Dynamical downscaling of the western North Pacific from CCSM4 simulations

Introduction Conclusions References


Introduction
Over the past 21 000 years, the Earth has undergone a substantial warming induced by natural vacillations in orbital geometry, a concomitant rebound in greenhouse gas levels, and changing boundary conditions (i.e., ice sheet retreat and rising sea level).At the height of the last glacial maximum (LGM) 21 000 years ago (21 ka), the drop in CO 2 levels to 185 ppm, the drop in CH 4 to 350 ppb, and the far greater extent of ice coverage at high latitudes, are the most important forcing changes for the climate of the LGM, while seasonal and latitudinal distribution of incoming solar radiation at the top of Earth's atmosphere was the second largest difference from those of today (e.g.Figures

Back Close
Full Otto- Bliesner et al., 2006).Thus, surface temperatures in the LGM were about 2 • C lower in the tropics (Broccoli, 2000) and about 30 • C colder over the Laurentide ice sheet (Braconnot et al., 2007).Downscaling of the paleoclimate can provide insights about the paleoclimate conditions that cannot be obtained otherwise by just compiling the proxy records.For example, horizontal and vertical spatial distributions of variables of interest can be inferred or conjectured realistically through the downscale modeling considering the large-scale climate condition as well as the proxy information.Geological studies have speculated about what synoptic scale patterns might have changed in the tropics, but global model simulations of paleoclimates offer synthetic data to compare with results from geologic proxies (Galewsky et al., 2006).Therefore, downscaling of general circulation model (GCM) output can provide a quantitative foundation for paleoenvironment research in a variety of applications.
The goal of this study is to downscale of the Community Climate System Model (version 4; CCSM4) LGM paleoclimate and twentieth century runs from the phase five of coupled model intercomparison project (CMIP5) and paleoclimate model intercomparison project version 3 (PMIP3) to understand the behavior of large-scale dynamics and thermodynamics over the western North Pacific under the LGM and present eras using the Weather Research and Forecasting (WRF) model.Specifically, the purpose of this paper is to address the following: (1) procedures to conduct a dynamical downscaling of the CCSM4 model outputs for the LGM and late modern simulations; and (2) evaluation of the downscaling performance of the WRF model by comparing the downscaling results with the GCM LGM paleoclimate and twentieth century simulation results.
This paper is organized as follows.In Sect.2, we describe the preparation methods and procedures for the WRF model simulation set-up, specifically, for the LGM period.
We discuss the validation of downscaling experiment results in Sect.3. A discussion will follow in Sect. 4. A summary and conclusion are addressed in Sect. 5. Introduction

Conclusions References
Tables Figures

Back Close
Full 2 Data and methods

Domain configuration and model physics scheme
For the dynamical downscaling of the CCSM4, the Advanced Research WRF (ARW) model version 3.1.5(Skamarock et al., 2005) was used.The WRF model has been shown capable of long-term climate simulations (e.g.Done et al., 2006;Yu et al., 2014).
The WRF model in this study employs a terrain following non-hydrostatic pressure vertical coordinate with 51 levels and the model top at 10 hPa.We set our model domain over the western North Pacific both for the LGM and the modern (Fig. 1).Each computational grid has a 36 km horizontal resolution with 171 latitude points by 282 longitude points.The WRF model domain latitudes range from 13.066 • S to 47.435 parameters (i.e., eccentricity, obliquity, and angular presession of the Earth) as recommended by the PMIP2 project (Braconnot et al., 2007; Table 1).

