We used climate projections under only one climate scenario

We used climate projections under only one climate scenario (i.e., RCP4.5). For a more comprehensive analysis, other scenarios should also be used. In addition, to improve the reproducibility of the downscaling result for the current climate, the frequent re-initialization method for a numerical weather prediction model could be used, as proposed in Lo et al. (2008). As is the case for climate scenarios, the study farnesoid x receptor is an important factor in the assessment of climate change. Eleven years were selected in this study for both current and future climate conditions (2000–2010 and 2060–2070), so that mean conditions in the two periods could be compared. However, these periods may not be sufficient to investigate the effects of decadal or inter-decadal variability. Longer period downscaling (for example, 30 years) is proposed for examining the effects of decadal/inter-decadal variations in global warming. In addition, spectral nudging in a dynamical downscaling is another powerful technique for improving the reproducibility of current climate conditions. Using such methods, the characteristics of HF-PGW conditions would be reflected more directly by the downscaling result, and thus the effects of global warming seen in HF-PGW conditions could be examined more accurately.
We used global warming projections from five different AOGCMs for preparation of the HF-PGW conditions, because the application of a multi-model ensemble is indispensable for assessment of the future climate. From a technical standpoint, more climate model outputs should be used for this preparation in order to reduce uncertainties. However, the evaluation of the model performance and AOGCM selection are also important. Knutti et al. (2010) discussed how to optimally combine the outputs from multiple AOGCMs in CMIP3, and they suggested that considerable improvement could be expected with ensembles of as many as five models. They noted that an ensemble degrades when poorly performing models are included. Some studies have investigated biases and dependence across different AOGCMs in CMIP3, and concluded that the effective number of models was much smaller than the actual number used for investigation (Jun et al., 2008; Pennell and Reichler, 2011). It was also noted that, depending on the application, lymph would be appropriate to weight the AOGCMs according to regions and seasons (Gleckler et al., 2008; Tebaldi et al., 2005). In future studies, AOGCM selection and the development of methods to evaluate multi-model downscaling results would be an important subject for research.

Acknowledgements
The author is grateful for use of CMIP5 products archived and published by the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and to all the research institutes contributing to this activity. The Japanese 25-year reanalysis data were provided by the Japan Meteorological Agency (JMA) and the Central Research Institute of Electric Power Industry (CRIEPI). The research was supported through Core Research for Evolutional Science and Technology (CREST) funded by the Japan Science and Technology Agency (JST). The author also greatly appreciates the comments and suggestions by the anonymous reviewers, which greatly improve the quality of this paper.

Introduction
Natural sciences, from geomorphology to vegetation sciences, show increasing interest in applications based on the accurate representation of topography, as provided by the most recent digital elevation models (DEMs) (Muñoz and Kravchenko, 2012; Elshehaby et al., 2013; Petroselli et al., 2013, 2014; Fan et al., 2014; Nourani and Zanardo, 2014). Hydrology is one discipline that has directly benefited from available terrain models. Virtually all watershed representations, however, contain flat areas or depression pixels that may be artifacts or actual landscape representations (Fisher and Tate, 2006; Pan et al., 2012). These features cause interruptions while calculating downstream flow through a DEM (Grimaldi et al., 2007; Arnold, 2010; Petroselli and Alvarez, 2012), which is the basis for every posterior hydrological modeling step. It has been found that even applications of more recent hydrological models can provide incorrect results when performed with the most detailed DEMs if depressions and flat areas are not properly addressed (Petroselli, 2012).