Projection of changes in future weather extremes using super-high-resolution global and regional atmospheric models in the KAKUSHIN Program : Results of preliminary experiments

Changes in future weather extremes are projected using a global atmospheric general circulation model and a non-hydrostatic regional climate model under the global warming environment in the near future (2030s) and at the end of the 21st century. The global 20-km mesh model can simulate tropical cyclones more realistically in their strength, structure and geographical distribution together with associated heavy rainfall and strong surface winds as compared with lower resolution models. According to the SRES A1B scenario, it is projected that at the end of the 21st century there will be a 40%∼60% increase in precipitation and a 15%∼20% increase in surface wind speeds within a 100 km radius of the tropical cyclone center. Ensemble simulations using the global 60-km model version are performed to obtain information on the uncertainty of projections. Downscaling with the 5-km mesh non-hydrostatic regional climate model is also performed for the Japanese summer rainy season. It is found that the frequency of heavy precipitation will increase in the future for the hourly as well as daily precipitation. In particular, the strong hourly precipitation will increase even in near future: 99.9%-ile value of hourly precipitation increases 7% in the near future and 21% at the end of the 21st century.


INTRODUCTION
In the Fourth Assess Report of the Intergovernmental Panel on Climate Change (IPCC), a dataset of more than 20 global coupled atmosphere-ocean general circulation models (AOGCMs) under the Coupled Modeling Intercomparison Project (CMIP3) are fully utilized to project future climate changes for various scenarios (IPCC 2007).These results give good information on future climate changes, but it cannot be denied that the low resolution of the models has an undesirable impact on the results particularly for extreme weather events, because the horizontal resolution of these models is of about 100 km to 400 km.Therefore, a high spatial resolution model is anticipated for use to study extreme weather events and to project their modification by climate changes in order to contribute to decisionmaking for adaptation studies and countermeasures.
Recently, a super-high resolution atmospheric general circulation model (AGCM) with a horizontal grid size of about 20 km has been developed for use in climate change studies (Mizuta et al. 2006) based on the Japan Meteorological Agency (JMA) numerical weather forecast model.The grid size of this model is several times higher than that previously used in climate model simulations.In the previous experiment, we performed the present-day simulation using the observed sea surface temperature (SST) and the global warming simulation by adding the SST anomalies obtained by the Meteorological Research Institute AOGCM (MRI-CGCM).Utilizing the results of this experiment, Kusunoki et al. (2006) investigated the Baiu rain band changes over East Asia at the end of the 21st century, while Kitoh et al. (2008) showed future climate projections over the Middle East.Moreover, Kamiguchi et al. (2006) discussed changes in extremes in precipitation and Oouchi et al. (2006) investigated changes in the frequency of wind intensity of tropical cyclones in a future warmed climate.
The previous experiment has used the future SST anomalies projected by one particular model.It is plausible that regional climate change largely depends on the spatial pattern of the SST anomalies used.In order to avoid such an uncertainty, we now use the ensemble mean SST anomalies of the available AOGCMs of CMIP3 (Mizuta et al. 2008).Although the global 20-km mesh model is a unique one in terms of its horizontal resolution for global change studies with multi-decades integrations, the computer power is still unable to make ensemble simulations with such a super-high-resolution model.To cover this caveat, ensemble simulations with the 60-km resolution version of the same model have also been performed.
Japan often suffered economic and human damage from short-term heavy precipitation over hundreds mm/day due to Baiu Front and typhoons.The 5-km mesh non-hydrostatic regional climate model can reproduce local extreme rainfall events realistically (Kanada et al. 2008).Therefore, further dynamical downscaling with a 5-km non-hydrostatic regional climate model is performed to project future changes of rainfall extremes for the summertime using the 20-km mesh model as the initial and lateral boundary conditions.Nakamura et al. (2008) discussed the changes of heavy precipitation during the Baiu season (June to July) in Japan.We have now extended the integration for the whole summer season.

GCM
The AGCM used in this study is a climate-model version of the JMA operational numerical weather prediction model (Mizuta et al. 2006).The simulations were performed at a triangular truncation of 959 with a linear Gaussian grid (TL959) in the horizontal.The transform grid uses 1920 × 960 grid cells, corresponding to a grid size of about 20 km.The model has 60 layers in the vertical with the model top at 0.1 hPa.Mizuta et al. (2006) showed that simulated global distributions of the seasonal mean atmospheric circulation fields, surface air temperature and precipitation agree well with the observations.Moreover, the model improves the representation of regional-scale phenomena and local climate by increasing horizontal resolution to better express topographical effects and physical processes without degrading the representation of global climate.Murakami et al. (2008) compared the 20km and the 60-km mesh models through medium-range forecast experiments for the twelve typhoons between 2002 and 2005.They addressed the importance of such a high resolution to predict the intensity and inner-core structure of typhoons.