2.2
LGM topography, ice sheet, and CCSM4 data preprocessing Brady et al. (2013) (Briegleb et al., 2004).Those components are combined through a coupler (Gent et al., 2011).For more detail on the CCSM4 model configuration, see Brady et al., 2013.Discernable differences in the LGM topography are the exposed land due to the globally retreated shore line (e.g.Southeast Asia) and the continental ice sheets in the North America and the Northern Europe.The North American ice sheet region of the PMIP3/CMIP5 LGM ice sheet topography, which is included in the LGM CCSM4 forcing files, was blended with the one degree grid resolution topography of PMIP2 ICE-5G (v.1.2) (Peltier, 2004) to represent the LGM topography and continental ice sheets correctly in the WRF model downscaling simulation with 36 km grid space resolution.The PMIP3/CMIP5 LGM ice sheets are also a blended average product among three different ice sheet reconstructions (Brady et al., 2013): ICE-6G v2.0 (Argus and Peltier, 2010), Meltwater routing and Ocean-Cryosphere-Atmosphere response (MOCA; Tarasov and Peltier, 2004), and Australian National University (ANU; Lambeck et al., 2010).The ice sheet mask of the ICE-5G data was used to delineate the ice sheet boundary over the blended LGM topography to assign the soil property appropri-Introduction

Conclusions References
Tables Figures

Back Close
Full ately in the WRF model boundary as ice-covered (Fig. 1).Although our western North Pacific domain did not need the ice sheet, it is essential for North American or tropical channel simulations.Since the CCSM4 LGM simulation is part of the PMIP3/CMIP5 set of simulations, the vegetation in the simulation of this study should be treated as in the CMIP5 preindustrial (PI) experiment where model vegetation was prescribed to PI with interactive leaf area index (LAI) (models with interactive carbon cycle, but no vegetation dynamics) for CCSM4 (B. Otto-Bliesner, 2013, personal communication).Although, the vegetation phenology (LAI and canopy heights) differ between the LGM and PI simulations (B. Otto-Bliesner, 2013, personal communication).Therefore, the land use land cover (LULC) in the default WRF model with the 24-Category USGS land use category are retained except for the LGM ice sheet area and the exposed land area due to the retrieved shoreline for the LGM simulation.For the LULC over the exposed land surface, Climate Long-Range Investigation, Mapping, and Prediction (CLIMAP) LGM vegetation was remapped to USGS 24-category land use categories referencing the vegetation types in "Details of the vegetation scheme used for the map reconstructions" (http://www.esd.ornl.gov/projects/qen/adams3.html) (Table 2 and Fig. 1).Albedo and green fraction information over the ice sheet and the exposed land in the LGM were also generated from the forcing files of the PMIP3/CMIP5 LGM experiment for the usage in the WRF LGM simulation.For the modern simulation, default WRF topography and LULC data were implemented (Fig. 1).Ruby Leung's Fortran program) with a slight modification to correct some critical errors existing in the code.The modification of the code also includes enhancing the vertical profiles of soil moisture and soil temperature in the four soil layers for the longterm climate simulation, eliminating the monotonicity in the soil properties in the vertical column by the community code.We used six hourly pressure level data of the CCSM4 as WRF input data for 3-dimensional pressure levels (e.g.pressure, temperature, zonal/meridional winds, and geopotential), and 2-dimensional surface such as 2 m air temperature (T 2), 2 m relative humidity (RH2), and 10 m winds (U10 and V 10).

Retrieving CCSM4 data
We also used daily outputs of CCSM4 data for sea surface temperature (SST), sea ice content, top layer soil water (SOILWATER_10CM), and top layer soil temperature (TSOI_10CM).The vertical profiles of soil moisture and soil temperature were created by blending the CCSM4 daily data of top layer soil values with the CLM4 monthly mean of soil vertical profiles at three different soil depths at 0.25, 0.7, and 1.5 m.Total model integrations periods of the LGM and the modern simulations are ten years each from PMIP3/CMIP5 model simulation years from 1871 to 1880 and from 1990 to 1999, respectively.Since Year 1870 was the initialization year of the PMIP3/CMIP5 paleoclimate LGM simulation, we chose Year 1871 to be the first year of the LGM simulation to avoid any issue associated with the CCSM4 model spin-up.WRF model simulations were re-initialized every year on 1 January at 00:00 UTC.

