Assessment of Dynamic Downscaling of the Extreme Rainfall over East Asia Using a Regional Climate Model

Transcription

1 ADVANCES IN ATMOSPHERIC SCIENCES, VOL. 28, NO. 5, 2011, Assessment of Dynamic Downscaling of the Extreme Rainfall over East Asia Using a Regional Climate Model GAO Yanhong 1 (p ù), Yongkang XUE 2, PENG Wen 1 ($ >), Hyun-Suk KANG 3,andDuaneWALISER 4 1 Laboratory for Climate Environment and Disasters of Western China, Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences, Lanzhou Department of Geography and Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, California, USA 3 Climate Research Laboratory National Institute of Meteorological Research Korea Meteorological Administration, Korea 4 Jet Propulsion Laboratory/California Institute of Technology, California, USA (Received 12 July 2010; revised 11 November 2010) ABSTRACT This study investigates the capability of the dynamic downscaling method (DDM) in an East Asian climate study for June 1998 using the fifth-generation Pennsylvania State University National Center for Atmospheric Research non-hydrostatic Mesoscale Model (MM5). Sensitivity experiments show that MM5 results at upper atmospheric levels cannot match reanalyses data, but the results show consistent improvement in simulating moisture transport at low levels. The downscaling ability for precipitation is regionally dependent. During the monsoon season over the Yangtze River basin and the pre-monsoon season over North China, the DDM cannot match observed precipitation. Over Northwest China and the Tibetan Plateau (TP), where there is high topography, the DDM shows better performance than reanalyses. Simulated monsoon evolution processes over East Asia, however, are much closer to observational data than reanalyses. The convection scheme has a substantial impact on extreme rainfall over the Yangtze River basin and the pre-monsoon over North China, but only a marginal contribution for Northwest China and the TP. Land surface parameterizations affect the locations and pattern of rainfall bands. The 10-day re-initialization in this study shows some improvement in simulated precipitation over some sub-regions but with no obvious improvement in circulation. The setting of the location of lateral boundaries (LLB) westward improves performance of the DDM. Including the entire TP in the western model domain improves the DDM performance in simulating precipitation in most sub-regions. In addition, a seasonal simulation demonstrates that the DDM can also obtain consistent results, as in the June case, even when another two months consist of no strong climate/weather events. Key words: DDM, MM5, cumulus convection scheme, land parameterization, re-initialization, lateral boundary location Citation: Gao, Y. H., Y. K. Xue, W. Peng, H.-S. Kang and D. Waliser, 2011: Assessment of dynamic downscaling of the extreme rainfall over East Asia using a regional climate model. Adv. Atmos. Sci., 28(5), , doi: /s Introduction In recognition of the economic significance of water resources in China, many studies have sought to examine the effects of climate change on components of the hydrologic budget. The common approach has been Corresponding author: GAO Yanhong, gaoyh@lzb.ac.cn. China National Committee for International Association of Meteorology and Atmospheric Sciences (IAMAS), Institute of Atmospheric Physics (IAP) and Science Press and Springer-Verlag Berlin Heidelberg 2011

2 1078 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 to combine basin-scale hydrologic models with climate change scenarios derived from general circulation model (GCM) output (see Watson et al., 1996). Due to their coarse resolution, however, GCMs overlook climatologically details necessary for accurate runoff estimation at the basin scale. The advance of higherresolution GCMs may improve the situation; however, hydrologic modeling at the basin scale requires climatologically information on scales that are generally far smaller than the typical grid size commonly used by GCMs for climate simulations (e.g., Phillips, 1995; Hay et al., 2002). In order to downscale (or translate) information from the coarse-resolution GCMs to the basin scales for hydrologic modeling, it is necessary to resolve subgrid scale information in the simulated fields. One way to achieve this is by statistical downscaling (Wilks, 1995; Wilby et al., 1999). In this approach, empirical relations are developed between features relatively reliably simulated by a GCM at grid-box scales (e.g., 500-hPa geopotential height) and surface variables at sub-grid scales (e.g., precipitation occurrence and amounts). Statistical downscaling (SDS), however, is ultimately limited by the assumption of stationarity in the empirical relations. Skillful SDS results for the present climate do not necessarily translate to skillful forecasts of future climate. The nonstationarity in empirical climate relations is well documented (e.g., Ramage, 1983). An alternative approach is the dynamical downscaling method (DDM), in which a regional climate model (RCM) uses GCM output as initial and lateral boundary conditions (LBCs) for much more spatiallydetailed climatologically simulations over a region of interest (e.g., Laprise et al., 2000; Feser et al., 2001; Denis et al., 2002; Miguez-Macho et al., 2004; Kanamitsu and Kanamaru, 2007; Castro et al., 2007). We refer to a limited-area model as an RCM when its integration time is greater than approximately two weeks, such that the sensitivity to initial atmospheric conditions is no longer a dominant factor (Jacob and Podzun, 1997; Castro et al., 2005). The most important issue is whether, and if so under what conditions, the DDM is really capable of improving/adding more climate information at different scales compared to the GCM or reanalysis that imposes LBCs to the RCMs. Leung et al. (2003) suggest that more accurate and spatially-detailed climate forecasts can be made with downscaling using regional climate modeling. Castro et al. (2005) proposed four types of dynamic downscaling and investigated Type 2 dynamic downscaling (i.e., reanalysis as LBCs for RCMs) in a suite of experiments for May 1998 using the Regional Atmospheric Modeling System (RAMS) model (Pielke et al., 1992) to evaluate its downscaling ability. They found that, for particular cases, the DDM does not retain values from the large-scale circulation, which exists in the reanalyses. Some studies demonstrated similar results and conclusions (e.g., Jones et al., 1995; Rockel et al., 2008). Lo et al. (2008) based on DDM work using the Weather Research and Forecasting (WRF) model, concluded that dynamical downscaling outperforms bilinear interpolation, especially for meteorological fields near the surface over mountainous regions. In their experiment, the 3-D nudging method was introduced and generates realistic regional-scale patterns that were not resolved by simply updating the LBCs as done traditionally, which, therefore, significantly improved the accuracy of simulated regional climate information. However, applying this method for future climate research is questionable because it is unclear which data can be applied for nudging. A number of studies, most of which are North American regional studies (e.g., Liang et al., 2001; Xue et al., 2007), have shown that the results from the DDM were sensitive to RCM settings and different parameterizations in RCMs. Among these studies, Xue et al. (2007) investigated the ability to dynamically downscale and found that domain size, LBCs, and grid spacing are crucial issues. In addition, convective parameterizations (Liang et al., 2004; Liu et al., 2006; Kang and Hong, 2008a) and land surface processes (Xue et al., 2001, 2007; Kang and Hong, 2008b) have also been identified as crucial factors in the DDM. In these studies, the impacts of different factors on the DDM were investigated separately. For example, Liang et al. (2001) focused on the buffer zone impact on the DDM; Xue et al. (2001), Gao et al. (2008a, b), and Liu et al. (2008) mainly focused on the land processes impact on the DDM; and Lee et al. (2005) and Liu et al. (2006) only emphasized the convection parameterization impact. A synthesized study that includes all major factors affecting the RCM s downscaling ability on the East Asian summer monsoon (EASM) is still lacking. Dynamic downscaling of monsoon precipitation for East Asia is a challenging task. Difficulties arise from the complexities associated with the thermal and mechanical effects of mountain ranges (Ding, 1994); multiscale interactions involving planetary, synoptic, and mesoscales; and various physical and dynamical processes that govern the circulation characteristics of different scales in the tropics, subtropics, and midlatitudes (Lau and Yang, 1996). Scale interactions are extremely complex in the Asian Pacific monsoon region (Holland, 1995) and are further complicated owing to the effects of the Tibetan Plateau (TP), land surface processes, ocean continent contrast, and air