RCM
We also use a non-hydrostatic model with a horizontal resolution of 5 km (NHM-5km; Saito et al. 2006).The model uses a cloud microphysics scheme but also uses the Kain-Fritsch cumulus parameterization scheme (Kain and Fritsch 1993).The experimental domain is 669 × 527 horizontal grids and 50 vertical layers (top height at 22 km), covering a wide region from the eastern part of the Asian Continent to the Japanese Islands.The spectral boundary coupling (SBC) method (Kida et al. 1991) is employed to reduce phase errors between the outer model and NHM-5km.
The model verification has been done by using the JMA operational regional analysis data (RANAL), which have a horizontal resolution of 20 km, as initial and boundary conditions.This experiment is conducted from June to October between 2002 and 2006.Kanada et al. (2008) give the model performance in the reproducibility of extreme precipitation over Japan by the NHM-5km.

GCM
The simulated present-day climate by the 20-km mesh model is almost the same as in the previous experiments in Mizuta et al. (2006), because the main difference of the experimental protocol is whether we use the climatological SST or interannually varying SST.The orographic nature of seasonal precipitation is therefore well represented by the model.This aspect is investigated by Rajendran and Kitoh (2008), who made the preliminary analysis of the Indian summer monsoon in the future climate projection using the first 10-years' data for the present-day and at the end of the 21st century with this 20-km mesh AGCM.They noted that the model was able to capture precipitation maxima to the west of the Western Ghats in southern India and along a southern periphery of the Himalaya range.
Simulated changes of the 25-year averaged global mean surface air temperature by the 20-km mesh model are 0.93°C in the near future and 2.73°C at the end of the 21st century, respectively.Those for the global mean precipitation change ratio are 2.1% and 7.0%, respectively.Due to the smaller changes in temperature increase, most of the projected mean climate changes may not be statistically significant in the near future.
An advantage of the 20-km mesh global model is that typhoons (tropical cyclones in the northwestern Pacific) are more realistically simulated in terms of their strength and structure than in lower-resolution models in the forecast mode (Murakami et al. 2008).This fact suggests that the 20-km mesh global model would give us a more reliable future projection of associated weather extremes.Tracks of the simulated tropical cyclones capture the major features of the observed Best Tracks.A total of 2086 tropical cyclones are detected for the present and 1667 at the end of the 21st century.Thus there is a reduction in the number of tropical cyclones.However, we do not find any statistically significant changes for the near future because the signal is small compared to natural variability.
The SST ensemble experiments with the 60-km mesh model can give us some assessment of uncertainty.It is found that the different future SST-changes make more significant differences in the future projections than the different atmospheric initial conditions.It is projected that the globally averaged occurrence of tropical cyclones decreases under global warming in all SST cases, but changes in occurrence differ from basin to basin.The latter suggests an importance of future SST pattern changes and associated atmospheric circulation changes.Sugi et al. (2009) discuss in detail a reduction in global tropical cyclone frequency due to global warming.
Figure 2 shows the radial profiles of precipitation and surface wind around all tropical cyclones.Although the model realistically reproduces the typhoon intensity in the forecast mode (Murakami et al. 2008), the model in the climate simulation has a weaker wind profile compared to the observations (Oouchi et al. 2006).
At the end of the 21st century, large changes occur near the inner-core region of tropical cyclones.There is an increase around 40 60% for precipitation and 15 20 % for surface wind.An increase of surface wind speed of more than 4% can be seen up to 500 km from the storm center.The increase in wind speed was more remarkable in the mid-troposphere than at the surface.In addition, the warm core was strengthened at the upper troposphere between 200 and 400 hPa.These changes were caused by enhanced heating near the eyewall region.The increase in precipitation is larger than that in wind speed because an increase in precipitation depends both on an increase in wind speed and an increase in moisture.There are only few studies available in the literature with high-resolution models.Using the same model with the current study but with different SST anomalies (20% smaller in global annual SST anomalies), Oouchi et al. (2006) showed an 11% increase in maximum wind.Knutson et al. (2008), although only over the Atlantic, showed a 3% increase in maximum wind and precipitation increase 37%, 23% and 10% within 50 km, 100 km and 400 km of the hurricane center.
Although the increase of surface wind is up to 20%, changes in storm energy and/or wind damage are com--51 - parable to those in precipitation.It is also noted that Figure 2 is calculated for all azimuth and there is an azimuth dependence on precipitation and wind intensity distribution around the storm center.The projection for the end of the 21st century indicates that the number of strong tropical cyclones increases and their life tends to be longer in the future.These aspects will be dealt with in detail in a separate paper.

