Development of latent resampling downscaling and its application to model bias and climate change projection

Development of latent resampling downscaling and its application to model bias and climate change projection | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development of latent resampling downscaling and its application to model bias and climate change projection Hajime Ohnishi, Masaru Inatsu, Ryo N. Matsuoka This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7266042/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract We have developed a cost-effective downscaling technique called latent resampling downscaling (LRDS) to address model uncertainty in regional climate change assessments. In this study, the LRDS was used to generate a surrogate dynamical downscaling (DDS) dataset for a coupled model intercomparison project phase 5 global climate model. This was achieved by sampling from a large ensemble of DDS datasets from d4PDF project. The sampling process was guided by the probability density functions of the global model’s weather patterns, which were classified using a self-organizing map algorithm. We applied LRDS to investigate summertime precipitation over Kyushu Island, Japan. The present-climate simulation revealed considerable inter-model variety in reproducing sea level pressure patterns around Japan in boreal summer. A storyline approach was employed to characterize three distinct behaviors of LRDS in simulating climatological and extreme rainfall over Kyushu under the present climate. Using LRDS, we also evaluated the projected response to climate change. Two contrasting storylines were identified: one showing an increase in rainfall over western Kyushu, and the other indicating increased rainfall over eastern Kyushu. Dynamical downscaling Regional climate change Large ensemble datasets Self-organizing map Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 1. Introduction Dynamical downscaling (DDS) refers to a regional atmospheric model (RAM) simulation applied over a limited area, using outputs from general circulation models (GCMs) as lateral boundary conditions (Giorgi and Bates 1989 ; Wang et al. 2004 ; Giorgi 2019 ). This technique has been widely used to compensate for the insufficient spatial resolution of GCMs in applications that require climate change information at approximately O(10 km) resolution. However, DDS results are strongly influenced by the lateral boundary conditions derived from GCMs. Consequently, DDS outputs often inherit biases from the driving GCMs and introduce additional biases from the RAMs themselves (McSweeney et al. 2015 ; Wu et al. 2005 ). Therefore, regional climate change projections should account for both uncertainties arising from GCMs and those associated with RAMs. This consideration has motivated multi-GCM and multi-RAM experiments to evaluate model uncertainty in regional climate in recently coordinated projects, such as the ENSEMBLES project (Déqué et al. 2012 ), NARCCAP (Mearns et al. 2012 ), CORDEX-SEA (Tangang et al. 2018), and the Japanese RECCA project (Inatsu et al. 2015 ). Applying DDS to all possible combinations of GCMs, RAMs, and emission scenarios would be impractical, especially considering that the coupled model intercomparison project phase5 (CMIP5; Taylor et al. 2012 ) involved 30 to 40 GCMs and included multiple representative concentration pathways (RCPs; Moss et al. 2010 ). Consequently, even in many recent projects that coordinated multi-GCM and multi-RAM experiments, the matrix of GCM–RAM combinations was mostly sparse. Furthermore, large-ensemble GCM experiments and their associated DDS simulations are essential for probabilistic risk assessments of natural hazards related to extreme weather events. Thus, the combination of multi-model and large-ensemble experiments poses a significant challenge to assessing model uncertainty in climate change impacts on such events through DDS. Recently, Mizuta et al. ( 2017 ) developed a large-ensemble dataset called d4PDF, which consists of GCM simulations and associated DDS results for the area around Japan under various climate conditions. The ensemble includes 3,000 years of simulations under 20th-century historical conditions and 5,400 years under a future climate condition with a global mean temperature increase of 4 K. This exceptionally large ensemble size allowed the assumption that d4PDF encompasses nearly the full range of internal variability under both current and future climate conditions. Therefore, unless the inter-model discrepancies are extremely large, it is feasible to approximate any GCM outputs by subsampling from d4PDF. This insight motivated us to assess model uncertainty in regional climate by selecting appropriate subsets from such a large-ensemble dataset, comprising a GCM and its associated DDS for the target region—such as d4PDF. The objective of this paper is to develop a method for estimating the downscaled results of any GCM outputs using a reference large-ensemble simulation without performing additional DDS computations. Assuming that any GCM results are encompassed within d4PDF result, we approximate a particular GCM result as a subset of d4PDF by focusing on the probability density function (PDF) of a specific climatic variable over the targeted area. The statistics of the DDS data for this d4PDF subset can then be regarded as the surrogated DDS data for the GCM. We named this method latent resampling downscaling (LRDS). In this study, we demonstrated the LRDS results focusing on summer precipitation in Kyushu. This paper is organized as follows: Section 2 discusses the data used in this study. Section 3 explains the procedures of LRDS. Sections 4 and 5 present the LRDS results for present and future climates, respectively. Finally, Section 6 provides a summary with discussion on the results. 2. Data 2.1. d4PDF We used the d4PDF dataset as the reference for LRDS. The d4PDF consists of a GCM experiment with a horizontal resolution of 60 km and a DDS experiment with a horizontal resolution of 20 km for a limited area around Japan (Fig. 1 a). The d4PDF experiments were conducted under several climate conditions. Historical experiments were coordinated as the time integration from various initial conditions with prescribed sea surface temperatures (SSTs) and other factors as observed from 1950 to 2010. The 4K experiments were also conducted, in which the detrended SST from 1950 to 2010 was increased to achieve a global mean temperature rise of 4K by scaling the value based on model’s climate sensitivity. The 4K experiment approximately corresponds to the period around 2090 of the RCP8.5 scenario experiment in the CMIP5. We here analyzed 50 ensemble members from the historical experiments and 90 ensemble members from the 4K experiment, in which GCM experiment and the associated DDS experiments were both available. Although the SST distributions in the 4K experiments had six different warming patterns to incorporate climate model uncertainties, this study did not take those uncertainties into account. The DDS experiments associated with the GCM provided outputs by integrating a regional atmospheric model with the GCM results imposed as lateral boundary conditions. In this study, we focused on precipitation over Kyushu Island (Fig. 1 b) in June, July, and August (JJA). 2.2. CMIP5 The CMIP5 ensemble data was used to demonstrate the LRDS results. We utilized 41 models from the 20th-century historical experiment and 33 models from the RCP8.5 scenario experiment (Table 1 ). Because CMIP5 had a different grid system from d4PDF, the CMIP5 data were linearly interpolated to match the grid points of d4PDF. Here, the period from 1951 to 2005 in the historical experiment corresponded to d4PDF historical experiment, and the period from 2081 to 2100 in the RCP8.5 scenario experiment corresponded to d4PDF 4K experiment. Table 1 Specifications of the CMIP5 models used in this study # Model Present climate (days) Future climate (days) 1 ACCESS1-0 5,060 1,840 2 ACCESS1-3 5,060 1,840 3 BNU-ESM 5,060 1,840 4 CCSM4 5,060 1,840 5 CESM1-BGC 5,060 1,840 6 CESM1-CAM5 5,060 1,840 7 CESM1-FASTCHEM 5,060 N/A 8 CMCC-CESM 5,060 1,840 9 CMCC-CM 5,060 1,840 10 CMCC-CMS 5,060 1,840 11 CNRM-CM5 5,060 1,840 12 CSIRO-Mk3-6-0 5,060 1,840 13 CSIRO-Mk3L-1-2 5,060 N/A 14 CanCM4 4,140 N/A 15 CanESM2 5,060 1,840 16 EC-EARTH 5,060 1,840 17 FGOALS-g2 5,060 1,840 18 FGOALS-s2 5,060 1,840 19 GFDL-CM3 5,060 1,840 20 GFDL-ESM2G 5,060 1,840 21 GFDL-ESM2M 5,060 1,840 22 GISS-E2-H 5,060 N/A 23 GISS-E2-R 5,060 N/A 24 HadCM3 4,950 N/A 25 HadGEM2-AO 4,950 1,800 26 HadGEM2-CC 4,950 1,800 27 HadGEM2-ES 4,950 1,800 28 IPSL-CM5A-LR 5,060 1,840 29 IPSL-CM5A-MR 5,060 1,840 30 IPSL-CM5B-LR 5,060 1,840 31 MIROC-ESM N/A N/A 32 MIROC-ESM-CHEM N/A N/A 33 MIROC4h 5,060 N/A 34 MIROC5 5,060 1,840 35 MPI-ESM-LR 5,060 1,840 36 MPI-ESM-MR 5,060 1,840 37 MPI-ESM-P 5,060 N/A 38 MRI-CGCM3 5,060 1,840 39 MRI-ESM1 5,060 1,840 40 NorESM1-M 5,060 1,840 41 bcc-csm1-1 5,060 1,840 42 bcc-csm1-1-m 5,060 1,840 43 inmcm4 5,060 1,840 3. Method 3.1. LRDS Procedure Figure 2 illustrates the procedure of LRDS. We assumed that a large-ensemble GCM dataset and its associated DDS dataset for a specific limited region were already provided as the reference. It was also assumed that the target dataset only contained GCM output with a smaller sample size than the reference and its associated DDS has not been performed. LRDS proposed a method, in such a case, to approximate the statistics of the associated DDS results without requiring additional DDS. LRDS involved two processes: PDF matching between reference GCM and target GCM data (the arrow from right to left in the upper level of Fig. 2 ), and DDS subsampling from the reference data (the arrow from left to right in the lower level). We here demonstrated the LRDS with the reference dataset d4PDF and the target datasets CMIP5. The PDF matching, the first half of the procedure, was conducted by comparing the reference and target GCM data on the latent space generated by the self-organizing map (SOM) algorithm (Kohonen, 1982 ). We here applied the algorithm to the gridded data of sea level pressures (SLPs) in the limited domain bounded by 20°N–50°N and 120°E–150°E (Fig. 1 a). Moreover, the SOM configuration was set to 10×10 in order to adequately represent the various weather patterns in both the reference and target GCM data. Given the larger sample size of the reference dataset, its subset was sufficient to reproduce the PDF of the target data by randomly sampling reference data assigned to each SOM node. Thus, the PDF matching enabled us to subsample the reference dataset in correspondence with the target dataset. In the second half of the procedure, by utilizing the linkage between the GCM and DDS in the reference data, the subsampling obtained through the PDF matching can be directly transferred to the DDS data. This approach assumes that regional phenomena represented in the DDS data are closely associated with large-scale atmospheric patterns captured in the SLP-based SOM. The DDS data corresponding to the target dataset could then be generated under this assumption. Based on the subsampled DDS data, we here created the statistics for the target dataset related to summertime rainfall over Kyushu (Fig. 1 b). 