3 NO. 5 GAO ET AL sea interaction associated with monsoonal activities. Preliminary investigations have been made to investigate the downscaling of East Asian climatology and extreme climate events using RCM simulation (e.g., Wang et al., 2000; Liu et al., 2002; Zheng et al., 2002; Kang and Hong, 2008b). Most of them focused on a single factor, such as cumulus parameterization or domain size and location. Although these studies with a single factor contributed to our understanding, they cannot give a broadly comprehensive picture and qualify the relative importance of these factors in the DDM. In addition, most of these studies focus on the Yangtze River basin, only a part of East Asia. The domain is not big enough to provide information for a comprehensive East Asian monsoon system study. Comprehensive investigation of DDM issues with multiple major factors and with the extensive use of observational data over the entire EASM area is lacking but necessary. This paper assesses the capability of the DDM in the East Asian monsoon study for June 1998, a heavy flooding month, under different conditions using the Pennsylvania State University-National Center for Atmospheric Research Fifth-Generation Mesoscale Model (MM5) (Dudhia et al., 2005). Impacts of the four major factors discussed above, i.e., cumulus convection parameterization, land parameterization, location of lateral boundary (LLB), and re-initialization on DDM performance over East Asia are discussed. This study intends to provide a synthesized assessment of all of major factors on East Asian DDM performance, their relative roles, and spatial characteristics. Different from most of the aforementioned East Asian studies, in which different climate variables were selected to assess different factors, this study systematically assesses the downscaling ability of these major factors in all major aspects of the EASM. The strength and weakness of each factor of the DDM will be comprehensively assessed. Furthermore, rather than selecting one reanalysis dataset, such as ECMWF or NCEP Reanalysis, for evaluation, as done in most previous studies, several observational data and reanalyses datasets are applied for downscaling evaluation in this study. The discrepancies among these datasets should provide a measurement regarding the uncertainty in the assessment. In this paper, the models used for experimental design are presented in section 2. DDM capability in simulating upper general circulation and lower-level moisture transport using the MM5 is discussed in section 3. DDM performance for precipitation is discussed in section 4. Finally, section 5 includes the summary and discussions. 2. Model, Data and Experiments 2.1 Model description MM5 is a mesoscale model that uses a terrainfollowing vertical coordinate. For this study, 23 vertical levels were specified from the surface to 50-hPa, where a radiative boundary condition is used. Details of the model can be found in Dudhia (1989, 1993), Warner et al. (1992), and Grell et al. (1994). The planetary boundary-layer parameterization is based on Hong and Pan (1996), a nonlocal scheme originally employed in the Medium Range Forecast (MRF) model of the NCEP. Two cumulus parameterizations are used. The Grell is a simple convective parameterization and is constructed to avoid first-order sources of errors (Grell et al., 1994). Clouds are described as two steady-state circulations, caused by an updraft and a downdraft. The KF2 (Kain and Fritsch, 1993; Kain, 2004; Anderson et al., 2007) scheme is a new version of the Kain-Fritsch scheme and uses a sophisticated cloud scheme to determine entrainment and detrainment processes. A simple explicit treatment of cloud microphysics is employed (Dudhia, 1989). Both ice and liquid phases are permitted for clouds and precipitation, but mixed phases are not permitted. The model uses a radiation scheme in which longwave and shortwave radiation interact with the clear atmosphere, clouds, precipitation, and the ground (Dudhia, 1989). A biophysical model known as the Simplified Simple Biosphere Model (SSiB, Xue et al., 1991, 2001) is coupled with MM5 (Zhang et al., 2003). MM5 version 3.4 (Dudhia et al., 2005) coupled with SSiB is employed for this study. SSiB is a biophysically based model of land-atmosphere interactions and is designed for global and regional studies. It consists of three soil layers and one vegetation layer. The model is intended to realistically simulate the controlling biophysical processes and to provide fluxes of radiation, momentum, and sensible and latent heat to RCMs. The Oregon State University (OSU) LSM (Mahrt and Ek, 1984; Mahrt and Pan, 1984; Pan and Mahrt, 1987), which was coupled in the original MM5, is also applied in this study to test effects of land parameterization. This model has been extended by Chen et al. (1996). It was later improved further in the WRF regional model. Since this paper focuses on the impact of vegetation biophysical processes on the RCM DDM, we apply the original land parameterization, which has no vegetation processes, in MM5 to tackle this issue and refer to it as the original MM5 land model in this paper.

