events and distribution of climate variables, spatial structure, and extremes. Based on the performance metrics, this study employs the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) to identify a robust statistical downscaling method.

The results show that both regression models produce compatible performance measures for PRCP, TMAX, and TMIN. However, IGISRM outperforms the KNU/RM in reproducing spatial distribution of precipitation events, particularly when a considerable level of spatial variability occurs in precipitation events. On the contrary, KNU/RM overpredicts many wet days with very small amounts of precipitation, referred to as the drizzling effect. Such results indicate that IGISRM has better skill for capturing the spatial heterogeneity in precipitation occurrence. In addition, IGISRM shows higher skill in reproducing extreme indices, particularly those related to wet and dry spells of precipitation owing mainly to overpredicted wet days of KNU/RM induced by an overly long radius of the influence circle. Regarding the inter-comparison of statistical downscaling methods, the downscaling skill is considerably affected by the skill of the general circulation model (GCM; (i.e., CFSR in this study)), and all methods lead to large improvements in representing all performance metrics. BCSD and BCCI, employing the spatial disaggregation algorithm, show slightly better skill in reproducing distributions and extremes, whereas BCCA and MACA, incorporating spatial weather patterns, outperform BCSD and BCCI in reproducing spatial structures. All statistical downscaling methods show lower skill at higher elevations where climate extremes are influenced by orographic and complex topographic effects. When TOPSIS is applied to the comprehensive performance metrics, MACA is identified as the most reliable and robust method for all variables, whereas BCCI shows the poorest performance owing mainly to lack of skill in simulating the spatial structure.

1. INTRODUCTION

2. PRACTICAL APPLICATIONS OF STATISTICAL DOWNSCALING METHODS

2.1 Coordinated Regional Climate Downscaling Experiment

2.2 NASA Earth Exchange Global Daily Downscaled Projections

2.3 Indian Ocean Climate Initiative Stage 3: Downscaled climate projections for Western Australia

2.4 Northwest Climate Science Center Project in the U.S.: Integrated Scenarios of the Pacific Northwest, U.S.

2.5 Downscaled CMIP3 and CMIP5 climate and hydrology projections

3. PHYSIOGRAPHIC AND CLIMATE DATA

4. METHDOLOGY

4.1 IIGISRM

4.2 Combined weighting of station and iterative station density algorithm

4.3 Precipitation and temperature predictions

4.4 Statistical downscaling methods

4.4.1 Simple Spatial Interpolation

4.4.2 Bias‐Correction/Spatial Disaggregation

4.4.3 Bias Correction/Constructed Analog

4.4.4 Multivariate Adapted Constructed Analogs

4.4.5 Bias‐Correction/Climate Imprint

4.5 Performance measures

4.5.1 Assessment of regression model skill

4.5.2 Assessment of downscaling skill

4.6 Technique for Order of Preference by Similarity to Ideal Solution

5. RESULTS AND DISCUSSION

5.1 Decision of influence radius

5.2 Assessment of IGISRM performance in selecting the spatial resolution of gridded data

5.3 Inter‐comparison of IGISRM and KNU/RM

5.4 Assessment of a suite of performance metrics for statistical downscaling methods

5.4.1 Sequencing of events

5.4.2 Distribution of daily climate data

5.4.3 Spatial correlations between stations

5.4.4 Extremes

5.5 Identifying robust statistical downscaling models with TOPSIS

6. CONCLUDING REMARKS

REFERENCES