Validation of model results
In part due to the lack of observational data during the LGM period for direct comparisons and in part due to the fact that our downscaling simulations are part of long-term simulations, model validations are conducted by evaluating the WRF forecast skills against the PMIP3/CMIP5 CCSM4 simulation results as the "control".We adopt correlation coefficient (r) and root mean square error (RMSE) as statistical measurements.Figures

Back Close
Full RMSE is one of the commonly used error statistics defined as where N is the total number of grid points and M and O represent the values of variables from WRF and CCSM4 simulations, respectively.The difference of domain average from the WRF and CCSM4 results was also computed.To facilitate comparisons between two different model grid resolutions, the WRF results (36 km; 171 × 282) were regridded horizontally into that of CCSM4 grids (1.25 • latitude × 0.9 • longitude; 66×91) using the Earth System Modeling Framework (ESMF) software.

Incoming solar radiation
Incoming solar radiation at the top of the model atmosphere is the fundamental source of energy driving the cascade of energy flow in the climate model.Incoming solar radiation at the top of the atmosphere does not change significantly annually but it is solely dependent on the radiation physics implemented in each model.Figure 2 shows 10 year averages of the latitudinal and temporal distributions of incoming solar radiation at the top of atmosphere (TOA) both in the CCSM4 (top) and the downscaled WRF model (bottom).Left, center, and right panels represent LGM, modern, and the difference between the two periods, respectively.The comparison between the two difference plots shows that the modified CAM radiation schemes for the downscaling experiments using the WRF model reproduced the characteristics of the spatiotemporal solar radiation distribution present in the CCSM4 LGM simulation outputs.
Their comparability between the CCSM4 and the WRF results suggests the validity of the modified CAM radiation schemes for the downscaling experiments using the WRF model.Introduction

Conclusions References
Tables Figures

Back Close
Full

2 m air temperature
Figure 3 shows comparison of ten year averages of 2 m air temperature (T 2) both from WRF and CCSM4 simulations for LGM and late modern periods.Ten-year domain average (standard deviation, SD) of T 2 from the LGM WRF simulation was 16.08 (10.07) • C, while that from the CCSM4 LGM simulation was 18.05 (9.17) • C. WRF bias (WRF-CCSM4) for the LGM simulations was −1.04 (1.45) • C and RMSE was 1.78 (Fig. 3).
Correlation coefficient (r) between the two long-term averages was 0.99.On the other hand, ten year domain average (SD) of T 2 from the Modern WRF simulation was 20.37 (8.42) • C, while that from the CCSM4 modern simulation was 22.07 (7.68) • C. WRF bias (WRF-CCSM4) for the modern WRF simulations was −0.95 (1.34) • C and RMSE was 1.64.r between the two long-term averages was also 0.99 for the Modern simulations (Fig. 3).
T 2 difference map between the LGM and Modern from the WRF downscaling simulations (Fig. 3 upper-right plot) resembles closely that from the CCSM4 simulations (Fig. 3 middle-right plot).The average of the difference between the two plots was −0.09 • C and r between the two was 0.93 (Fig. 3 lower-right).This suggests that the WRF model simulations reproduced the cold LGM climate correctly geographically.Note that decrease of boundary layer temperature during the LGM is greater in the high latitudes than in the lower latitudes, which is consistent with literature (Broccoli, 2000;Braconnot et al., 2007;Korty et al., 2012) ocean combined.However, the WRF bias was slightly reduced with T 2 over the land only.
