Evaluation of the spatial characteristics of climate scenarios based on statistical and dynamical downscaling for impact assessments in Japan

Abstract

AbstractRepresentation of spatial dependence on temporal variations in climate scenarios is quite important when performing impact assessments in various sectors, including water resource management, agriculture, and energy. Although the complex terrain of Japan is closely related to the formation of regional climates, the spatial aspects of climate scenarios have not been investigated. To examine how well spatial dependency on temporal variations in precipitation is represented, we analysed the dependence of the correlation of daily precipitation time series for station pairs on the geographical distance between stations in several downscaled products for Japan. Although the bias correction (BC) method using quantile mapping was sufficient for removing the inherent biases at each station, we found minimal improvement in the spatial dependency. The performance was strongly dependent on the grid spacing of the parent model when the downscaling process was simple; it could not be improved by applying BC to the model output statistics. When analogue‐type statistical downscaling was applied, the observed spatial pattern of interstation correlations was much better reproduced; this was comparable to the results obtained using a dynamical downscaling technique with grid spacing of a few kilometres. Our results emphasize the importance of selecting appropriate data to represent the spatial variability of meteorological variables because mean bias‐based or accessibility‐based data selection may produce misleading results. The combination of an analogue method with BC of model output statistics has the potential to effectively represent the marginal and spatial aspects of climate scenarios.