3.2. PDF matching We first constructed a latent space using SOM, based on training SLP data sampled every 24 hours from six ensemble members of the d4PDF dataset and a single ensemble from each CMIP5 model. For the analysis of the present climate, a slice of latent space was created using daily data from JJA months, combining six ensembles from the d4PDF historical experiment with 20th-century experiments from 41 CMIP5 models (Table 1 ). Since each d4PDF ensemble comprised 60 years of data, the combined d4PDF dataset totaled 33,120 days. For CMIP5, we utilized the full range of the 20th-century experiment from 1951 to 2005, resulting in approximately 5,040 days per model (Table 1 ). Exceptions included CanCM4 (model #14) that provided only 45 years of data after 1961 and the Hadley Centre model series (models #24–#27) that adopted a 360-day calendar. Multi-model JJA-mean patterns were computed for the present climate, and the principal component (PC) analysis was conducted using SLP anomalies, defined as deviations from the mean SLP. For the future climate analysis, another slice of latent space was created using daily JJA data from both present and future climate conditions. The dataset included six ensembles from d4PDF historical experiments (33,120 days in total) and six ensembles from d4PDF 4K experiments, each imposing different SST patterns (also totaling 33,120 days). Additionally, we included outputs from 33 CMIP5 model: 20 years of data (1981 to 2000) from the 20th-century experiments and 20 years of data (2081 to 2100) from RCP8.5 experiments. These periods provided 1,840 days of data per experiment per model, except for models #25–#27 that adopted a 360-day calendar (Table 1 ). Multi-model JJA-mean patterns were calculated for both present and future climates, and the PC analysis was again conducted using SLP anomalies. For computational efficiency, the SOM was trained using the projection of the SLP anomalies onto the first 60 PC modes over the limited domain (Fig. 1 a). All non-training data were subsequently projected onto the latent space by assigning each timestep to the SOM node whose SLP pattern was closest in Euclidean distance. As a result, every timestep in the 6-hourly d4PDF and daily CMIP5 datasets was associated with a specific SOM node, uniquely identifying the timestamp, ensemble member, and model. The PDF on the latent space was proportional to the number of timestamps assigned to each SOM node over 10×10 SOM nodes. A model-specific PDF was also constructed using all available data from each model. In the present climate analysis, 6-hourly data from the d4PDF historical experiment were randomly sampled without duplication to match the PDF of a CMIP5 model’s 20th century experiment on the present-climate latent space, with a total of 10,000 samples selected. In future climate analysis, using the latent space created from the combined present and future climate datasets, the 6-hourly data from the d4PDF historical experiment were randomly sampled to match the PDF of a CMIP5 model’s 20th century experiment, and the 6-hourly data from the d4PDF 4K experiment were randomly sampled to match the PDF of the same model’s RCP 8.5 experiment. The sampled number was 10,000 as well. 3.3. DDS subsampling The DDS subsampling was performed based on a sampling set for each CMIP5 model, which included 10,000 discontinuous timestamps and ensemble members. The GCM sampling had a 6-hourly interval, whereas the DDS model output was available at an hourly interval. The d4PDF DDS results were then sliced within a three-hour window before and after each timestamp in the GCM sampling. In this paper, based on this subsampled data, we took the time average to check the climatological precipitation and calculated the 99th percentile values in hourly precipitation intensity to demonstrate the extreme precipitation. It should be noted that timeseries analysis could not be conducted by LRDS due to the discontinuous nature of the subsampled data. 3.4. Storyline-based approach We employed a storyline-based approach, following the framework proposed by Shepherd ( 2019 ), to illustrate representative patterns emerging from the LRDS results across multiple models. In the present-climate analysis, we identified three typical patterns among 41 CMIP5 models (Table 1 ) by classifying their PDFs in the SOM latent space, based on SLP anomalies over the target domain (Fig. 1 a). The classification was performed using Ward’s method (Matsuoka and Inatsu, 2024 ), and an arbitrary threshold was applied to divide the models into three groups with similar PDFs in the latent space. A representative model for each group was selected as the one whose PDF had the smallest Kullback–Leibler divergence from the reference PDF defined as the group’s mean. These three storylines reflected characteristic patterns in the models' ability to reproduce synoptic-scale environments under present climate conditions. In the future climate analysis, we identified two representative patterns among 33 CMIP5 models (Table 1 ) by classifying the differences in their PDFs between present and future climates on the SOM latent space. The subsequent procedures followed those used in the present climate analysis. A representative model for each group was selected as the one whose PDF difference between the future and present climates had the smallest root-mean-square difference from the group-mean PDF difference, which served as the reference. As a result, we derived two storylines that represented characteristic model responses to climate change. 4. Present climate analysis 4.1 LRDS results Figure 3 illustrates the SLP patterns projected onto the 10×10 latent space generated by the SOM algorithm (Section 3.1 ) based on the training data that combined six ensemble members from d4PDF historical experiment (Section 2.1 ) with CMIP5 20th-century experiment for years 1951 to 2005 (Section 2.2 ). The SOM nodes captured typical weather patterns in boreal summer around Japan: a westward extension of the Bonin high (upper left nodes), intense low-pressure anomalies southwest of Japan (lower right nodes), and weak low-pressure anomalies zonally elongated over Japan (middle right nodes). The SOM algorithm uniformly mapped various SLP patterns in training d4PDF and CMIP5 data into the latent space. The sample counts exhibited a relatively uniform distribution across nodes, ranging from 1,659 at node #99 to 3,310 at node #02 (Fig. 4 a). Because most of CMIP5 models deviated from the training data, their SOM projections exhibited non-uniform distribution (Fig. A1 ). Even the d4PDF data showed a non-uniform PDF in the latent space (Fig. 4 b). The LRDS was accomplished by DDS subsampling (Section 3.3 ), in which the d4PDF DDS outputs were assembled for timeframes corresponding to GCM data subsampled in the PDF matching process. This procedure was applied to each CMIP5 model output for the present climate analysis. The multi-model mean rainfall (Fig. 5 a) estimated by the LRDS almost ranged approximately from 8 to 13 mm/day over Kyushu, with local maxima exceeding 12 mm/day on the eastern and western sides of the island. Model-to-model variation was particularly large around these local maxima of mean rainfall. In contrast, the extreme rainfall evaluated by the 99%-tile value showed a more horizontally uniform distribution, ranging from 12 to 16 mm/hour (Fig. 5 b). The inter-model standard deviation was around 1.0 mm/hour, considerably smaller than the multi-model mean. As above, the LRDS results showed little variation in the 99%-tile rainfall over Kyushu, presumably because extreme rainfall events were not primarily controlled by large-scale weather patterns. Rather, they were typically triggered by random processes under background conditions with abundant moisture in the boundary layer. 4.2. Storylines The storyline-based approach effectively streamlined the interpretation of LRDS results across 41 CMIP5 model outputs. By applying the Ward clustering method to the 41 PDFs on the latent space (Fig. 6 ), we were able to classify the models and select representative ones. An arbitrary threshold of 0.25 was used to divide the models into three distinct groups, as visualized in the dendrogram (Fig. 6 ), which represented model-to-model similarity in PDF distributions. Based on this classification, models #18, #09, and #25 (Table 1 ) were selected as three representative storylines, each capturing typical features of SLP anomalies around Japan. By retrieving the mean SLP fields for these three CMIP models, we identified distinct differences in summertime weather patterns around Japan (Figs. 7 a-c), as well as in the associated PDFs in the latent space (Figs. 7 d-f). CMIP5 model #18 exhibited relatively higher probabilities in the upper left side of the latent space (Fig. 7 d), corresponding to a more frequent westward extension of the Bonin high compared to d4PDF (Fig. 7 a). CMIP5 model #25 showed a mean SLP field characterized by a low-pressure anomaly over the subtropical ocean (Fig. 7 c), which corresponded to the SOM nodes in the lower right corner (Fig. 7 f). CMIP5 model #09, which represented the intermediate group including d4PDF, displayed a relatively uniform distribution in the latent space (Fig. 7 e) and showed less deviation from d4PDF (Fig. 7 b). Figure 8 showed the LRDS results for JJA-mean rainfall over Kyushu, along with a storyline-based approach. Model #18 exhibited lower JJA-mean rainfall over eastern Kyushu (Fig. 8 a) compared to model #25 (Fig. 8 c). The model #09 showed an intermediate pattern. These differences in local rainfall among the models were likely linked to the JJA-mean SLP fields in CMIP5 GCMs. Typhoon passages associated with weather patterns located in the lower right of the SOM map (Fig. 3 ) substantially contributed to rainfall events in eastern Kyushu. Therefore, the GCM tendency toward more frequent tropical cyclone occurrence (Fig. 7 c) was associated with heavier mean rainfall in this region (Fig. 8 c). 5. Future climate analysis 5.1. LRDS results Figure 9 displays the SLP patterns in the latent space based on the training data, in which present climate data were equally combined with future climate data (Section 3.2 ). The SOM nodes captured typical weather patterns in boreal summer around Japan, quite similar to the present climate analysis (Fig. 3 ). Therefore, the conditional PDF for present climate data (Fig. 10 a) somewhat resembled the PDF in present climate analysis (Fig. 4 a). The conditional PDF for future climate data (Fig. 10 b) was also uniform in the latent space but was slightly different from that for present climate data. Limited to the d4PDF dataset, the PDF difference between present and future climates (Fig. 10 c) ranged ± 0.4%, with an increase in probability around central nodes and a decrease in probability around lower nodes. This feature of PDF difference was consistent with the climatological SLP response to climate change in d4PDF dataset (Fig. 10 d; Takabatake and Inatsu 2021 ; Kawase et al. 2019 ): the north-to-south SLP contrast was more prominent in future climate conditions. The LRDS was conducted by applying DDS subsampling through a PDF matching process for both the present and future climates. Based on this, we therefore estimated the multi-model mean rainfall for present and future climates. Figure 11 a displays their difference in climatological rainfall over Kyushu. In the sense of multi-model mean, the rainfall increased by ~ 3 mm/day over western Kyushu and ~ 1 mm/day in eastern Kyushu. The inter-model variation was not negligible around these two peak regions. In contrast, the extreme rainfall clearly increased by ~ 1.2 mm/hour all over Kyushu with small intermodal variations (Fig. 11 b). 5.2. Storylines The storyline-based approach was also applied to the PDF difference between present and future climates across 33 CMIP5 model outputs. The Ward clustering method with an arbitrary threshold of 0.15 allowed us to divide the models into two distinct groups, as visualized in the dendrogram (Fig. 12 ). Based on this classification, models #10 and #41 (Table 1 ) were selected as two representative storylines, each capturing typical features of SLP responses to climate change around Japan. Model #10 exhibited an increase of approximately 2 mm/day in eastern Kyushu and a decrease of about 1 mm/day in northern Kyushu (Fig. 13 a). This rainfall response was consistent with a climatological cyclone tendency to the south and southeast of Japan (Fig. 13 b), suggesting an increase in typhoon passages in that region. In contrast, model #41 showed an increase of around 3 mm/day in western Kyushu and about 2 mm/day in southern Kyushu (Fig. 13 c). Climate change in this model strengthened two anticyclones located to the north and south of Japan (Fig. 13 d), leading to an enhancement of the stagnant low-pressure region. This circulation pattern was generally associated with heavy rainfall events, particularly in western Kyushu (Hirockawa et al. 2021). 