4 1080 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL Summer 1998 heavy precipitation event in China The June 1998 flood, which occurred in the Yangtze River Basin (26 32 N, E) during the monsoon/mei-yu season, was the most severe event recorded since It caused large economic losses in China. The flooding event was induced by severe precipitation events associated with the activities of the summer monsoon/mei-yu fronts over East Asia (Ding and Liu, 2001). The abnormal monsoon rainfall was also related to the prolonged impact of the strongest 1997/98 El Niño event (Ding and Liu, 2001). A full description of the 1998 EASM and severe precipitation event over China can be found in Ding and Liu (2001). This extreme event has been identified to be related to physics mechanisms at both planetary and local scales (Wang et al., 2003; Liu et al., 2006; Xu et al., 2008) such as two northward-propagating intraseasonal oscillation (ISO) events during the monsoon season in East Asia, a quasi-stationary front and an orographic uplift, and, thus, is an ideal case for evaluation of the DDM. This study focuses on downscaling ability for the June 1998 monsoon in China. Due to discrepancies in observational and reanalyses data, several datasets are used to evaluate the MM5 s downscaling ability for EASM. The datasets include: (1) observed daily precipitation data with 0.5 resolution interpolated from 743 stations in China referred to as OBS in this paper; (2) the Asian Precipitation Highly-Resolved Observational Data Integration towards Evaluation of the Water Resources (APHRODITE s Water Resources); The APHRODITE project develops state-of-the-art daily precipitation datasets with high-resolution grids (0.25 and 0.5 ) for Asia. The datasets are created primarily from a rain-gauge observation network, remotely sensed data, and geographic information (e.g., topography) over Asia, especially the Himalayas and the TP ( html). It greatly offsets the deficiency of the OBS data in western China and is used to evaluate the downscaling performance over the TP. (3) the NCEP- National Center for Atmospheric Research (NCEP- NCAR) Global Reanalysis (NCEP R-1: Kalnay et al., 1996). (4) the European Center for Medium- Range Weather Forecasts (ECMWF) global reanalysis (ERA40; Gibson et al., 1997). (5) The GEWEX/Asian Monsoon Experiment (GAME, Yamazaki et al., 2003) reanalysis. The GAME reanalysis has a horizontal resolution of 0.5 at 17 standard pressure levels and was created based on a data assimilation system with observational data from the GAME IOP (Intensive Observation Period, 1 April to 31 October 1998), GTS (Global Telecommunication System), and data from several other field experiments including TIPEX, GAME/Tibet, and HUPEX, as well as soundings and wind profiler data from Japan, Korea, Indonesia, Malaysia, Thailand, Vietnam, et al. 2.3 Experimental design The DDM experiments in this study are listed in Table 1. The simulations are conducted for the period of 1 30 June The NCEP R-1 at a 6-hour interval is used as both initial and LBCs for this study. Case 1 serves as the control case. The model domain covers dimensions of horizontal grid points with spacing of 30 km (Fig. 1, domain A). Cases 1 4 are centered at 30 N, 115 E, and Cases 5 7 are centered 5 10 northward and/or westward compared to Case 1 (see Table 1 and Fig. 1 for details). This domain consists of northward moisture sources from the South China Sea and the Bay of Bengal, the upper level westerly jet (ULJ) and low level jet (LLJ), the TP, and westward moisture sources from the eastern boundary, all of which have an impact on the EASM. Liang et al. (1997) North American RCM results are sensitive to the convective scheme. In this study, Cases 1 and 2 apply different cumulus convection parameterizations the Grell Scheme for Case 1 and KF2 for Case 2 and are used to assess the effects of convection schemes on EASM study. Furthermore, studies have also shown that surface variability affects environmental processes on meso and regional scales (Avissar and Pielke, 1989; Giorgi and Avissar, 1997; Pielke et al., 2002; Niyogi et al., 2006; Holt et al., 2006) because land surface characteristics exert an impor- 50N 40N 30N 20N 10N B IV V I III II 80E 90E 100E 110E 120E 130E Fig. 1. Experimental domains and topography (units: m) for Cases 1 (A) and 6 (B). Five sub-domains enclosed with bold lines are for DDM analysis. I: Analysis Area; II: Yangtze River basin; III: North China; IV: Northwest China; V: Tibetan Plateau. A