On the other hand, latitudinal and monthly distributions of ten year averages of T 2 from the WRF and CCSM4 simulations both for the LGM and modern periods (Fig. 4) suggest also that the WRF downscaling simulations reproduced the temporal variance of T 2 in the CCSM4 results well in general.It is notable, however, that compared to the CCSM4 simulations, the WRF simulations underpredicted T 2 during the summer season and overpredicted T 2 during the fall and winter seasons over the region poleward of 15 • N, while they slightly overpredicted T 2 over tropical region ( 13• S-10 • N) from March to December.The tendency of underprediction of T 2 was stronger in the modern simulation than in the LGM simulation.These deviations from the CCSM4 results may be attributable in part to the model physics (Lo et al., 2008) and the higher spatial resolution in the WRF simulations.Model bias can be introduced also from calculating the zonal averages and regridding the relatively higher resolution of WRF results to match with the coarse CCSM4 grids.
Considering only land values, 10 year monthly climate of T 2 shows that the WRF simulation under-predicted the T 2 through the year (Fig. 5).But the model bias remained within the magnitude of 1.3 ). SD of T 2 has a strong seasonality with high in summer and low in winter in both the WRF and CCSM4 simulations.However, WRF model bias and SD of the bias did not vary much through the season.It is also notable that some of model biases are attributable to the regridding of WRF result, which is unavoidable in the model comparison.
To evaluate the WRF forecast error with the time varying T 2 further in detail, we compared the time series of domain averages of T 2 between the WRF and CCSM4.To keep data processing at manageable levels, a sampling method was used here.
That is, out of 14 600 time stamps of the whole 10 year six-hourly model simulation results for the WRF and the CCSM4 each, weekly data (total of 522 samples) were collected and compared from the initial time of the simulations.All the time stamps of the sampled model results should be 00:00 UTC.Over the entire model domain, T 2 average of the sampled CCSM4 LGM simulation results was 17.03 • C, while that of the WRF simulation was 15.62 • C. Also, the mean of the WRF forecast error (bias: WRF -CCSM4) was −0.46 • C, and the average of RMSE was 2.17 (Fig. 6, Table 3).The counterparts in the modern simulations were 22.02, 19.98, −1.25, and 3.09 • C for the CCSM4, WRF model simulations, model forecast error, and the RMSE, respectively (not shown).Note that the WRF model bias and RMSE in the LGM are less than those of modern.It is clearly notable that the downscaled WRF LGM simulation result was about 4 • C colder than that of the modern simulation, which is consistent with one of the defining characteristics of the LGM (Brady et al., 2013).Table 3 summarizes WRF model performance with average, (normalized) forecast errors, and RMSE from the time series of major variables over the ten year simulation period.Variables include T 2, soil moisture at 5 cm depth (SMOIS), soil temperature at 5 cm depth (TSLB), RH2, 10 m U wind (U10), 10 m V wind (V 10), sea level pressure (SLP), geopothential height (GHT), 850 hPa temperature (TT), 850 hPa Uwind (UU), and 850 hPa V-wind (VV).It is notable that WRF model forecast errors were relatively small in general but wind component variables at 10 m height and 850 hPa level have relatively high errors both in the LGM and Modern simulations.