6. Concluding Remarks In this study, we developed LRDS to reduce the computational cost of DDS for climate projection data. LRDS estimates the results of DDS for arbitrary climate models based on the PDF in a latent space, utilizing a large-ensemble experiment of a single climate model and its DDS that has already been conducted. We applied LRDS to summer precipitation over Kyushu under both present and future climate conditions, generating high-resolution data for the target domain corresponding to multiple GCMs. For summer precipitation in Kyushu under the current climate, we identified three distinct spatial distribution patterns: one with a precipitation maximum in the west, another with a maximum in the east, and a third exhibiting intermediate features between the two. Under future climate projections, substantial inter-model variation emerged around the peak precipitation regions in both the west and east. One model projected a greater increase in precipitation in the west, associated with a relatively low SLP response along the Japanese Islands. In contrast, another model projected an increase in precipitation in the east, linked to a low SLP response south of Japan—suggesting more frequent typhoon passages in that region. To validate the results obtained in this study, a direct comparison with actual DDS outputs would be necessary. However, the primary motivation for proposing LRDS was to avoid the high computational cost of DDS itself. Therefore, performing DDS solely for validation purposes would be impractical, though a storyline-based approach may offer some mitigation of this difficulty. As an alternative solution, we propose dividing the d4PDF ensemble into two groups. The first group includes both large-ensemble simulations and their corresponding DDS outputs, referred to as d4PDF in this study. The second group consists only of global outputs from multiple climate models, referred to here as CMIP5. By applying LRDS to the latter dataset, we can estimate downscaling statistics and compare them with the actual DDS results from the former group. This approach enables an indirect validation of LRDS. It should be noted when interpreting the results that extreme precipitation showed less model uncertainty in its climate change response. Since the LRDS in this study employed a single regional atmospheric model for DDS, the results did not account for uncertainties associated with different regional models. In general, the simulation of extreme precipitation is sensitive to model resolution and physical parameterizations. Ban et al. ( 2021 ) demonstrated that kilometer-scale models tended to produce more intense precipitation and fewer wet hours compared to 10-km scale models. Furthermore, Supari et al. (2020) reported an increase in extreme precipitation under climate change in the Asian monsoon region, although the uncertainty arising from regional models was not explicitly discussed. These previous studies as well as our LRDS results suggested that extreme precipitation might be more sensitive to the increase in atmospheric moisture associated with warming, as governed by the Clausius–Clapeyron relationship (Yamada et al. 2014 ), rather than to changes in synoptic-scale weather systems. Extreme precipitation is often caused by cumulonimbus clouds embedded within mesoscale convective systems, though 10-km resolution models often generate extreme precipitation caused by orographic uplifting. Whatever the underlying reason, extreme precipitation may be interpreted as a stochastic outcome within a fundamentally moist environment. The regional model uncertainty in extreme precipitation could then be smaller than that of mean precipitation. This implication might be validated by kilometer-scale model ensemble simulation in the future. Declarations Conflict of interest The authors declare that they have no competing interests. Funding MI is supported by the Environment Research and Technology Development Fund JPMEERF20232003 of the Environmental Restoration and Conservation Agency provided by Ministry of the Environment of Japan and by JPMXD0722680734 of MEXT, Japan. Acknowledgements We would like to thank Prof. Yousuke Sato and Prof. Tomohito J. Yamada, who give us many insightful comments for our earlier studies. This study used d4PDF produced with the Earth Simulator jointly by science programs (SOUSEI, TOUGOU, SI-CAT, DIAS) of the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), Japan. We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1 of this paper) for producing and making available their model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. Data availability The CMIP5 outputs can be obtained from https://pcmdi.llnl.gov/mips/cmip5/data-portal.html . The d4PDF outputs can be obtained from https://search.diasjp.net/en/dataset/d4PDF_GCM . References Ban N, Brisson E, Caillaud C, Coppola E, Pichelli E, Sobolowski S, Adinolfi M, Ahrens B, Alias A, Anders I, Bastin S, Belušić D, Berthou S, Brisson E, Cardoso RM, Chan SC, Christensen OB, Fernández J, Fita L, Frisius T, Gašparac G, Girogi F, Goergen K, Haugen JE, Hodnebrog Ø, Kartsios S, Katragkou E, Kendon EJ, Keuler K, Lavin-Gullon A, Lenderink G, Leutwyler D, Lorenz T, Maraun D, Mercogliano P, Milovac J, Panitz H-J, Raffa M, Remedio AR, Schär C, Soares PMM, Srnec L, Steensen BM, Stocchi P, Tölle MH, Truhetz H, Vergara-Temprado J, de Vries H, Warrach-Sagi K, Wulfmeyer V, Zander MJ (2021) The first multi-model ensemble of regional climate simulations at the kilometer-scale resolution, Part I: Evaluation of precipitation. Clim Dyn 57:275–302. https://doi.org/10.1007/s00382-021-05708-w Déqué M, Somot S, Sanchez-Gomez E, Goodess CM, Jacob D, Lenderink G, Christensen OB (2012) The spread among ENSEMBLES regional scenarios: Regional climate models, driving general circulation models and interannual variability. Clim Dyn 38:951–964. https://doi.org/10.1007/s00382-011-1053-x Giorgi F, Bates GT (1989) The climatological skill of a regional model over complex terrain. Mon Wea Rev 117:2325–2347. https://doi.org/10.1175/1520-0493(1989)1172.0.CO;2 Giorgi F (2019) Thirty years of regional climate modeling: Where are we and where are we going next? J Geophys Res Atmos 124:5696–5723. https://doi.org/10.1029/2018JD030094 Hawkins E, Ortega P, Suckling E, Schurer A, Hegerl G, Jones P, Joshi M, Osborn TJ, Masson-Delmotte V, Mignot J, Thorne P, van Oldenborgh GJ (2017) Estimating Changes in Global Temperature since the Preindustrial Period. Bull Amer Meteor Soc 98:1841–1856. https://doi.org/10.1175/BAMS-D-16-0007.1 Hirockawa Y, Kato T, Tsuruti H, Seino N (2020) Identification and classification of heavy rainfall areas and their characteristic features in Japan. J Meteor Soc Japan 98:835–857. https://doi.org/10.2151/jmsj.2020-043 IPCC (2023) Sections. In: Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J. Romero (eds.)]. IPCC Geneva Switzerland 35–115. 10.59327/IPCC/AR6-9789291691647 Inatsu M, Sato T, Yamada TJ, Kuno R, Sugimoto S, Farukh MA, Pokhrel YN, Kure S (2015) Multi-GCM by multi-RAM experiments for dynamical downscaling on summertime climate change in Hokkaido. Atmos Sci Lett 16:297–304. https://doi.org/10.1002/asl2.557 Kawase H, Imada Y, Sasaki H, Nakaegawa T, Murata A, Nosaka M, Takayabu I (2019) Contribution of historical global warming to local-scale heavy precipitation in western Japan estimated by large ensemble high-resolution simulations. J Geophys Res 124:6093–6103. https://doi.org/10.1029/2018JD030155 Kohonen T (1982) Self-organized formation of topologically correct feature maps. Biol Cybern 43:59–69. https://doi.org/10.1007/BF00337288 Matsuoka RN, Inatsu M (2024) Weather classifications for high temperatures in Japanese cities. SOLA 20:298–305. https://doi.org/10.2151/sola.2024-040 Mearns LO, Arritt R, Biner S, Bukovsky MS, McGinnis S, Sain S, Caya SD, Correia J Jr, Flory D, Gutowski W, Takle ES, Jones R, Leung R, Moufouma-Okia W, McDaniel L, Nunes AMB, Quan Y, Roads J, Sloan L, Snyder M (2012) The North American regional climate change assessment program: Overview of phase I results. Bull Amer Meteor Soc 93:1337–1362. https://doi.org/10.1175/BAMS-D-11-00223.1 McSweeney C, Jones R, Lee R, Powell D (2015) Selecting CMIP5 GCMs for downscaling over multiple regions. Clim Dyn 44:3237–3260. https://doi.org/10.1007/s00382-011-1278-8 Mizuta R, Murata A, Ishii M, Shiogama H, Hibino K, Mori N, Arakawa O, Imada Y, Yoshida K, Aoyagi T, Kawase H, Mori M, Okada Y, Shimura T, Nagatomo T, Ikeda M, Endo H, Nosaka M, Arai M, Takahashi C, Tanaka K, Takemi T, Tachikawa Y, Temur K, Kamae Y, Watanabe M, Sasaki H, Kitoh A, Takayabu I, Nakakita E, Kimoto M (2017) Over 5,000 years of ensemble future climate simulations by 60-km global and 20-km regional atmospheric models. Bull Amer Meteor Soc 98:1383–1398. https://doi.org/10.1175/BAMS-D-16-0099.1 Moss RH, Edmonds JA, Hibbard KA, Manning MR, Rose SK, van Vuuren DP, Carter TR, Emori S, Kainuma M, Kram T, Meehl GA, Mitchell JFB, Nakicenovic N, Riahi K, Smith SJ, Stouffer RJ, Thomson AM, Weyant JP, Wilbanks TJ (2010) The next generation of scenarios for climate change research and assessment. Nature 463:747–756. https://doi.org/10.1038/nature08823 Shepherd TG (2019) Storyline approach to the construction of regional climate change information. Proc R Soc A 475:20190013. http://dx.doi.org/10.1098/rspa.2019.0013 Supari TF, Juneng L, Cruz F, Chung JX, Ngai ST, Salimun E, Mohd MSF, Santisirisomboon J, Singhruck P, PhanVan T, Ngo-Duc T, Narisma G, Aldrian E, Gunawan D, Sopaheluwakan A (2021) Multi-model projections of precipitation extremes in Southeast Asia based on CORDEX-Southeast Asia simulations. Environ Res 184:109350. https://doi.org/10.1016/j.envres.2020.109350 Takabatake D, Inatsu M (2021) Summertime precipitation in Hokkaido and Kyushu, Japan in response to global warming. Clim Dyn 58:1671–1682. https://doi.org/10.1007/s00382-021-05983-7 Tangang F, Chung JX, Juneng L, Supari, Salinun E, Ngai ST, Jamaluddin AF, Mohd MSF, Cruz F, Narisma G, Santisirisomboon J, Ngo-Duc T, Tan PV, Singhruck P, Gunawan D, Aldrian E, Sopaheluwakan A, Grigory N, Remedio ARC, Sein DV, Hein-Griggs D, McGregor JL, Yang H, Sasaki H, Kumar P (2020) Projected future changes in rainfall in Southeast Asia based on CORDEX–SEA multi–model simulations. Clim Dyn 55:1247–1267. https://doi.org/10.1007/s00382-020-05322-2 Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Amer Meteor Soc 93:485–498. https://doi.org/10.1175/BAMS-D-11-00094.1 Wang Y, Leung LR, Mcgregor JL, Lee D-K, Wang W-C, Ding Y, Kimura F (2004) Regional Climate Modeling: Progress, Challenges, and Prospects. J Meteor Soc Japan 82:1599–1628. https://doi.org/10.2151/jmsj.82.1599 Yamada TJ, Farukh MA, Fukushima T, Inatsu M, Sato T, Pokhrel YN, Oki T (2014) Extreme precipitation intensity in future climates associated with the Clausius-Clapeyron-like relationship. Hydrol Res Lett 8:108–113. https://doi.org/10.3178/hrl.8.108 Wu W, Lynch AH, Rivers A (2005) Estimating the uncertainty in a regional climate model related to initial and lateral boundary conditions. J Clim 18:917–933. https://doi.org/10.1175/JCLI-3293.1 Supplementary Files FigureA1.png FigureA2.png Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Major Revision 13 Oct, 2025 Reviewers agreed at journal 31 Aug, 2025 Reviewers invited by journal 26 Aug, 2025 Editor assigned by journal 09 Aug, 2025 First submitted to journal 04 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7266042","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":505997800,"identity":"bd427ac5-b4f4-482f-98fa-086174a764a7","order_by":0,"name":"Hajime Ohnishi","email":"","orcid":"","institution":"Hokkaido University","correspondingAuthor":false,"prefix":"","firstName":"Hajime","middleName":"","lastName":"Ohnishi","suffix":""},{"id":505997801,"identity":"9ef3858c-8ef4-46b3-87a3-5770b8d49143","order_by":1,"name":"Masaru Inatsu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5ElEQVRIiWNgGAWjYBACNhDxgYGBh0GCgRnE5QELSxDQwjiDJC0gwMwDUcUMtZQA4JNufvzZ5o+dDIN082FjnjIgg/3wAwbLHXgcJnPMwDi3LZmHQeZYcjLPOSCDJ82AQfIMHi0SCQbJuQ0HgH7JMT7M2wZyZA4Dg2QbPi3pHw5b/IFrqedh4H9DSEuOYTMDG0RLMm/bYRCDoJZixl6gX9gk0pIN55w7DmQ8MziAzy/yM9I3f/jxx86eXyL5sMSbsmp7fv7kh48l8YQYwjpkxmHJBiK0oADGjyRrGQWjYBSMgmEMAKPOPgf9K8V7AAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-1693-1024","institution":"Hokkaido University","correspondingAuthor":true,"prefix":"","firstName":"Masaru","middleName":"","lastName":"Inatsu","suffix":""},{"id":505997802,"identity":"b0da9217-2561-4f08-8872-4a7c6e2edc88","order_by":2,"name":"Ryo N. Matsuoka","email":"","orcid":"","institution":"Hokkaido University","correspondingAuthor":false,"prefix":"","firstName":"Ryo","middleName":"N.","lastName":"Matsuoka","suffix":""}],"badges":[],"createdAt":"2025-07-31 23:48:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7266042/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7266042/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":90532275,"identity":"dc0becb2-21e5-45c1-8be4-7876375a0cd5","added_by":"auto","created_at":"2025-09-03 18:46:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":268778,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Geographical map with land shaded including the target domain with (area enclosed by dotted lines) model domain in the DDS, (black solid line) analysis domain for SOM, and (red) Kyushu domain. (b) DDS surface height (m) in Kyushu domain with the color shading as in the bottom\u003c/p\u003e","description":"","filename":"Figure001.