RCM
The NHM-5km results forced with the 20-km AGCM data for the present climate are first compared with the observed rainfall and temperature data over Japan.It was found that detailed regional characteristics of monthly precipitation were well reproduced, but the simulated precipitation has bias.The bias is induced by poor reproducibility of Baiu with the 20-km AGCM for June and July, and the underestimation of the number of typhoon approaches to Japan for September.Therefore, some bias correction is necessary when present and future climate data are directly used in a quantitative manner, in order to improve the ability of the global model in reproducing the regional climate.The probability density functions of daily and hourly precipitation were also simulated realistically even for quantitative comparison with observation.
Here we present results of the preliminary analysis of future climate changes by comparing the 10-year integrations for the present and the future climates.Firstly, total precipitation for the five-month period (June to October) averaged for all Japan does not change much, although there are month-to-month changes.However, it is found that the areas experiencing extreme daily precipitation over around 100 mm/day increase by more than 25% at the end of this century.Figure 3 shows the probability density function (pdf) and cumulative relative frequency with the NHM-5km of daily precipitation and hourly precipitation for the present, near-future and future simulations.The value is averaged for land grid points over Japan during June October.When compared to the observations, the present-day simulation of the NHM-5km overestimates the high-end daily precipitation by about 10%, while the model underestimates the hourly precipitation by about 10%.
At the end of the 21st century, the frequency of heavy precipitation will increase both for the daily precipitation and hourly precipitation.The 99 percentile and 99.9 percentile values are plotted in Figure 3.Both the 99%-ile and 99.9%-ile values increase 17% for daily precipitation at the end of the 21st century, while those for hourly precipitation increase 16% and 21%, respectively.It is interesting to note that in the near future the pdf of daily precipitation will not change much from the present, but the pdf of hourly precipitation will increase by 7%.We do not have a measure of statistical significance on such changes, and more work is needed on whether very extreme cases could appear even in the near future and more extreme cases intensify more in a warmer world.

CONCLUDING REMARKS
This study used the 20-km mesh global model and the 5-km mesh cloud-resolving regional model for the climate change projections.These models are developed based on the JMA operational global numerical forecast model for the short-and medium-range forecast and the regional mesoscale model for the very short-range forecast, respectively.Resolution does not have much effect on future projected changes in seasonal mean precipitation, but does affect their extremes.The 20-km mesh model simulates the characteristics of tropical cyclones more realistically as compared with lower resolution models.It is found that the frequency of heavy precipitation will increase in the future for the hourly as well as daily precipitation.In particular, the strong hourly precipitation will increase even in the near future almost linearly with temperature increase (99.9%ile value of hourly precipitation increases 7% in the near future and 21% at the end of the 21st century).
The global and regional super-high-resolution models, which are able to simulate present weather realistically, project more frequent occurrence of future and near-future weather extremes in a different manner as was seen in Figure 3 for example.Those results together with information on the uncertainty are significant for decision-making processes.These data are used for various application studies on disaster prevention in collaboration with other national institutes and univer--52 - sities.In particular, changes of disaster environment regarding landslide, debris flow, flood, drought, storm surge and strong wind are being evaluated.Cooperation is extended internationally, including the adaptation studies in monsoon Asia under the Japan International Cooperation Agency (JICA) funds and the adaptation studies in Latin America and Caribbean regions, such as climate impacts in coastal zones, high-mountains and the Amazon, under the World Bank funds.

ExperimentA
Figure 1.A schematic of the experiment.

Figure 2 .
Figure 2. Radial profile of (a) precipitation and (b) surface wind around all simulated tropical cyclones.The four digits in parentheses show the number of tropical cyclones.

Figure 3 .
Figure 3.The probability density function (right hand scale) and cumulative relative frequency (left hand scale) distributions in Japan during June October with the NHM-5km of (a) daily precipitation and (b) hourly precipitation.Blue, green and red lines denote the present-day, near-future and future simulations, respectively.For comparison, observations are shown in black lines.The 99%-ile and 99.9%-ile values of precipitation are shown in each panel.