[1] In this study, we evaluate the accuracy of four regional climate models (NHRCM, NRAMS, TRAMS, and TWRF) and one bias correction-type statistical model (CDFDM) for daily precipitation indices under the present-day climate (1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004) over Japan on a 20 km grid interval. The evaluated indices are (1) mean precipitation, (2) number of days with precipitation ≥1 mm/d (corresponds to number of wet days), (3) mean amount per wet day, (4) 90th percentile of daily precipitation, and (5) number of days with precipitation ≥90th percentile of daily precipitation. The boundary conditions of the dynamical models and the predictors of the statistical model are given from the single reanalysis data, i.e., JRA25. Both types of models successfully improved the accuracy of the indices relative to the reanalysis data in terms of bias, seasonal cycle, geographical pattern, cumulative distribution function of wet-day amount, and interannual variation pattern. In most aspects, NHRCM is the best model of all indices. Through the intercomparison between the dynamical and statistical models, respective strengths and weaknesses emerged. Briefly, (1) many dynamical models simulate too many wet days with a small amount of precipitation in humid climate zones, such as summer in Japan, relative to the statistical model, unless the cumulus convection scheme improved for such a condition is incorporated; (2) a few dynamical models can derive a better high-order percentile of daily precipitation (e.g., 90th percentile) than the statistical model; (3) both the dynamical and statistical models are still insufficient in the representation of the interannual variation pattern of the number of days with precipitation ≥90th percentile of daily precipitation; (4) the statistical model is comparable to the dynamical models in the long-term mean geographical pattern of the indices even on a 20 km grid interval if a dense observation network is applicable; (5) the statistical model is less accurate than the dynamical models in the temporal variation pattern due to the strong dependence of the predictand on the relatively less accurate predictor (daily reanalysis precipitation); and (6) the simple statistical model is less plausible in the physical sense because of the oversimplification of underlying physical processes compared to the dynamical models and more sophisticated statistical models.
Under the framework of the Regional Climate Model Intercomparison Project (RMIP III), simulation results from six regional climate models (RCMs) and two global climate models (GCMs) were used to generate climate extreme indices for the present and future over China using two ensemble methods. All the models reasonably captured the observed climate extremes, and performance‐based ensemble averaging (PEA) outperformed the individual model and equal‐weighted averaging (MME) for the control climate. However, noticeable cold deficiencies in temperature extremes were found over areas with complex topography, and too frequent heavy precipitation at smaller intensities was simulated using the multiple model ensembles. Under the A1B scenario for 2041–2060, widespread increases in the 90th percentiles of the maximum temperatures (Tmax90p) and the 10th percentile of the minimum temperatures (Tmin10p) were projected, with larger increases in winter than in summer. Greater intensities in precipitation extremes were projected over China, with the exception of Inner Mongolia. Large uncertainties exist in the projected mean diurnal temperature range (Trange), number of days with precipitation exceeding 10 mm (R10) and the maximum number of consecutive dry days (CDD) because of disagreements in both the magnitudes and signs of the climate model projections, and even the two ensemble methods presented opposite signs over some regions.
[1] This study proposes the stochastic weather generator (WG)-based bootstrap approach to provide the probabilistic climate change information on mean precipitation as well as extremes, which applies a WG (i.e., LARS-WG) to daily precipitation under the present-day and future climate conditions derived from dynamical and statistical downscaling models. Additionally, the study intercompares the precipitation change scenarios derived from the multimodel ensemble for Japan focusing on five precipitation indices (mean precipitation, MEA; number of wet days, FRE; mean precipitation amount per wet day, INT; maximum number of consecutive dry days, CDD; and 90th percentile value of daily precipitation amount in wet days, Q90). Three regional climate models (RCMs: NHRCM, NRAMS and TWRF) are nested into the high-resolution atmosphere-ocean coupled general circulation model (MIROC3.2HI AOGCM) for A1B emission scenario. LARS-WG is validated and used to generate 2000 years of daily precipitation from sets of grid-specific parameters derived from the 20-year simulations from the RCMs and statistical downscaling model (SDM: CDFDM). Then 100 samples of the 20-year of continuous precipitation series are resampled, and mean values of precipitation indices are computed, which represents the randomness inherent in daily precipitation data. Based on these samples, the probabilities of change in the indices and the joint occurrence probability of extremes (CDD and Q90) are computed. High probabilities are found for the increases in heavy precipitation amount in spring and summer and elongated consecutive dry days in winter over Japan in the period 2081-2100, relative to 1981-2000. The joint probability increases in most areas throughout the year, suggesting higher potential risk of droughts and excess water-related disasters (e.g., floods) in a 20 year period in the future. The proposed approach offers more flexible way in estimating probabilities of multiple types of precipitation extremes including their joint probability compared to conventional approaches.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.