2 m relative humidity and total precipitable water
Along with T 2, atmospheric moisture in the near surface and in the vertical column is a critical component in the model thermodynamics.To evaluate the downscaling performance in this feature, relative humidity at 2 m (RH2) and total precipitable water (TPW) are examined in this section.Ten years domain average (SD) of RH2 from the WRF LGM simulation was 85.88 (9.8) %, while that from the CCSM4 was 79.14 (5.96) %.Average WRF bias for the LGM simulations was 6.41 (7.98) % and the RMSE was 10.22.Correlation coefficient of RH2 between the WRF and the CCSM4 simulation averages was 0.51 (Fig. 7).Likewise, ten years domain average (SD) of RH2 from the WRF Modern simulation was 87.5 (7.71) %, while that from the CCSM4 Modern simulation was 79.91 (4.13) %.Av-Introduction

Conclusions References
Tables Figures

Back Close
Full erage WRF bias for the Modern simulations was 7.39 (6.57) % and the RMSE was 9.89.Correlation coefficient of RH2 between the WRF and the CCSM4 Modern simulation averages was 0.42 (Fig. 7).Comparisons of the mean RH2 between the WRF and the CCSM4 simulations suggests that (1) the WRF model produced wet biases both in the LGM and the Modern simulations in general, (2) in both the LGM and the Modern simulations, the WRF model produced dry biases over the land in general while producing strong wet biases over Tibet Plateau and Korea/Japan regions.Spatial distributions of RH2 differences between the LGM and the Modern simulations resemble each other for the WRF and the CCSM4 simulations in general.The WRF downscaling simulations reproduced well the drier continental surface during the LGM than that in the Modern climate.It is notable that RH2 over the open ocean does not differ between the LGM and the Modern both for the WRF and the CCSM4 simulations but the WRF simulations produced relatively strong dry biases compared to CCSM4 over the exposed land area at the LGM, in particular, near the Maritime continent.It seems that differences in model physics and in land surface model that are implemented in the WRF downscaling simulations are attributable to those spatial difference of RH2.
Considering LGM RH2 over land values only, monthly climate of RH2 shows that the downscaled WRF simulation underpredicted RH2 about 10 % yearly compared to the CCSM4 (Fig. 8).The WRF model bias contains a weak seasonal cycle ranging from −11.01 % in December to −2.7 % in August which is coupled with its RSME.Interestingly, both the WRF and CCSM4 simulations have two months of RH2 minima in April (WRF: 65.11 % and CCSM4: 74.74 %) and October (67.99 and 74.13 %).Latitudinal and monthly distributions of ten year averages of RH2 (Fig. 9) also suggest that compared to the CCSM4 results, the WRF simulations produced relatively strong wet biases over 10 ∼ 30 • N latitudes.It seems that the wet biases over the ocean in the WRF simulations are mostly attributable to these positive biases in the zonal mean in both the LGM and Modern simulations (see Fig. 7).This tendency seems to be also associated with T 2 distributions (see Fig. 4).Figures

Back Close
Full Meanwhile, ten years domain average (SD) of TPW from the WRF LGM simulation was 26.68 (12.08) mm vs. 29.55(11.44) mm from the CCSM4 simulation (Fig. 10).Domain average WRF model bias (SD) and the RMSE for the LGM simulations were −1.79 (2.80) mm and 3.33, respectively.r between TPWs from the WRF LGM and the CCSM4 LGM simulations was 0.97.On the other hand, ten years domain average (SD) of TPW from the WRF Modern simulation was 35.39 (14.55) mm vs. 38.13 (13.14) mm from the CCSM4 Modern simulation.Domain average WRF bias and the RMSE in TPW for the Modern simulations were −1.42 (3.36) mm and 3.65, respectively.r of TPW between the WRF and the CCSM4 simulations was 0.97 in the modern case.Thus, WRF model both in LGM and Modern simulations reproduced TPWs of the CCSM4 results relatively well (r = 0.97).Still, the WRF has a tendency to overpredict TPWs over the western tropical Pacific and to underpredict over the southeastern China Sea while the spatial distributions of TPWs resemble each other between the WRF and the CCSM4 simulations (Fig. 10).

Vertical distribution of 3-dimensional variables
Examining vertical distribution of key dynamic and thermodynamic variables of threedimension is useful and necessary to evaluate the performance of the downscaling model simulations.Figure 11  tion (not shown).It is notable that the WRF downscaling simulations reproduced the vertical and latitudinal characteristics of relative humidity reasonably well (Fig. 11d): relatively high RH in the tropical latitudes and relatively low RH in the mid-latitude at the midtroposphere, resembling the Hadley circulation.
Figure 12 presents the comparisons between the CCSM4 and WRF results from their LGM simulations with the ten years climatology of zonal averages of zonal wind (U) and meridional wind (V ).Apparently, both the U and V from the CCSM4 LGM simulation were reproduced comparably in the WRF LGM simulation.The WRF downscaling experiment has closely simulated tropical easterly winds and midlatitde westerly winds balancing each other over the extratropical latitudes (10-30 • N) over the model domain.
Note that lower-level easterly winds were slightly enhanced in the WRF simulations over the CCSM4 simulations.The zonal averages of U from the WRF downscaled simulation clearly depicts the zone of enhanced easterly winds in the low-level atmosphere over the western North Pacific, which is one of the critical components of tropical cyclone development over the region, as well as the enhancement of westerly winds in the upper-level atmosphere in the midlatitude, which represents the midlatitude upper-level jet stream.On the other hand, the meridional wind (V ) for the CCSM LGM simulation (Fig. 12c) shows the large-scale general circulation patterns with reduced magnitudes.In contrast, the zonal climatology of V from the WRF LGM downscaling simulation (Fig. 12d) clearly represents tropical convergence (divergence) in the low-level (upper-level) atmosphere which is the characteristic of Inter-Tropical Convergence Zone (ITCZ).Also depicted are the upper-level convergent flow over 20 • N and the low-level divergence over 30 • N. The same plots but for the Modern simulation show the similar vertical distributions of U and V in the CCSM4 and WRF simulations (not shown).Compared to the WRF LGM simulation, the WRF Modern simulation produced slightly enhanced magnitudes both in U and V in the Southern Hemisphere.Introduction