5 NO. 5 GAO ET AL Table 1. Experiment setting for the 1998 cases. Case Convective scheme LSM Initial date Domain centers Case 1 Grell SSiB 1 June 30 N, 115 E Case 2 KF2 SSiB 1 June 30 N, 115 E Case 3 Grell original MM5 1 June 30 N, 115 E Case 4 Grell SSiB Re-initialization every 10 days 30 N, 115 E Case 5 Grell SSiB 1 June 35 N, 115 E Case 6 Grell SSiB 1 June 35 N, 110 E Case 7 Grell SSiB 1June 40 N, 110 E tant control on the surface-atmosphere exchange and atmospheric processes (Xue et al., 2001; LeMone et al., 2002, 2006; Gao et al., 2008a, b). In this study, we apply the original MM5 land scheme in Case 3, in which vegetation maps and the corresponding vegetation and soil characteristics are also different from Case 1 accordingly. Therefore, the comparison between Case 1 and Case 3 shows the effects of land surface parameterization and land characteristics on downscaling ability. Long-term continuous integration with one single initialization of large scale fields and frequent updates of LBCs is currently the most common approach in regional climate simulation (Leung et al., 2003). We use this approach for most cases in this study. Due to lower skill with the one-initialization approach (e.g., Davies, 1976; Warner et al., 1992), re-initialization has been introduced to RCM simulations. This approach has been successful in weather forecasting to mitigate the problem of systematic error growth in long-term integrations. Some RCM studies also point out that this approach could be beneficial to RCM simulations (Druyan et al., 2001; Qian et al., 2003). Lo et al. (2008), after assessing the DDM over the U.S. based on WRF, concluded that a re-initialization run outperforms continuous integration. Their run with more frequent (weekly) re-initialization outperforms that with less (monthly) frequent re-initialization. Caldwell et al. (2009); however, they found that re-initialization interferes with the RCM s ability to develop realistic surface moisture. In this study, Case 4 is designed to reinitialize every 10 days, a medium forecast range between one day in Xue et al. (2001) and one month in the control run. Comparison between Case 1 and Case 4 can be used to assess the effect of frequency of initialization on DDM performance. The LBC influence on the DDM is illustrated well by Xue et al. (2007). Here, Cases 5 7 are set to illustrate the impact of southern and western boundaries on the DDM. Case 5 is centered at 35 N, 115 Ewith the same dimensions and spacing as Case 1 4. It extends the model domain center northward by 5 compared with that of Case 1 to reduce the coverage of the ocean, where reanalysis normally is not as reliable as over land (Liang et al., 1997). Case 6 extends the model domain westward by 5 compared to Case 5 to avoid Pamir Mountain as a lateral boundary (Fig. 1, domain B). Case 7 moves the domain center 5 farther northward with respect to Case 6 to avoid the South China Sea. In the following analyses of DDM experiments, we first assess the performance of the DDM on atmospheric circulation and then discuss the precipitation downscaling. Atmospheric circulation simulation is evaluated using the GAME reanalysis. Precipitation from rain gauge observation (referred to as OBS hereafter) is interpolated to 0.5, the same resolution as the GAME reanalysis. To address the simulation uncertainty, NCEP R-1 and ERA40 data at 2.5 resolution are included in the comparison and interpolated into the same resolution as GAME reanalysis. For precipitation analysis, the entire simulation domain is divided into several sub-regions. Sub-region I (26 45 N, E), a relatively large domain, is used to evaluate circulation. In addition, we set four subregions II-V (Fig. 1) for precipitation evaluation because each sub-region has different precipitation characteristics and the model performs differently for each domain. Statistics similar to those used by Giorgi et al. (1993a, b) and Wang et al. (2003) are utilized to assess the different aspects of model performance including the spatial pattern correlation coefficient (SC) between the model simulation and observation, model bias, and root mean squared errors (RMSE). 3. Circulation downscaling For the atmospheric circulation, three fundamental climate variables which control regional climate over East Asia are considered in the DDM evaluation: mid-latitude upper-level 200-hPa jet streams (ULJ), 500-hPa geopotential height, and moisture transport accompanying the southwesterly low-level jet (LLJ). 3.1 Reanalyses and control run simulation GAME reanalysis data, which exist on a highresolution spatial grid, have been widely used to diagnose the water and energy cycle around the TP in 1998 (e.g., Yamazaki et al., 2003). It is used as true