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/fbc577f6e6b0ce83400b4b70.png"},{"id":90532274,"identity":"ac6adef1-a9d9-4c41-8eca-fc2cf80788e6","added_by":"auto","created_at":"2025-09-03 18:46:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":38321,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of the LRDS procedure\u003c/p\u003e","description":"","filename":"Figure002.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/b3e405c43cbe8a6295b893a1.png"},{"id":90533731,"identity":"f0f567cf-a1f2-4e95-9de2-8a700f9da3da","added_by":"auto","created_at":"2025-09-03 19:10:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1908301,"visible":true,"origin":"","legend":"\u003cp\u003eThe SLP (hPa) in JJA projected onto the latent space created by the SOM algorithm with six ensembles of d4PDF historical experiment results combined with CMIP5 20th century experiment results for years 1951 to 2005 (Table 1) as the training data. The SOM node number was indicated at the top-left corner of each panel. The contour interval is 2 hPa with every 10 hPa contours thickened. The color shading indicates SLP anomaly as per the reference in the bottom\u003c/p\u003e","description":"","filename":"Figure003.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/712f89fb6673a6302bb831a9.png"},{"id":90532927,"identity":"6904dfc7-9a0c-4fcd-8362-3836ab2ed09d","added_by":"auto","created_at":"2025-09-03 18:54:07","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":435907,"visible":true,"origin":"","legend":"\u003cp\u003eThe number of samples fallen into each SOM node for (a) training data and (b) all d4PDF historical experiment results. The color shading is as per the reference in the bottom right\u003c/p\u003e","description":"","filename":"Figure004.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/02a493740c0fbadb1fc06524.png"},{"id":90532283,"identity":"dc451cb5-1919-4800-b485-e35d63bcca58","added_by":"auto","created_at":"2025-09-03 18:46:07","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":309900,"visible":true,"origin":"","legend":"\u003cp\u003e(a; Contour) Multi-model mean precipitation (mm/day) and (shading) its standard deviation among CMIP5 models in LRDS results in which samples were selected to match the PDF in the latent space for CMIP5 models. The contour interval is 2 mm/day and color shading is as per the reference in the bottom. (b) Multi-model mean of 99%-tile precipitation (mm/hour) and (shading) its standard deviation among CMIP5 models in LRDS results. The contour interval is 2 mm/hour and color shading is as per the reference in the bottom\u003c/p\u003e","description":"","filename":"Figure005.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/8f179d4a957c586b0b84516c.png"},{"id":90533237,"identity":"03293e74-759c-413b-ab2b-ad5641d7e7e5","added_by":"auto","created_at":"2025-09-03 19:02:08","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":71013,"visible":true,"origin":"","legend":"\u003cp\u003eDendrogram for PDF in the latent space based on present climate SLP data for d4PDF and CMIP5 models, classified into three groups denoted by red, green and blue by Ward method. CMIP5 models are numbered as in Table 1. The dotted line at 0.25 denotes the threshold separating the three groups\u003c/p\u003e","description":"","filename":"Figure006.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/d3e39c4728c106f1120c3d1f.png"},{"id":90533232,"identity":"a84cfb70-e62f-4360-91bf-4c75d21bcf46","added_by":"auto","created_at":"2025-09-03 19:02:07","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":407608,"visible":true,"origin":"","legend":"\u003cp\u003e(a-c; Contour) Climatological SLP for the 20th-century experiment results for CMIP5 models (a) #18, (b) #09, and (c) #25. The contour interval is 2 hPa with every 10 hPa contours thickened. The color shading indicates the deviation from d4PDF climatology as per the reference in the bottom of upper panels. (d-f) The relative frequency distribution (%) in the 10×10 latent space generated by the SOM algorithm for CMIP5 models (d) #18, (e) #09, and (f) #25. The color shading is as per the reference in the bottom of lower panels\u003c/p\u003e","description":"","filename":"Figure007.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/815ebd76c319cf6f5177f814.png"},{"id":90532285,"identity":"69ab8e66-f70f-4ead-8536-1d47358ff512","added_by":"auto","created_at":"2025-09-03 18:46:07","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":181018,"visible":true,"origin":"","legend":"\u003cp\u003e(a-c) Climatological precipitation (mm/day) in the LRDS results for CMIP5 models (a) #18, (b) #09, and (c) #25. The color shading is as per the reference of the bottom of upper panels\u003c/p\u003e","description":"","filename":"Figure008.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/f266bc45e3fcd2814f912806.png"},{"id":90532278,"identity":"c0e4195c-cdc4-445c-8e2a-c65974583363","added_by":"auto","created_at":"2025-09-03 18:46:07","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":1924426,"visible":true,"origin":"","legend":"\u003cp\u003eSame as Fig. 3, but for the SLP data combined equally from present and future climates\u003c/p\u003e","description":"","filename":"Figure009.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/6f2f050fce37f8e75854a96d.png"},{"id":90533732,"identity":"387a322a-535c-4e6f-8c31-a7bfb3f6762a","added_by":"auto","created_at":"2025-09-03 19:10:07","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":252919,"visible":true,"origin":"","legend":"\u003cp\u003e(a,b) The number of samples fallen into each SOM node for training data of (a) present and (b) future climates. The color shading is as per the reference in the bottom of (b). (c) The frequency difference (‰) between present and future climates in the latent space for d4PDF dataset. The color shading is as per the reference in the bottom. (d; Contour) Climatological SLP (hPa) in the d4PDF present climate simulation and (shading) its difference between future and present climates. Contour interval is 2 hPa and color shading is as per the reference in the bottom.\u003c/p\u003e","description":"","filename":"Figure010.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/bf163081d1807216ef5223f2.png"},{"id":90533234,"identity":"e55e98bd-3e64-432f-ae44-5a08144b01a6","added_by":"auto","created_at":"2025-09-03 19:02:07","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":475693,"visible":true,"origin":"","legend":"\u003cp\u003ea; Shade) Multi-model mean precipitation difference between present and future climates (mm/day) and (contour) its standard deviation based on LRDS results. The contour interval is 0.2 mm/day and color shading is as per the reference in the bottom. (b; Shade) Multi-model mean of 99%-tile precipitation difference (mm/hour) and (contour) its standard deviation. The contour interval is 0.02 mm/hour and color shading is as per the reference in the bottom\u003c/p\u003e","description":"","filename":"Figure011.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/45280a3b0a071fcfe386df27.png"},{"id":90532929,"identity":"d8217b56-a8ae-4181-8cbf-058d265194df","added_by":"auto","created_at":"2025-09-03 18:54:07","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":63731,"visible":true,"origin":"","legend":"\u003cp\u003eDendrogram for PDF in the latent space based on present and future climate SLP data for d4PDF and CMIP5 models, classified into two groups denoted by red and blue. CMIP5 models are numbered as in Table 1. The dotted line at 0.15 denotes the threshold\u003c/p\u003e","description":"","filename":"Figure012.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/b8a43acda6920e580eb3dfaf.png"},{"id":90533236,"identity":"5f0a03ec-b195-4a19-b841-f93c659ecd41","added_by":"auto","created_at":"2025-09-03 19:02:07","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":367489,"visible":true,"origin":"","legend":"\u003cp\u003e(a,b) Climatological rainfall difference (mm/day) between present and future climates based on the LRDS results for CMIP5 models (a) #10 and (b) #41. The color shading is as per the reference in the bottom. (c,d; contour) Climatological SLP (hPa) in the present climate and (shading) the difference between present and future climates based on the LRDS results for CMIP5 models (a) #10 and (b) #41. The contour interval is 2 hPaand color shading is as per the reference of the bottom\u003c/p\u003e","description":"","filename":"Figure013.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/574aa2afd2a08822e8cd9a9f.png"},{"id":90533955,"identity":"e5d05aba-9bfe-436f-948c-2cbe257ce7ec","added_by":"auto","created_at":"2025-09-03 19:18:12","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5850808,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/8a39feb9-64fb-49bc-8c9c-4b4d698897d7.pdf"},{"id":90532937,"identity":"968281b9-63b7-4278-8169-4384487a4499","added_by":"auto","created_at":"2025-09-03 18:54:08","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"supplement","size":254584,"visible":true,"origin":"","legend":"","description":"","filename":"FigureA1.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/8171f3e0c37aba10f07c59c1.png"},{"id":90532297,"identity":"e8e7b92c-3607-4fde-b468-f916ae005a66","added_by":"auto","created_at":"2025-09-03 18:46:08","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"supplement","size":262365,"visible":true,"origin":"","legend":"","description":"","filename":"FigureA2.png","url":"https://assets-eu.researchsquare.com/files/rs-7266042/v1/47dfe9b933586af236fc7445.png"}],"financialInterests":"","formattedTitle":"Development of latent resampling downscaling and its application to model bias and climate change projection","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eDynamical downscaling (DDS) refers to a regional atmospheric model (RAM) simulation applied over a limited area, using outputs from general circulation models (GCMs) as lateral boundary conditions (Giorgi and Bates \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Wang et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Giorgi \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). This technique has been widely used to compensate for the insufficient spatial resolution of GCMs in applications that require climate change information at approximately O(10 km) resolution. However, DDS results are strongly influenced by the lateral boundary conditions derived from GCMs. Consequently, DDS outputs often inherit biases from the driving GCMs and introduce additional biases from the RAMs themselves (McSweeney et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Wu et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). Therefore, regional climate change projections should account for both uncertainties arising from GCMs and those associated with RAMs. This consideration has motivated multi-GCM and multi-RAM experiments to evaluate model uncertainty in regional climate in recently coordinated projects, such as the ENSEMBLES project (D\u0026eacute;qu\u0026eacute; et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), NARCCAP (Mearns et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), CORDEX-SEA (Tangang et al. 2018), and the Japanese RECCA project (Inatsu et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eApplying DDS to all possible combinations of GCMs, RAMs, and emission scenarios would be impractical, especially considering that the coupled model intercomparison project phase5 (CMIP5; Taylor et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2012\u003c/span\u003e) involved 30 to 40 GCMs and included multiple representative concentration pathways (RCPs; Moss et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Consequently, even in many recent projects that coordinated multi-GCM and multi-RAM experiments, the matrix of GCM\u0026ndash;RAM combinations was mostly sparse. Furthermore, large-ensemble GCM experiments and their associated DDS simulations are essential for probabilistic risk assessments of natural hazards related to extreme weather events. Thus, the combination of multi-model and large-ensemble experiments poses a significant challenge to assessing model uncertainty in climate change impacts on such events through DDS.