Conclusions References
Tables Figures

Back Close
Full

Discussion
Boundary or surface conditions play a critical role in the climate model simulations.In a long-term climate simulation of the modern climate, among the boundary variables updating SST once a day may be the most important input variable to check to avoid the model simulation drift from the long-term average.The other time-varying boundary inputs (such as vegetation, soil moisture/temperature, albedo, etc.) are controlled by climatology data, which do not vary much over time under the current climate.However, to simulate a different climate state like the LGM, modelers should be careful with their choice in constructing those time-varying boundary variables.While SST will be provided by the GCM, the other boundary variables should be determined and provided by the modeler.The more difficult choice should be made to determine the vegetation over the exposed land only during the LGM period due to the retrieved shorelines along the coast region in the Asia (Fig. 1).Since model simulation cannot be executed without the boundary conditions reconstructed for the LGM condition and those conditions cannot be modified during the execution, there is a clear limitation of long-term paleoclimate simulation, especially with the LGM.Paleoclimate reconstruction research with proxy records might reduce the uncertainty in the geographical extents of boundary input variables in the future.

Summary and conclusion
A In this paper, we described procedures to conduct a dynamical downscaling of the CCSM4 model outputs for the LGM and late modern simulations.In particular, to realize the LGM topography properly in the downscaling WRF model simulation, PMIP 2 ICE-5G data (v.1.2) (Peltier, 2004) and PMIP3/CMIP5 LGM ice sheets data was used to modify the WRF geographic data (GEOG) to represent the LGM topography and continental ice sheets close to those of the CCSM4 LGM simulation.
For the model validation, the results from the WRF simulations were compared with the CCSM4 LGM paleoclimate and twentieth century simulation results using the ESMF regrid software for the quantitative and statistical comparison.Using 10 year averages of the forecast error (the difference between the WRF and CCSM4 simulations) for evaluated variable (T 2, SMOIS, TSLB, RH2, U10, V 10, SLP, GHT, TT, UU, and VV) (Table 3), it was shown that the WRF downscaling experiments reproduced the thermodynamic conditions closely to those of the CCSM4 LGM and Modern simulation results in general.
Overall, results of the dynamical downscaling of the CCSM4 LGM paleoclimate and twentieth century using the WRF model suggest that the WRF model is capable of long-term simulations in a different climate state in the past.Moreover, comparisons of vertical distributions of three-dimensional variables (T , RH, U, and V ) suggest that the WRF model can correct biases in the GCM model, partly attributable to the low grid resolution, and produce more realistic spatial distribution patterns of the pressure-level variables, presumably, partly due to model physics and enhancement in the spatial resolution.It seems that the downscaling of a GCM model using the 36 km grid resolution WRF model for the regional climate simulation is computationally cost-effective and reliable from the perspectives of model thermodynamics in general while there are some forecast errors still existing with dynamic variables.This study might thus profitably contribute to dynamical downscaling studies of the paleoenvironment as well as regional climate change studies.Figures