6 1082 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 data for the circulation here to evaluate coarse resolution reanalysis data and model-simulated atmospheric circulations. Figures 2a 5a show the monthly mean 200-hPa zonal wind (referred to as U200 in this paper), geopotential height at 500-hPa (H500), moisture transport at 850-hPa (VQ850), and a cross section of mean meridional wind along 25 N from GAME reanalysis, respectively. Performances based on the NCEP R-1 (Figs. 2b 5b), ERA40 (Figs. 2c 5c), and each DDM experiment from the GAME reanalysis are shown in Figs To quantify the model performance in the simulation of atmospheric circulation over sub-region I, we calculate the statistics for these fields and air temperature at 850-hPa (T850), including SC, biases (Er), and RMSEs (Table 2). Several global reanalysis results are presented in this paper. Since this study uses NCEP R-1 as the LBC, it is expected that the DDM should produce better features than, or at least keep the same features as, NCEP R-1 for large-scale circulation. We also show ERA40 to indicate the uncertainty in global reanalyses to help our assessment of DDM ability. Figures 2b 5b and 2c 5c and Table 2 show that U200, H500, and T500 for two reanalyses (i.e., NCEP R-1 and ERA40) are very similar and do not show systematic biases compared to the GAME reanalysis. For VQ850, NCEP R-1 is better than ERA40 in correlation but worse in bias. Figures 2d 5d show that Case 1 produces reasonable simulations. The 200-hPa ULJ moves northward, producing a slight positive bias over northeast East Asia and a slight negative bias to the south (Fig. 2d) compared to the GAME reanalysis. The bias of H500 produces a dipole pattern: a positive bias in the east and a negative bias over the TP (Fig. 3d). It inherits the H500 bias pattern from NCEP R-1. However, the area associated with the maximum biases is enlarged compared to the reanalyses. The stream flow at 850-hPa is well simulated, but moisture transport is quite weak, although the pattern is consistent with the reanalysis (Fig. 4d). Figure 5d shows that Case 1 reproduces very well the LLJ to the west slope of the TP at around 97 E, the LLJ to the east of the TP, and the ULJ at 100 E, superiorly to global reanalyses due to the RCM s high resolution. In addition, a jet around 600-hPa at the top of the TP is also well simulated (Fig. 5d). The values for Case 1 listed in Table 2 are generally lower than reanalyses values, except for the spatial correlation of VQ Impacts of physical parameterizations In this section, we focus on the effects of two physical parameterizations: convective scheme and land parameterizations Convection schemes The simulation with the Grell cumulus convection scheme obtains a much higher correlation coefficient and lower RMSE for U200 (Table 2). The bias contour lines of 8 m s 1 over both North China and the area between 10 Nand20 N appear in the simulation with the KF2 scheme but not with the Grell scheme (Figs. 2d and 2e). For the H500 simulation, although the KF2 scheme reduces the area of positive bias, the area with negative bias is expanded compared with the Grell Scheme (Figs. 3d and 3e). The KF2 cumulus convection scheme (Case 2) and Grell scheme (Case 1) produce similar results for air temperature at 500-hPa and 850-hPa (not shown). On the other hand, the KF2 scheme improves the simulation of moisture transport at low levels greatly in terms of the correlation coefficient (Table 2). Stronger moisture transport is produced by the KF2 (Figs. 4d and 4e). Positive bias and RMSE of meridian wind speed are greatly reduced by the KF2 cumulus convection scheme compared to the Grell scheme (Table 2) and are even better than the reanalyses, consistent with other Indian downscaling studies (Ratnam and Kumar, 2005; Mukhopadhyay et al., 2010). The two cumulus schemes in this study show different dynamic downscaling abilities for circulation at different heights. While the KF2 scheme enhances downscaling ability for low-level moisture transport, Grell generally produces better simulations for the upper level circulation, which are related to their parameterizations. The KF2 scheme uses a Lagrangian scheme of a one-dimensional, entraining-detraining, steady-state plume with downdraft to determine entrainment and detrainment processes (Kain and Fritsch, 1993; Kain, 2004). The vertical profile of cloud mass is supplied to the host model for explicit simulation through the grid-resolved dynamical equations and parameterized microphysical processes. The combination of high resolution and the treatment of vertical profile of cloud mass results in more frequent and more realistically clustered propagating of nocturnal mesoscale precipitation events (Kain, 2004; Anderson et al., 2007), which should be particularly useful for the low-level atmosphere with a sub-grid process. The Grell scheme (Grell, 1993; Grell et al., 1994), on the other hand, has no direct mixing between cloudy air and environmental air, except at the top and the bottom of the atmosphere, which is related to the large scale circulation. Mass flux is constant with height, and there is no entrainment or detrainment along the cloud edges. It was constructed to use several closure assumptions to avoid first-order sources of errors, of which the quasiequilibrium assumption related the strength and the location of the convection to the larger-scale destabi-

7 NO. 5 GAO ET AL Fig. 2. (a) 200-hPa zonal wind for GAME reanalysis, (b, c) bias of monthly mean 200-hPa zonal wind for NCEP R-1/ERA40 reanalyses and (d j) seven DDM cases compared with GAME at June 1998 (units: m s 1 ).

8 1084 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 Fig. 3. (a) 500-hPa geopotential height (zonal mean removed) for GAME reanalysis, (b, c) bias of monthly mean 500-hPa geopotential height for NCEP R-1/ERA40 reanalyses and (d j) seven DDM cases compared with GAME at June 1998 (units: 10 m).

9 NO. 5 GAO ET AL Fig. 4. Water vapor flux (units: g kg 1 ms 1 ) and stream at 850-hPa for GAME/NCEP R- 1/ERA40 reanalyses and DDM cases at June 1998.

10 1086 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 Fig. 5. Cross section of mean meridional wind (m s 1 ) (contour lines) along 25 Nfor(a) GAME reanalysis, (b) NCEP R-1, (c) ERA40, and (d j) DDM cases. The black shade stands for terrain.

11 NO. 5 GAO ET AL Table 2. Bias (Er), RMSE, and Correlations (SC) for June between the GAME reanalysis and simulations of atmospheric variables for June 1998 over sub-region I (26 45 N, E, Fig. 1). U200 H500 SC Er RMSE SC Er RMSE NCEP R ERA Case Case Case Case Case Case Case T850 VQ850 SC Er RMSE SC Er RMSE NCEP R ERA Case Case Case Case Case Case Case Note: Units: U200: m s 1 ; H500: 10 m, T850: C; and VQ850: g kg 1 ms 1. Zonal mean was removed for the first 3 variables. lization. It is likely that these treatments help the large-scale simulation. Two convective schemes performances in this study are consistent with their parameterization and with other studies focusing on North American downscaling (e.g., Liang et al., 2004), which also found that, over the Great Plains where the timing of the diurnal cycle of convection is mainly influenced by large-scale vertical motion, Grell produces a more realistic simulation, whereas the KF scheme works better in the Southeast U.S. where convection is largely governed by near-surface forcing Land surface parameterizations The difference between Case 1 (SSiB land scheme) and Case 3 (original MM5 land scheme) shows the effect of land surface parameterizations in the DDM. The different physical surface processes and the different initialization of the land surface state variables affect the surface water and energy balances, which would, in turn, modify the atmospheric conditions and precipitation (Xue et al., 2004). It is interesting to note that the land parameterizations have a substantial impact on high-level circulation as shown in Xue (1996) and Xue et al. (2001, 2004). Case 3 is not able to catch the upper-level 200-hPa jet stream s pattern very well for the simulation region (Figs. 2d and 2f, Table 2), and it also shows an area with a contour line of positive bias of 8 m s 1 over the northern part of the domain, which does not appear in Case 1. Furthermore, the area with a negative bias contour line of 8 ms 1 over South China is much larger than in Case 1 (Figs. 2d and 2f). Previous land-atmosphere interaction studies using the SSiB model (Xue et al., 2001, 2004) have demonstrated that land surface parameterizations affect high-level circulation via its impact on the turbulence exchange, then PBL evolution. For instance, in an RCM study (Xue et al., 2001), it has been found that an unstable PBL in the RCM-SSiB extends much higher in the daytime, which helps transfer surface fluxes to high atmospheric levels. In another East Asian study (Xue et al., 2004), it has been found that, without SSiB, the area to the south of 30 Nin East Asia was a heat sink and the area to the north of 30 N was a source. In contrast, the entire East Asian continent was a heat source in the case with SSiB. A different heat gradient produces very different circulation through geostrophic balance. In this study, Case 1 and Case 3 also reflect similar differences in simulated surface heating (not shown) and gradients of