\u003c/p\u003e\u003cp\u003eRecently, Mizuta et al. (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) developed a large-ensemble dataset called d4PDF, which consists of GCM simulations and associated DDS results for the area around Japan under various climate conditions. The ensemble includes 3,000 years of simulations under 20th-century historical conditions and 5,400 years under a future climate condition with a global mean temperature increase of 4 K. This exceptionally large ensemble size allowed the assumption that d4PDF encompasses nearly the full range of internal variability under both current and future climate conditions. Therefore, unless the inter-model discrepancies are extremely large, it is feasible to approximate any GCM outputs by subsampling from d4PDF. This insight motivated us to assess model uncertainty in regional climate by selecting appropriate subsets from such a large-ensemble dataset, comprising a GCM and its associated DDS for the target region\u0026mdash;such as d4PDF.\u003c/p\u003e\u003cp\u003eThe objective of this paper is to develop a method for estimating the downscaled results of any GCM outputs using a reference large-ensemble simulation without performing additional DDS computations. Assuming that any GCM results are encompassed within d4PDF result, we approximate a particular GCM result as a subset of d4PDF by focusing on the probability density function (PDF) of a specific climatic variable over the targeted area. The statistics of the DDS data for this d4PDF subset can then be regarded as the surrogated DDS data for the GCM. We named this method latent resampling downscaling (LRDS). In this study, we demonstrated the LRDS results focusing on summer precipitation in Kyushu. This paper is organized as follows: Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e discusses the data used in this study. Section \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003e3\u003c/span\u003e explains the procedures of LRDS. Sections \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e5\u003c/span\u003e present the LRDS results for present and future climates, respectively. Finally, Section \u003cspan refid=\"Sec16\" class=\"InternalRef\"\u003e6\u003c/span\u003e provides a summary with discussion on the results.\u003c/p\u003e"},{"header":"2. Data","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1. d4PDF\u003c/h2\u003e\u003cp\u003eWe used the d4PDF dataset as the reference for LRDS. The d4PDF consists of a GCM experiment with a horizontal resolution of 60 km and a DDS experiment with a horizontal resolution of 20 km for a limited area around Japan (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). The d4PDF experiments were conducted under several climate conditions. Historical experiments were coordinated as the time integration from various initial conditions with prescribed sea surface temperatures (SSTs) and other factors as observed from 1950 to 2010. The 4K experiments were also conducted, in which the detrended SST from 1950 to 2010 was increased to achieve a global mean temperature rise of 4K by scaling the value based on model\u0026rsquo;s climate sensitivity. The 4K experiment approximately corresponds to the period around 2090 of the RCP8.5 scenario experiment in the CMIP5.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eWe here analyzed 50 ensemble members from the historical experiments and 90 ensemble members from the 4K experiment, in which GCM experiment and the associated DDS experiments were both available. Although the SST distributions in the 4K experiments had six different warming patterns to incorporate climate model uncertainties, this study did not take those uncertainties into account. The DDS experiments associated with the GCM provided outputs by integrating a regional atmospheric model with the GCM results imposed as lateral boundary conditions. In this study, we focused on precipitation over Kyushu Island (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb) in June, July, and August (JJA).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2. CMIP5\u003c/h2\u003e\u003cp\u003eThe CMIP5 ensemble data was used to demonstrate the LRDS results. We utilized 41 models from the 20th-century historical experiment and 33 models from the RCP8.5 scenario experiment (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Because CMIP5 had a different grid system from d4PDF, the CMIP5 data were linearly interpolated to match the grid points of d4PDF. Here, the period from 1951 to 2005 in the historical experiment corresponded to d4PDF historical experiment, and the period from 2081 to 2100 in the RCP8.5 scenario experiment corresponded to d4PDF 4K experiment.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSpecifications of the CMIP5 models used in this study\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003e#\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePresent climate\u003c/p\u003e\u003cp\u003e(days)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFuture climate\u003c/p\u003e\u003cp\u003e(days)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eACCESS1-0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eACCESS1-3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBNU-ESM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCCSM4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCESM1-BGC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCESM1-CAM5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCESM1-FASTCHEM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCMCC-CESM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCMCC-CM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCMCC-CMS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCNRM-CM5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCSIRO-Mk3-6-0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCSIRO-Mk3L-1-2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e14\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCanCM4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4,140\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e15\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCanESM2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEC-EARTH\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e17\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFGOALS-g2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFGOALS-s2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGFDL-CM3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGFDL-ESM2G\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGFDL-ESM2M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e22\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGISS-E2-H\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e23\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGISS-E2-R\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e24\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHadCM3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4,950\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHadGEM2-AO\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4,950\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,800\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e26\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHadGEM2-CC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4,950\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,800\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e27\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHadGEM2-ES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4,950\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,800\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e28\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIPSL-CM5A-LR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIPSL-CM5A-MR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e30\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIPSL-CM5B-LR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e31\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMIROC-ESM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMIROC-ESM-CHEM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMIROC4h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e34\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMIROC5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMPI-ESM-LR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e36\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMPI-ESM-MR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e37\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMPI-ESM-P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e38\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMRI-CGCM3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e39\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMRI-ESM1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e40\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNorESM1-M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e41\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ebcc-csm1-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e42\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ebcc-csm1-1-m\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003einmcm4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5,060\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1,840\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Method","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e3.1. LRDS Procedure\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrates the procedure of LRDS. We assumed that a large-ensemble GCM dataset and its associated DDS dataset for a specific limited region were already provided as the reference. It was also assumed that the target dataset only contained GCM output with a smaller sample size than the reference and its associated DDS has not been performed. LRDS proposed a method, in such a case, to approximate the statistics of the associated DDS results without requiring additional DDS. LRDS involved two processes: PDF matching between reference GCM and target GCM data (the arrow from right to left in the upper level of Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), and DDS subsampling from the reference data (the arrow from left to right in the lower level). We here demonstrated the LRDS with the reference dataset d4PDF and the target datasets CMIP5.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe PDF matching, the first half of the procedure, was conducted by comparing the reference and target GCM data on the latent space generated by the self-organizing map (SOM) algorithm (Kohonen, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1982\u003c/span\u003e). We here applied the algorithm to the gridded data of sea level pressures (SLPs) in the limited domain bounded by 20\u0026deg;N\u0026ndash;50\u0026deg;N and 120\u0026deg;E\u0026ndash;150\u0026deg;E (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). Moreover, the SOM configuration was set to 10\u0026times;10 in order to adequately represent the various weather patterns in both the reference and target GCM data. Given the larger sample size of the reference dataset, its subset was sufficient to reproduce the PDF of the target data by randomly sampling reference data assigned to each SOM node. Thus, the PDF matching enabled us to subsample the reference dataset in correspondence with the target dataset.