Back Close
Full  Full  Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | • N and the domain longitude range from 93.75 • E to 206.25 • E. The adaptive time step option was applied for the actual integrations with an average time step (dt) of about 74 s.Except for the modified community atmosphere model (CAM) shortwave and longwave radiation schemes for the LGM climate simulation, model physics and dynamic schemes employed for the LGM and the modern simulations are identical.The ARW (version 3) modeling system user's guide (National Center for Atmospheric Research (NCAR), 2014) was referenced to configure the model physics schemes: WRF Single-Moment 6-class scheme for microphysics; CAM scheme for shortwave and longwave radiation schemes; the Pennsylvania State University/NCAR mesoscale model version 5 (MM5) similarity scheme based on Monin-Obukhov with Carslon-Boland viscous sub-layer for surface layer option; Noah Land Surface Model represents surface processes; Yonsei University planetary boundary layer (PBL) scheme represents boundary layer; Cumulus convection is parameterized with the Kain-Fritsch (new Eta) scheme; model sea surface temperature (SST) and skin temperature are updated every time step.CAM radiation scheme for the LGM simulation was modified to be consistent with the LGM radiation of the PMIP3/CMIP5 experiment accounting for the different forcing that are related with atmospheric concentration ratios of CO 2 , CH 4 , NO 2 and Earth's orbital Figures Back Close Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | We obtained 6 hourly PMIP3/CMIP5 data in NetCDF format of both paleoclimate runs and the 20th century MOAR control simulation from the University Corporation for Atmospheric Research (UCAR) Yellowstone data storage (a.k.a.HPSS).The latter are also available at http://www.cesm.ucar.edu/experiments/cesm1.0/.The CCSM4 NetCDF file was converted into GRIB file for the WRF model input by using a community package called cam_to_wps (B.Fisel, 2014, personal communication; adopting 6 Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | . Comparisons of T 2 between the WRF and CCSM4 simulations during both the LGM (left column) and late modern (center column) show that the WRF model produced negative biases slightly overall.It seems that those negative biases are spatially clustered over the midlatitude coastal region and its downwind region.Considering only for the LGM simulation results, ten year average (SD) of T 2 over the land only from the WRF simulation was 8.62 (11.01) • C, while that from the CCSM4 was 11.81 (10.97) • C. WRF bias (WRF-CCSM4) was −0.90 (1.65) • C and RMSE was 1.88 (not shown).Apparently, T 2 over the land was colder than over the ocean and T 2 over the land has more variability (with higher SD) than when the Figures Back Close Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | shows vertical plots of the ten years zonal averages of atmospheric temperature (T ) and relative humidity (RH) comparing CCSM4 and WRF results from their LGM simulations.The WRF LGM simulation reproduced the vertical distribution of T in the CCSM4 LGM simulation closely.RH plot from the WRF LGM simulation resembles that from the CCSM4 LGM simulation except in the tropical upper-level atmosphere.It seems that relatively high RH in the tropical upper-level atmosphere in the CCSM4 results are model bias in the CCSM4 LGM simulation.However, the WRF simulation seems to have corrected the moist bias in the tropical upperlevel RH in the CCSM4, producing reasonable relative humidity in vertical as well as in latitudes.The same plot but for the Modern simulations shows the similar bias in the CCSM4 simulation and the corrected vertical distributions of RH in the WRF simula-Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | study of a dynamical downscaling of the CCSM4 LGM paleoclimate and twentieth century runs from the CMIP5/PMIP3 was conducted using the WRF model in 36 km grid spatial resolution.The goal of the study is to investigate the behavior of largescale dynamic and thermodynamic variables in the downscaling experiments over the western North Pacific under the LGM and modern climates.The model integrations periods of the LGM and the modern simulations were ten years from 1871 to 1880 and from 1990 to 1999, respectively.Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Table3.CCSM4 and WRF Model Comparison.The results from the WRF simulations were compared with the CCSM4 LGM paleoclimate and twentieth century simulation results using the ESMF regrid software for the quantitative comparison.10 year averages of the forecast error (the difference between the WRF and CCSM4 simulations), RMSE, and Normalized forecast error (FE) are shown for evaluated variable (T 2, SMOIS, TSLB, RH2, U10, V 10, SLP, GHT, TT, UU, and VV.Normalized FE was obtained from (100 × FE/WRF).

Figure 1 .Figure 2 .Figure 3 .
Figure 1.WRF model domains with topography for the Modern (upper left) and LGM (upper right) periods and land use land cover category plots for the Modern (bottom left) and LGM (bottom right) periods.The LGM LULC was reclassified from CLIMAP LGM vegetation into the USGS 24-Category.SeeTable 1 for the vegetation categories for detail.

Table 1 .
Boundary conditions, forcing that are related with atmospheric concentration ratios of trace gases and Earth's orbital parameters that are applied for the simulations of the modern and the LGM as recommended by the PMIP2 project.