12 1088 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 geopotential height in southeastern China (Fig. 3f) as shown in Xue et al. (2004) for results with and without SSiB. The unrealistically larger gradient of the zonal wind in Case 3 at 200-hPa at 35 N is consistent with the large gradient of geopotential height there. For H500, as discussed above, the gradients of geopotential height in southeastern China are stronger in Case 3 than in Case 1. Although the positive bias over the northeastern domain is reduced compared to Case 1, the negative bias increases greatly over the western domain in Case 3 (Figs. 3d and 3f). Overall, bias and RMSE of H500 in Case 3 are higher (Table 2). Case 3 s positive bias for T500 over eastern East Asia is less than in Case 1, and the warm center over the TP is well produced in Case 3 but not in Case 1 (not shown). Table 2 shows that Case 3 produces a slightly higher correlation for air temperature at 850- hpa than Case 1. In addition, the ability for moisture transport in Cases 1 and 3 is similar (Figs. 4d and 4f). The meridian wind speed to the east of the TP is unrealistically higher in Case 3 than in Case 1, and the jet at the top of the TP at 600-hPa is weak in Case 3 compared to reanalysis (Figs. 5d and 5f). Comparing the differences between Case 1 and Case 2 and between Case 1 and Case 3, it is shown that convective schemes produce larger differences for the variables at 850-hPa and the land parameterizations for those at 200-hPa, respectively (Table 2). But both parameterization effects on the DDM are substantial. 3.3 Impacts of 10-day re-initialization and location of lateral boundaries (LLB) Case 4 reinitialized the simulation every 10 days. Based on Figs. 2d 5d (Case 1) and 2g 5g (Case 4) and Table 2, it is clear that 10-day re-initialization does not lead to improvement in the monthly mean regional circulation simulation. It does not yield a higher correlation coefficient or smaller RMSE and bias either for ULJ or middle-level geopotential height (Table 2). Moisture transport from the southwesterly LLJ is not improved either. Cases 5 7 test the impact of LLB on downscaling ability. Case 5, which moved the southern boundary northward, produces the poorest performance with the largest biases and RMSEs of all 7 cases for this region for all the levels, from the upper atmospheric level to the low level (Table 2 and Figs. 2h 5h). At 200 hpa, it produces a great positive bias over North China and a negative bias to the south (Fig. 2h). For H500, it produces the worst positive bias over eastern East Asia, where most areas have more than a 20-m bias, while for other cases; the bias is only around 5 10 m (Fig. 3h). This case indicates that the northward move of the lateral boundary, which reduces the South China Sea area, does not help the DDM, different from a North American DDM study (Xue et al., 2007) in which the reduction of the Gulf of Mexico in the southern boundary area greatly improved the downscaling results. In Case 6, moving the western boundary farther to the west improved the simulation substantially. For VQ850, Case 6 produces the best performance among all cases, even better than NCEP R-1 and ERA40 (Table 2). The pattern of moisture transport at the low level is vastly improved (Fig. 3i). Although Case 6 s other statistics shown in Table 2 are similar to those of the other cases, a careful examination reveals that this case greatly improves the spatial distribution of some variables besides the VQ850 (Figs. 2i 5i). Compared to Case 1, it substantially reduces the simulation bias for zonal wind at 200-hPa (Figs. 2d and 2i) and reduces the positive bias of H500 over eastern East Asia to a large extent although the negative bias over northeast East Asia is increased (Figs. 3d and 3i). That indicates that inclusion of the TP plays an important role in producing a proper circulation pattern of EASM and should not be ignored. The TP has been considered to be an important factor in East Asia regional simulation (Ding et al., 2006). Most of the previous EASM simulation domains have their western boundary located at E, which passes through the eastern periphery of the TP. In doing so, the effects of both dynamical and thermal forcing are not properly considered, especially their effects on the East Asian monsoon circulation and the rain-producing weather systems which originate in the TP and move eastward along the Yangtze River basin. Furthermore, differences in the representation of surface topography between the coarseresolution and regional models can produce gravity inertia waves that propagate to the interior of the domain so that regional simulations can be seriously affected (Leung et al., 1999). Case 6 confirms the discussions (or discoveries or hypotheses) presented in these studies. Case 7 has similar results to Case 6, but they are consistently slightly worse because its domain moves farther away from the moisture source. This study reveals that the western LLB has a crucial impact on DDM performance in East Asia. 3.4 Discussion In all seven simulation cases, Case 1 has the highest correlation coefficient for the upper-level 200-hPa jet stream (ULJ), 0.91, while Case 5 has the lowest correlation, 0.16 (Table 2), with a strong northward shift of the jet (Fig. 2h). RMSE in all seven cases is much larger than in the reanalyses. Only Cases 1, 3, and 4 show smaller biases than the reanalyses (Table