\u003c/p\u003e\u003cp\u003eIn the second half of the procedure, by utilizing the linkage between the GCM and DDS in the reference data, the subsampling obtained through the PDF matching can be directly transferred to the DDS data. This approach assumes that regional phenomena represented in the DDS data are closely associated with large-scale atmospheric patterns captured in the SLP-based SOM. The DDS data corresponding to the target dataset could then be generated under this assumption. Based on the subsampled DDS data, we here created the statistics for the target dataset related to summertime rainfall over Kyushu (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e3.2. PDF matching\u003c/h2\u003e\u003cp\u003eWe first constructed a latent space using SOM, based on training SLP data sampled every 24 hours from six ensemble members of the d4PDF dataset and a single ensemble from each CMIP5 model. For the analysis of the present climate, a slice of latent space was created using daily data from JJA months, combining six ensembles from the d4PDF historical experiment with 20th-century experiments from 41 CMIP5 models (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Since each d4PDF ensemble comprised 60 years of data, the combined d4PDF dataset totaled 33,120 days. For CMIP5, we utilized the full range of the 20th-century experiment from 1951 to 2005, resulting in approximately 5,040 days per model (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Exceptions included CanCM4 (model #14) that provided only 45 years of data after 1961 and the Hadley Centre model series (models #24\u0026ndash;#27) that adopted a 360-day calendar. Multi-model JJA-mean patterns were computed for the present climate, and the principal component (PC) analysis was conducted using SLP anomalies, defined as deviations from the mean SLP.\u003c/p\u003e\u003cp\u003eFor the future climate analysis, another slice of latent space was created using daily JJA data from both present and future climate conditions. The dataset included six ensembles from d4PDF historical experiments (33,120 days in total) and six ensembles from d4PDF 4K experiments, each imposing different SST patterns (also totaling 33,120 days). Additionally, we included outputs from 33 CMIP5 model: 20 years of data (1981 to 2000) from the 20th-century experiments and 20 years of data (2081 to 2100) from RCP8.5 experiments. These periods provided 1,840 days of data per experiment per model, except for models #25\u0026ndash;#27 that adopted a 360-day calendar (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Multi-model JJA-mean patterns were calculated for both present and future climates, and the PC analysis was again conducted using SLP anomalies.\u003c/p\u003e\u003cp\u003eFor computational efficiency, the SOM was trained using the projection of the SLP anomalies onto the first 60 PC modes over the limited domain (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). All non-training data were subsequently projected onto the latent space by assigning each timestep to the SOM node whose SLP pattern was closest in Euclidean distance. As a result, every timestep in the 6-hourly d4PDF and daily CMIP5 datasets was associated with a specific SOM node, uniquely identifying the timestamp, ensemble member, and model.\u003c/p\u003e\u003cp\u003eThe PDF on the latent space was proportional to the number of timestamps assigned to each SOM node over 10\u0026times;10 SOM nodes. A model-specific PDF was also constructed using all available data from each model. In the present climate analysis, 6-hourly data from the d4PDF historical experiment were randomly sampled without duplication to match the PDF of a CMIP5 model\u0026rsquo;s 20th century experiment on the present-climate latent space, with a total of 10,000 samples selected. In future climate analysis, using the latent space created from the combined present and future climate datasets, the 6-hourly data from the d4PDF historical experiment were randomly sampled to match the PDF of a CMIP5 model\u0026rsquo;s 20th century experiment, and the 6-hourly data from the d4PDF 4K experiment were randomly sampled to match the PDF of the same model\u0026rsquo;s RCP 8.5 experiment. The sampled number was 10,000 as well.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e3.3. DDS subsampling\u003c/h2\u003e\u003cp\u003eThe DDS subsampling was performed based on a sampling set for each CMIP5 model, which included 10,000 discontinuous timestamps and ensemble members. The GCM sampling had a 6-hourly interval, whereas the DDS model output was available at an hourly interval. The d4PDF DDS results were then sliced within a three-hour window before and after each timestamp in the GCM sampling. In this paper, based on this subsampled data, we took the time average to check the climatological precipitation and calculated the 99th percentile values in hourly precipitation intensity to demonstrate the extreme precipitation. It should be noted that timeseries analysis could not be conducted by LRDS due to the discontinuous nature of the subsampled data.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.4. Storyline-based approach\u003c/h2\u003e\u003cp\u003eWe employed a storyline-based approach, following the framework proposed by Shepherd (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), to illustrate representative patterns emerging from the LRDS results across multiple models. In the present-climate analysis, we identified three typical patterns among 41 CMIP5 models (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) by classifying their PDFs in the SOM latent space, based on SLP anomalies over the target domain (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). The classification was performed using Ward\u0026rsquo;s method (Matsuoka and Inatsu, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and an arbitrary threshold was applied to divide the models into three groups with similar PDFs in the latent space. A representative model for each group was selected as the one whose PDF had the smallest Kullback\u0026ndash;Leibler divergence from the reference PDF defined as the group\u0026rsquo;s mean. These three storylines reflected characteristic patterns in the models' ability to reproduce synoptic-scale environments under present climate conditions.\u003c/p\u003e\u003cp\u003eIn the future climate analysis, we identified two representative patterns among 33 CMIP5 models (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) by classifying the differences in their PDFs between present and future climates on the SOM latent space. The subsequent procedures followed those used in the present climate analysis. A representative model for each group was selected as the one whose PDF difference between the future and present climates had the smallest root-mean-square difference from the group-mean PDF difference, which served as the reference. As a result, we derived two storylines that represented characteristic model responses to climate change.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Present climate analysis","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e4.1 LRDS results\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates the SLP patterns projected onto the 10\u0026times;10 latent space generated by the SOM algorithm (Section \u003cspan refid=\"Sec6\" class=\"InternalRef\"\u003e3.1\u003c/span\u003e) based on the training data that combined six ensemble members from d4PDF historical experiment (Section \u003cspan refid=\"Sec3\" class=\"InternalRef\"\u003e2.1\u003c/span\u003e) with CMIP5 20th-century experiment for years 1951 to 2005 (Section \u003cspan refid=\"Sec4\" class=\"InternalRef\"\u003e2.2\u003c/span\u003e). The SOM nodes captured typical weather patterns in boreal summer around Japan: a westward extension of the Bonin high (upper left nodes), intense low-pressure anomalies southwest of Japan (lower right nodes), and weak low-pressure anomalies zonally elongated over Japan (middle right nodes).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe SOM algorithm uniformly mapped various SLP patterns in training d4PDF and CMIP5 data into the latent space. The sample counts exhibited a relatively uniform distribution across nodes, ranging from 1,659 at node #99 to 3,310 at node #02 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). Because most of CMIP5 models deviated from the training data, their SOM projections exhibited non-uniform distribution (Fig. \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003eA1\u003c/span\u003e). Even the d4PDF data showed a non-uniform PDF in the latent space (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe LRDS was accomplished by DDS subsampling (Section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e3.3\u003c/span\u003e), in which the d4PDF DDS outputs were assembled for timeframes corresponding to GCM data subsampled in the PDF matching process. This procedure was applied to each CMIP5 model output for the present climate analysis. The multi-model mean rainfall (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003ea) estimated by the LRDS almost ranged approximately from 8 to 13 mm/day over Kyushu, with local maxima exceeding 12 mm/day on the eastern and western sides of the island. Model-to-model variation was particularly large around these local maxima of mean rainfall.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIn contrast, the extreme rainfall evaluated by the 99%-tile value showed a more horizontally uniform distribution, ranging from 12 to 16 mm/hour (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). The inter-model standard deviation was around 1.0 mm/hour, considerably smaller than the multi-model mean. As above, the LRDS results showed little variation in the 99%-tile rainfall over Kyushu, presumably because extreme rainfall events were not primarily controlled by large-scale weather patterns. Rather, they were typically triggered by random processes under background conditions with abundant moisture in the boundary layer.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e4.2. Storylines\u003c/h2\u003e\u003cp\u003eThe storyline-based approach effectively streamlined the interpretation of LRDS results across 41 CMIP5 model outputs. By applying the Ward clustering method to the 41 PDFs on the latent space (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003e), we were able to classify the models and select representative ones. An arbitrary threshold of 0.25 was used to divide the models into three distinct groups, as visualized in the dendrogram (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003e), which represented model-to-model similarity in PDF distributions. Based on this classification, models #18, #09, and #25 (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) were selected as three representative storylines, each capturing typical features of SLP anomalies around Japan.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eBy retrieving the mean SLP fields for these three CMIP models, we identified distinct differences in summertime weather patterns around Japan (Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ea-c), as well as in the associated PDFs in the latent space (Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ed-f). CMIP5 model #18 exhibited relatively higher probabilities in the upper left side of the latent space (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ed), corresponding to a more frequent westward extension of the Bonin high compared to d4PDF (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ea). CMIP5 model #25 showed a mean SLP field characterized by a low-pressure anomaly over the subtropical ocean (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ec), which corresponded to the SOM nodes in the lower right corner (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ef). CMIP5 model #09, which represented the intermediate group including d4PDF, displayed a relatively uniform distribution in the latent space (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ee) and showed less deviation from d4PDF (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003eb).