13 NO. 5 GAO ET AL ). The DDM, in general, could not produce better results for the upper-level 200-hPa jet stream compared with reanalyses in our cases. For H500 at the middle level, all seven cases produce very high spatial correlation, comparable with reanalyses. Case 1 s bias is smaller than in the reanalyses, and the RMSE is close to those in the reanalyses (Table 2). The DDM with the MM5-Grell-SSiB could generally reproduce 500-hPa geopotential height comparable to the reanalyses. At the low level, the RMSE and spatial correlations with the DDM could not match reanalyses results in T850 (Table 2). Only Case 3 has T850 results comparable to the reanalysis and has the best performance amongst 7 cases. The most substantial improvement in the DDM is the moisture transport at the low level. Correlation coefficients for VQ850 between GAME and NCEP R- 1 and between GAME and ERA40 are 0.81 and 0.71, respectively (Table 2). Most DDM cases, except Case 5, obtain higher correlation coefficients with GAME than the reanalyses. In addition, Cases 2, 6, and 7 produce lower RMSE than the reanalyses (Table 2). 4. Precipitation downscaling 4.1 Characteristics of June 1998 precipitation of observation, reanalyses, and the control run simulation Figure 6 presents the June 1998 precipitation for OBS, the APHRODITE dataset, GAME, NCEP R- 1, and the ERA40 reanalysis. There are two rainfall bands in South China and two rainfall centers in North China in the OBS figure (Fig. 6a). The greatest W- E rainfall belt is located over South China extending between N and E. The second rainfall band is located along the southern coast. There is a rather weak NW-SE rainfall band along the River West Corridor (Silk Road) centered at 37 N, 100 E. A relatively smaller and weaker center is located in North China (40 N, 117 E). Meanwhile, a south-north precipitation gradient over the TP is apparent in Fig. 6a. This gradient was confirmed by station records over the TP (Yang et al., 2007a, b). The rainfall center at the south slope of the TP as shown in the reanalyses and the APHRODITE dataset, however, is not represented well in Fig. 6a due to the rain gauge coverage. We believe that precipitation over the south TP is underestimated in Fig. 6a. In general, reanalysis underestimates the monsoon rainfall in South China and overestimates the rainfall over North China. Compared to the other two reanalyses, GAME has a higher correlation for West China and a smaller bias over the TP due to its higher resolution. To better evaluate DDM performance, four subregions are divided for further investigation (Fig. 1): sub-region II: Yangtze River basin (26 32 N, E), sub-region III: North China (32 42 N, E), sub-region IV: Northwest China (36 42 N, E), and sub-region V: TP (28 36 N, E). These four sub-regions have different climate characteristics: sub-region II is in the mei-yu monsoon season with severe rainstorms; sub-region III is still in the pre-monsoon stage. Sub-region IV is an arid/semiarid region with complex topography. Subregion V covers the eastern TP with a high altitude gradient. The correlation coefficient, bias, and RMSE for the GAME, NCEP R-1, ERA40, and 7 case simulations compared with OBS are shown in Table 3. APHRODITE has a very high correlation coefficient with OBS, about 0.9, for the Yangtze River basin and North China (Table 3). In Northwest China, the spatial correlation coefficient between APHRODITE and OBS is not as high as in the two former sub-regions (Table 3), only The topography rainfall center along the Silk Road Corridor in Northwest China is not obvious in the APHRODITE data (Fig. 6b). Since rain gauges are sparse over the TP and the interpolation there is unreliable (Fig. 6a), the APHRODITE precipitation, which had higher correlation with OBS in other sub-regions, is used to evaluate the reanalyses and DDM performance in sub-region V. Case 1 s result indicates that the MM5 simulation could not match reanalysis for severe rainstorms over the Yangtze River basin and the pre-monsoon season over North China. The likely explanation for the low correlation of precipitation over East Asia for Case 1 is the meridional wind simulation at the low atmospheric level. Figure 5d shows that the simulated northward meridional wind is stronger than in the reanalysis between E and weaker to the east of 120 E. This stronger northward flow in the west and weaker northward flow in the east make the rain belt turn clockwise into more of a W-E direction, rather than the NE-SW direction in the observation. For Northwest China, the correlation between Case 1 and OBS is 0.74, similar to GAME and higher than the two global reanalyses. The RMSE and Bias with Case 1, however, are smaller than with the GAME reanalysis, but larger than with the NCEP R-1 and ERA40 (Table 3). Over the TP, Case 1 has a higher correlation with APHRODITE s precipitation than with all of the reanalyses and a smaller or comparable bias to NCEP R-1/ERA40 (Table 3). In general, the regional DDM could yield similar or better performance than reanalysis in Northwest China and over the TP, where the topographic effect is important, and