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003e showed the LRDS results for JJA-mean rainfall over Kyushu, along with a storyline-based approach. Model #18 exhibited lower JJA-mean rainfall over eastern Kyushu (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003ea) compared to model #25 (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003ec). The model #09 showed an intermediate pattern. These differences in local rainfall among the models were likely linked to the JJA-mean SLP fields in CMIP5 GCMs. Typhoon passages associated with weather patterns located in the lower right of the SOM map (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e) substantially contributed to rainfall events in eastern Kyushu. Therefore, the GCM tendency toward more frequent tropical cyclone occurrence (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003ec) was associated with heavier mean rainfall in this region (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003ec).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Future climate analysis","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e5.1. LRDS results\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e9\u003c/span\u003e displays the SLP patterns in the latent space based on the training data, in which present climate data were equally combined with future climate data (Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e). The SOM nodes captured typical weather patterns in boreal summer around Japan, quite similar to the present climate analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Therefore, the conditional PDF for present climate data (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e10\u003c/span\u003ea) somewhat resembled the PDF in present climate analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). The conditional PDF for future climate data (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e10\u003c/span\u003eb) was also uniform in the latent space but was slightly different from that for present climate data. Limited to the d4PDF dataset, the PDF difference between present and future climates (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e10\u003c/span\u003ec) ranged\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4%, with an increase in probability around central nodes and a decrease in probability around lower nodes. This feature of PDF difference was consistent with the climatological SLP response to climate change in d4PDF dataset (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e10\u003c/span\u003ed; Takabatake and Inatsu \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Kawase et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e): the north-to-south SLP contrast was more prominent in future climate conditions.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe LRDS was conducted by applying DDS subsampling through a PDF matching process for both the present and future climates. Based on this, we therefore estimated the multi-model mean rainfall for present and future climates. Figure\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e11\u003c/span\u003ea displays their difference in climatological rainfall over Kyushu. In the sense of multi-model mean, the rainfall increased by ~\u0026thinsp;3 mm/day over western Kyushu and ~\u0026thinsp;1 mm/day in eastern Kyushu. The inter-model variation was not negligible around these two peak regions. In contrast, the extreme rainfall clearly increased by ~\u0026thinsp;1.2 mm/hour all over Kyushu with small intermodal variations (Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e11\u003c/span\u003eb).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e5.2. Storylines\u003c/h2\u003e\u003cp\u003eThe storyline-based approach was also applied to the PDF difference between present and future climates across 33 CMIP5 model outputs. The Ward clustering method with an arbitrary threshold of 0.15 allowed us to divide the models into two distinct groups, as visualized in the dendrogram (Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e12\u003c/span\u003e). Based on this classification, models #10 and #41 (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) were selected as two representative storylines, each capturing typical features of SLP responses to climate change around Japan.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eModel #10 exhibited an increase of approximately 2 mm/day in eastern Kyushu and a decrease of about 1 mm/day in northern Kyushu (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e13\u003c/span\u003ea). This rainfall response was consistent with a climatological cyclone tendency to the south and southeast of Japan (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e13\u003c/span\u003eb), suggesting an increase in typhoon passages in that region. In contrast, model #41 showed an increase of around 3 mm/day in western Kyushu and about 2 mm/day in southern Kyushu (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e13\u003c/span\u003ec). Climate change in this model strengthened two anticyclones located to the north and south of Japan (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e13\u003c/span\u003ed), leading to an enhancement of the stagnant low-pressure region. This circulation pattern was generally associated with heavy rainfall events, particularly in western Kyushu (Hirockawa et al. 2021).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"6. Concluding Remarks","content":"\u003cp\u003eIn this study, we developed LRDS to reduce the computational cost of DDS for climate projection data. LRDS estimates the results of DDS for arbitrary climate models based on the PDF in a latent space, utilizing a large-ensemble experiment of a single climate model and its DDS that has already been conducted. We applied LRDS to summer precipitation over Kyushu under both present and future climate conditions, generating high-resolution data for the target domain corresponding to multiple GCMs. For summer precipitation in Kyushu under the current climate, we identified three distinct spatial distribution patterns: one with a precipitation maximum in the west, another with a maximum in the east, and a third exhibiting intermediate features between the two. Under future climate projections, substantial inter-model variation emerged around the peak precipitation regions in both the west and east. One model projected a greater increase in precipitation in the west, associated with a relatively low SLP response along the Japanese Islands. In contrast, another model projected an increase in precipitation in the east, linked to a low SLP response south of Japan\u0026mdash;suggesting more frequent typhoon passages in that region.\u003c/p\u003e\u003cp\u003eTo validate the results obtained in this study, a direct comparison with actual DDS outputs would be necessary. However, the primary motivation for proposing LRDS was to avoid the high computational cost of DDS itself. Therefore, performing DDS solely for validation purposes would be impractical, though a storyline-based approach may offer some mitigation of this difficulty. As an alternative solution, we propose dividing the d4PDF ensemble into two groups. The first group includes both large-ensemble simulations and their corresponding DDS outputs, referred to as d4PDF in this study. The second group consists only of global outputs from multiple climate models, referred to here as CMIP5. By applying LRDS to the latter dataset, we can estimate downscaling statistics and compare them with the actual DDS results from the former group. This approach enables an indirect validation of LRDS.\u003c/p\u003e\u003cp\u003eIt should be noted when interpreting the results that extreme precipitation showed less model uncertainty in its climate change response. Since the LRDS in this study employed a single regional atmospheric model for DDS, the results did not account for uncertainties associated with different regional models. In general, the simulation of extreme precipitation is sensitive to model resolution and physical parameterizations. Ban et al. (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) demonstrated that kilometer-scale models tended to produce more intense precipitation and fewer wet hours compared to 10-km scale models. Furthermore, Supari et al. (2020) reported an increase in extreme precipitation under climate change in the Asian monsoon region, although the uncertainty arising from regional models was not explicitly discussed. These previous studies as well as our LRDS results suggested that extreme precipitation might be more sensitive to the increase in atmospheric moisture associated with warming, as governed by the Clausius\u0026ndash;Clapeyron relationship (Yamada et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), rather than to changes in synoptic-scale weather systems. Extreme precipitation is often caused by cumulonimbus clouds embedded within mesoscale convective systems, though 10-km resolution models often generate extreme precipitation caused by orographic uplifting. Whatever the underlying reason, extreme precipitation may be interpreted as a stochastic outcome within a fundamentally moist environment. The regional model uncertainty in extreme precipitation could then be smaller than that of mean precipitation. This implication might be validated by kilometer-scale model ensemble simulation in the future.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eMI is supported by the Environment Research and Technology Development Fund JPMEERF20232003 of the Environmental Restoration and Conservation Agency provided by Ministry of the Environment of Japan and by JPMXD0722680734 of MEXT, Japan.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e\u003cp\u003eWe would like to thank Prof. Yousuke Sato and Prof. Tomohito J. Yamada, who give us many insightful comments for our earlier studies. This study used d4PDF produced with the Earth Simulator jointly by science programs (SOUSEI, TOUGOU, SI-CAT, DIAS) of the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), Japan. We acknowledge the World Climate Research Programme\u0026rsquo;s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table\u0026nbsp;1 of this paper) for producing and making available their model output. For CMIP the U.S. Department of Energy\u0026rsquo;s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals.\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e\u003cp\u003eThe CMIP5 outputs can be obtained from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pcmdi.llnl.gov/mips/cmip5/data-portal.html\u003c/span\u003e\u003cspan address=\"https://pcmdi.llnl.gov/mips/cmip5/data-portal.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. The d4PDF outputs can be obtained from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://search.diasjp.net/en/dataset/d4PDF_GCM\u003c/span\u003e\u003cspan address=\"https://search.diasjp.net/en/dataset/d4PDF_GCM\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBan N, Brisson E, Caillaud C, Coppola E, Pichelli E, Sobolowski S, Adinolfi M, Ahrens B, Alias A, Anders I, Bastin S, Belušić D, Berthou S, Brisson E, Cardoso RM, Chan SC, Christensen OB, Fern\u0026aacute;ndez J, Fita L, Frisius T, Gašparac G, Girogi F, Goergen K, Haugen JE, Hodnebrog \u0026Oslash;, Kartsios S, Katragkou E, Kendon EJ, Keuler K, Lavin-Gullon A, Lenderink G, Leutwyler D, Lorenz T, Maraun D, Mercogliano P, Milovac J, Panitz H-J, Raffa M, Remedio AR, Sch\u0026auml;r C, Soares PMM, Srnec L, Steensen BM, Stocchi P, T\u0026ouml;lle MH, Truhetz H, Vergara-Temprado J, de Vries H, Warrach-Sagi K, Wulfmeyer V, Zander MJ (2021) The first multi-model ensemble of regional climate simulations at the kilometer-scale resolution, Part I: Evaluation of precipitation. Clim Dyn 57:275\u0026ndash;302. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00382-021-05708-w\u003c/span\u003e\u003cspan address=\"10.1007/s00382-021-05708-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eD\u0026eacute;qu\u0026eacute; M, Somot S, Sanchez-Gomez E, Goodess CM, Jacob D, Lenderink G, Christensen OB (2012) The spread among ENSEMBLES regional scenarios: Regional climate models, driving general circulation models and interannual variability. Clim Dyn 38:951\u0026ndash;964. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00382-011-1053-x\u003c/span\u003e\u003cspan address=\"10.1007/s00382-011-1053-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGiorgi F, Bates GT (1989) The climatological skill of a regional model over complex terrain. Mon Wea Rev 117:2325\u0026ndash;2347. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1175/1520-0493(1989)117\u0026lt;2325:TCSOAR\u0026gt;2.0.CO;2\u003c/span\u003e\u003cspan address=\"10.1175/1520-0493(1989)117%3C2325:TCSOAR%3E2.0.CO;2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGiorgi F (2019) Thirty years of regional climate modeling: Where are we and where are we going next? J Geophys Res Atmos 124:5696\u0026ndash;5723. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1029/2018JD030094\u003c/span\u003e\u003cspan address=\"10.1029/2018JD030094\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHawkins E, Ortega P, Suckling E, Schurer A, Hegerl G, Jones P, Joshi M, Osborn TJ, Masson-Delmotte V, Mignot J, Thorne P, van Oldenborgh GJ (2017) Estimating Changes in Global Temperature since the Preindustrial Period. Bull Amer Meteor Soc 98:1841\u0026ndash;1856. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1175/BAMS-D-16-0007.1\u003c/span\u003e\u003cspan address=\"10.1175/BAMS-D-16-0007.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHirockawa Y, Kato T, Tsuruti H, Seino N (2020) Identification and classification of heavy rainfall areas and their characteristic features in Japan. J Meteor Soc Japan 98:835\u0026ndash;857. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2151/jmsj.2020-043\u003c/span\u003e\u003cspan address=\"10.2151/jmsj.2020-043\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIPCC (2023) Sections. In: Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J. Romero (eds.)]. IPCC Geneva Switzerland 35\u0026ndash;115. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.59327/IPCC/AR6-9789291691647\u003c/span\u003e\u003cspan address=\"10.59327/IPCC/AR6-9789291691647\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eInatsu M, Sato T, Yamada TJ, Kuno R, Sugimoto S, Farukh MA, Pokhrel YN, Kure S (2015) Multi-GCM by multi-RAM experiments for dynamical downscaling on summertime climate change in Hokkaido. Atmos Sci Lett 16:297\u0026ndash;304. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/asl2.557\u003c/span\u003e\u003cspan address=\"10.1002/asl2.557\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKawase H, Imada Y, Sasaki H, Nakaegawa T, Murata A, Nosaka M, Takayabu I (2019) Contribution of historical global warming to local-scale heavy precipitation in western Japan estimated by large ensemble high-resolution simulations. J Geophys Res 124:6093\u0026ndash;6103. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1029/2018JD030155\u003c/span\u003e\u003cspan address=\"10.1029/2018JD030155\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKohonen T (1982) Self-organized formation of topologically correct feature maps. Biol Cybern 43:59\u0026ndash;69. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/BF00337288\u003c/span\u003e\u003cspan address=\"10.1007/BF00337288\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMatsuoka RN, Inatsu M (2024) Weather classifications for high temperatures in Japanese cities. SOLA 20:298\u0026ndash;305. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2151/sola.2024-040\u003c/span\u003e\u003cspan address=\"10.2151/sola.2024-040\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMearns LO, Arritt R, Biner S, Bukovsky MS, McGinnis S, Sain S, Caya SD, Correia J Jr, Flory D, Gutowski W, Takle ES, Jones R, Leung R, Moufouma-Okia W, McDaniel L, Nunes AMB, Quan Y, Roads J, Sloan L, Snyder M (2012) The North American regional climate change assessment program: Overview of phase I results. Bull Amer Meteor Soc 93:1337\u0026ndash;1362. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1175/BAMS-D-11-00223.1\u003c/span\u003e\u003cspan address=\"10.1175/BAMS-D-11-00223.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcSweeney C, Jones R, Lee R, Powell D (2015) Selecting CMIP5 GCMs for downscaling over multiple regions. Clim Dyn 44:3237\u0026ndash;3260. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00382-011-1278-8\u003c/span\u003e\u003cspan address=\"10.1007/s00382-011-1278-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMizuta R, Murata A, Ishii M, Shiogama H, Hibino K, Mori N, Arakawa O, Imada Y, Yoshida K, Aoyagi T, Kawase H, Mori M, Okada Y, Shimura T, Nagatomo T, Ikeda M, Endo H, Nosaka M, Arai M, Takahashi C, Tanaka K, Takemi T, Tachikawa Y, Temur K, Kamae Y, Watanabe M, Sasaki H, Kitoh A, Takayabu I, Nakakita E, Kimoto M (2017) Over 5,000 years of ensemble future climate simulations by 60-km global and 20-km regional atmospheric models. Bull Amer Meteor Soc 98:1383\u0026ndash;1398. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1175/BAMS-D-16-0099.1\u003c/span\u003e\u003cspan address=\"10.1175/BAMS-D-16-0099.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoss RH, Edmonds JA, Hibbard KA, Manning MR, Rose SK, van Vuuren DP, Carter TR, Emori S, Kainuma M, Kram T, Meehl GA, Mitchell JFB, Nakicenovic N, Riahi K, Smith SJ, Stouffer RJ, Thomson AM, Weyant JP, Wilbanks TJ (2010) The next generation of scenarios for climate change research and assessment. Nature 463:747\u0026ndash;756. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nature08823\u003c/span\u003e\u003cspan address=\"10.1038/nature08823\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShepherd TG (2019) Storyline approach to the construction of regional climate change information. Proc R Soc A 475:20190013. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://dx.doi.org/10.1098/rspa.2019.0013\u003c/span\u003e\u003cspan address=\"10.1098/rspa.2019.0013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSupari TF, Juneng L, Cruz F, Chung JX, Ngai ST, Salimun E, Mohd MSF, Santisirisomboon J, Singhruck P, PhanVan T, Ngo-Duc T, Narisma G, Aldrian E, Gunawan D, Sopaheluwakan A (2021) Multi-model projections of precipitation extremes in Southeast Asia based on CORDEX-Southeast Asia simulations. Environ Res 184:109350. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envres.2020.109350\u003c/span\u003e\u003cspan address=\"10.1016/j.envres.2020.109350\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTakabatake D, Inatsu M (2021) Summertime precipitation in Hokkaido and Kyushu, Japan in response to global warming. Clim Dyn 58:1671\u0026ndash;1682. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00382-021-05983-7\u003c/span\u003e\u003cspan address=\"10.1007/s00382-021-05983-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTangang F, Chung JX, Juneng L, Supari, Salinun E, Ngai ST, Jamaluddin AF, Mohd MSF, Cruz F, Narisma G, Santisirisomboon J, Ngo-Duc T, Tan PV, Singhruck P, Gunawan D, Aldrian E, Sopaheluwakan A, Grigory N, Remedio ARC, Sein DV, Hein-Griggs D, McGregor JL, Yang H, Sasaki H, Kumar P (2020) Projected future changes in rainfall in Southeast Asia based on CORDEX\u0026ndash;SEA multi\u0026ndash;model simulations. Clim Dyn 55:1247\u0026ndash;1267. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00382-020-05322-2\u003c/span\u003e\u003cspan address=\"10.1007/s00382-020-05322-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTaylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Amer Meteor Soc 93:485\u0026ndash;498. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1175/BAMS-D-11-00094.1\u003c/span\u003e\u003cspan address=\"10.1175/BAMS-D-11-00094.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Y, Leung LR, Mcgregor JL, Lee D-K, Wang W-C, Ding Y, Kimura F (2004) Regional Climate Modeling: Progress, Challenges, and Prospects. J Meteor Soc Japan 82:1599\u0026ndash;1628. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2151/jmsj.82.1599\u003c/span\u003e\u003cspan address=\"10.2151/jmsj.82.1599\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYamada TJ, Farukh MA, Fukushima T, Inatsu M, Sato T, Pokhrel YN, Oki T (2014) Extreme precipitation intensity in future climates associated with the Clausius-Clapeyron-like relationship. Hydrol Res Lett 8:108\u0026ndash;113. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3178/hrl.8.108\u003c/span\u003e\u003cspan address=\"10.3178/hrl.8.108\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu W, Lynch AH, Rivers A (2005) Estimating the uncertainty in a regional climate model related to initial and lateral boundary conditions. J Clim 18:917\u0026ndash;933. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1175/JCLI-3293.1\u003c/span\u003e\u003cspan address=\"10.1175/JCLI-3293.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"climate-dynamics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"cldy","sideBox":"Learn more about [Climate Dynamics](https://www.springer.com/journal/382)","snPcode":"382","submissionUrl":"https://submission.nature.com/new-submission/382/3","title":"Climate Dynamics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Dynamical downscaling, Regional climate change, Large ensemble datasets, Self-organizing map","lastPublishedDoi":"10.21203/rs.3.rs-7266042/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7266042/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWe have developed a cost-effective downscaling technique called latent resampling downscaling (LRDS) to address model uncertainty in regional climate change assessments. In this study, the LRDS was used to generate a surrogate dynamical downscaling (DDS) dataset for a coupled model intercomparison project phase 5 global climate model. This was achieved by sampling from a large ensemble of DDS datasets from d4PDF project. The sampling process was guided by the probability density functions of the global model\u0026rsquo;s weather patterns, which were classified using a self-organizing map algorithm. We applied LRDS to investigate summertime precipitation over Kyushu Island, Japan. The present-climate simulation revealed considerable inter-model variety in reproducing sea level pressure patterns around Japan in boreal summer. A storyline approach was employed to characterize three distinct behaviors of LRDS in simulating climatological and extreme rainfall over Kyushu under the present climate. Using LRDS, we also evaluated the projected response to climate change. Two contrasting storylines were identified: one showing an increase in rainfall over western Kyushu, and the other indicating increased rainfall over eastern Kyushu.\u003c/p\u003e","manuscriptTitle":"Development of latent resampling downscaling and its application to model bias and climate change projection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-03 18:46:02","doi":"10.21203/rs.3.rs-7266042/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major Revision","date":"2025-10-13T19:42:58+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2025-08-31T05:17:28+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-26T18:30:39+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-09T10:28:25+00:00","index":"","fulltext":""},{"type":"submitted","content":"Climate Dynamics","date":"2025-08-04T23:42:11+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"climate-dynamics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"cldy","sideBox":"Learn more about [Climate Dynamics](https://www.springer.com/journal/382)","snPcode":"382","submissionUrl":"https://submission.nature.com/new-submission/382/3","title":"Climate Dynamics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"1e836de0-1b3c-447c-b9d2-0bc1ef551d55","owner":[],"postedDate":"September 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-18T12:35:06+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-03 18:46:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7266042","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7266042","identity":"rs-7266042","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}