14 1090 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 Fig. 6. (a) observed precipitation, (b) APHRODITE dataset, (c, d, e) GAME/NCEP R-1/ERA40 reanalyses at June 1998 (unit: mm m 1 ). could not match the reanalyses for South and North China, where the precipitation is mostly controlled by moisture flux transport. 4.2 Impacts of physical parameterizations Case 2 (with the KF2 scheme) has a higher correlation with OBS for the Yangtze River basin (0.47) and for North China (0.5) compared to Case 1 (only 0.27 and 0.42, respectively) but is still worse than the reanalyses (see Table 3 and Fig. 7b). In Northwest China and the TP, Case 2 has similar or slightly higher correlations with OBS and APHRODITE s precipitation (0.75 and 0.78, respectively), compared to Case 1. This is related to the great improvement of the simulation of moisture transport at low levels from the KF2 scheme, as indicated in their correlation coefficients (Table 2). Enhanced moisture transport in Case 2 is consistent with the reduced bias and RMSE of meridian wind speed by the KF2 cumulus convection scheme compared to the Grell scheme (Table 2). The bias and RMSE in precipitation for Case 2, however, show no substantial difference from Case 1. In general, the KF2 cumulus convection scheme (Case 2) produces better results for the spatial distribution of precipitation, similar to Mukhopadhyay et al. (2010) and Ratnam and Kumar (2005) studies over India and Liang et al. (2004) study in the Southeast U.S. Case 3 (with original MM5 land parameterization) produces the lowest correlation with OBS and APHRODITE except over North China. This is because it misses the heavy rainfall band located to the south of the Yangtze River (Fig. 7c); the rainfall band

15 NO. 5 GAO ET AL Table 3. Bias (Er), RMSE, and Correlations (SC) for June between observation and reanalyses and simulated precipitations (mm m 1 ) over different sub-regions (Fig. 1). Sub-region II (26 32 N, E) Sub-region III (32 42 N, E) SC Er RMSE SC Er RMSE APHRODITE GAME NCEP R ERA Case Case Case Case Case Case Case Sub-region IV (36 42 N, E) Sub-region V (28 36 N, E) SC Er RMSE SC Er RMSE APHRODITE GAME NCEP R ERA Case Case Case Case Case Case Case shifts to the north of the river. For Northwest China and over the TP, the rainfall center shifts more to the west of the TP than in other cases. Only in North China does it have a higher correlation with OBS but a larger RMSE than other cases. The mismatch in the rainfall band is consistent with the deficiency in simulating the large-scale circulation pattern as discussed earlier. The zonal wind in Case 3 is rather weak (Fig. 2f). The stronger northward meridional wind (Fig. 5f) pushes the warm, wet air farther northward, which causes the mismatch of the rain band and low correlation with OBS. Comparison between Case 3 and Case 1 indicates that land parameterization has a great impact on the regional precipitation position simulation. 4.3 Impacts of re-initialization and location of lateral boundaries Case 4 simulates the rainfall center located at 24 N, 110 E over the Yangtze River. The relatively smaller and weaker center is located in North China (40 N, 117 E) more clearly than in Case 1 (Fig. 7d). Case 4 produces a higher correlation with OBS, 0.41, for sub-region II than Case 1 (0.27). It has the highest correlation in North China with OBS among the 7 DDM experiments, which is 0.53 (Table 3). However, the correlations are still lower than in the reanalyses. For Northwest China and over the TP, the weak NW- SE rainfall band along the River West Corridor (Silk Road) is correctly centered at 37 N, 102 Eandthe south-north precipitation gradient over the TP is well situated also (Fig. 7d). Case 4 s correlations with OBS and APHRODITE in these two sub-regions are comparable to or better than those of the reanalyses, but similar to those of Case 1. The bias and RMSE in Case 4 are generally smaller than in Case 1 and NCEP R-1 (Table 3) for these two sub-domains. Apparently, the 10-day re-initialization does lead to some improvement in the simulation of precipitation spatial distribution for North China and improvement in the bias and RMSE for Northwest China and over the TP with its complex topography by refreshing LBC every 10 days. In the LLB experiment, Case 5 has worse performance for precipitation than the others, consistent with its circulation simulation. It misses the heavy rainfall band located to the south of the Yangtze River (Fig. 7e), which shifts to the north of the river. The location of the topographic rainfall center in North-

16 1092 ASSESSMENT OF DYNAMIC DOWNSCALING OVER EAST ASIA USING A RCM VOL. 28 Fig. 7. Simulated precipitation for 7 DDM cases at June 1998 (units: mm m 1 ).

17 NO. 5 GAO ET AL Fig. 8. Temporal evolution of the daily precipitation (mm d 1 ) averaged over Efor(a) observation; (b) GAME; (c) ERA40; (d j) 7 DDM cases at June west China and over the TP in Case 5 extends more to the west than with OBS or APHRODITE, reducing the correlation coefficients (Table 3). A northward shift of the domain causes less water transportation from the southern boundary (Fig. 4h). Together with the much weaker eastward zonal wind (Fig. 4h), water vapor is transferred farther westward. After shifting the western boundary westward, however, Case 6 obtains the highest correlation coefficient among the 7 DDM experiments, 0.60 (Table 3) for the Yangtze River, comparable to the GAME and NCEP R-1 reanalyses (Table 3). It is again consistent with the circulation simulation. Case 7 has similar performance to Case 6. Inclusion of the TP plays an important role in producing South China s precipitation pattern. More water vapor reaches South China by air flow along the south edge of the TP (Fig. 4i). However, the westward shift of the domain (Cases 6 and 7) deteriorates the DDM s ability for North China (Table 3 and Figs. 7e g). It seems that westward shifting of the western domain boundary causes more easterly moisture flux towards South China while reducing the flow to North China. In Northwest China and the TP, the correlation of Case 6, 7 with OBS/APHRODITE ranges from 0.74 to 0.76, similar to that of Case 1, which is close to GAME s performance but higher than the two global reanalyses. The bias and RMSE in Case 6 are similar to those in Case 1. By and large, a westward shift of the LLB produces better spatial distribution of precipitation in South China with monsoon precipitation, while it deteriorates the DDM s performance for North China in the pre-monsoon season. 4.4 DDM performance of monsoon evolution Figure 8 shows the zonal mean rainfall temporal evolution averaged over E. The obser-