Assimilation of conventional and satellite wind observations in a mesoscale atmospheric model for studying atmospheric dispersion

Atmospheric Environment 44 (2010) 2846e2864

Contents lists available at ScienceDirect

Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv

Assimilation of conventional and satellite wind observations in a mesoscale atmospheric model for studying atmospheric dispersion C.V. Srinivas*, R. Venkatesan, V. Yesubabu, C. Nagaraju, K.M. Somayajai, P. Chellapandi, Baldev Raj Radiological Safety Division, Safety Group, Indira Gandhi Centre for Atomic Research, Kalpakkam 603102, India

a r t i c l e i n f o

a b s t r a c t

Article history: Received 4 March 2010 Received in revised form 28 April 2010 Accepted 29 April 2010

A mesoscale atmospheric model PSU/NCAR MM5 is used to provide operational weather forecasts for a nuclear emergency response decision support system on the southeast coast of India. In this study the performance of the MM5 model with assimilation of conventional surface and upper-air observations along with satellite derived 2-d surface wind data from QuickSCAT sources is examined. Two numerical experiments with MM5 are conducted: one with static initialization using NCEP FNL data and second with dynamic initialization by assimilation of observations using four dimensional data assimilation (FDDA) analysis nudging for a pre-forecast period of 12 h. Dispersion simulations are conducted for a hypothetical source at Kalpakkam location with the HYSPLIT Lagrangian particle model using simulated wind field from the above experiments. The present paper brings out the differences in the atmospheric model predictions and the differences in dispersion model results from control and assimilation runs. An improvement is noted in the atmospheric fields from the assimilation experiment which has led to significant alteration in the trajectory positions, plume orientation and its distribution pattern. Sensitivity tests using different PBL and surface parameterizations indicated the simple first order closure schemes (Blackadar, MRF) coupled with the simple soil model have given better results for various atmospheric fields. The study illustrates the impact of the assimilation of the scatterometer wind and automated weather stations (AWS) observations on the meteorological model predictions and the dispersion results. Ó 2010 Elsevier Ltd. All rights reserved.

Keywords: Mesoscale Assimilation AWS QuickSCAT FDDA Radiological dispersion

1. Introduction Modeling the environmental dispersion of radioactive effluents is an essential aspect in nuclear emergencies to assess the radiological consequences to the members of the public. Real-time estimates of transport and diffusion of hazardous contaminants is required for response management in such situations (Knox et al., 1981). Three-dimensional dispersion modeling tools have been developed and used for forecasting the radiological and chemical releases from normal and off-normal scenarios (Dickerson et al., 1979). Often it is needed to determine the dose rates resulting from the atmospheric transport and dispersion beyond a limited zone called the off-site range and long ranges extending up to a few hundreds of kilometers. In these ranges a numerical weather prediction model is generally used to predict the meteorological fields needed in dispersion assessment (e.g., Satomura et al., 1994; Lagzi et al., 2004). The dispersion results are influenced by the

* Corresponding author. Tel.: þ91 44 27480062. E-mail address: [email protected] (C.V. Srinivas). 1352-2310/$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2010.04.051

meteorological fields and the uncertainty in the estimated concentration or dose rates depends to a large extent on the accuracy of the meteorological fields apart from the uncertainty in the source term and dispersion parameters. Essentially the physical processes of the atmospheric transport of the contaminants and their mixing, dilution and deposition need to be precisely calculated which depend on many meteorological inputs. Outputs from mesoscale atmospheric models that directly influence the dispersion simulations include the wind field, temperature profiles, water vapour mixing ratio, boundary layer depth, turbulence, surface pressure and rainfall/precipitation in the lowest 2 or 3 km (Hanna, 1994; Seaman, 2000). The performance of atmospheric models depends on the accuracy of the initial meteorological fields and the ability of the model to realistically simulate atmospheric physical and dynamical processes. Assimilation of observations through suitable methods improves the atmospheric predictions needed in air quality and emergency response systems for decision support (Seaman, 2000; Zheng et al., 2007; Kovalets et al., 2003). Several methods are developed on data assimilation in recent times, such as Four Dimensional Data Assimilation (FDDA), 3DVAR etc. FDDA is a continuous data assimilation

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

technique in which the model state is relaxed toward the observed state by augmenting some of the prognostic equations with forcing terms based on the difference between the observed state and the model state (Stauffer et al., 1991; Stauffer and Seaman, 1990, 1994). The land based surface automated weather stations (AWS) data on wind, temperature and humidity and the 2-D wind data from satellite platforms such as QuickSCAT over the oceanic region can be used to define the model initial conditions accurately. A non-hydrostatic mesoscale atmospheric model MM5 developed by PSU/NCAR (Grell et al., 1995) is used for near real-time atmospheric predictions for application in a nuclear emergency response decision support system (Srinivas et al., 2006) for the Kalpakkam nuclear site situated on the southeast coast of India. The performance of this model with assimilation of observations is tested in this study for the prediction of lower atmospheric fields used in plume dispersion. In recent times more land based surface observations are becoming available over India for the purpose of precise short-range weather prediction (Rao, 2008). Assimilation of these observations needs to be tested in mesosclae models for their utility in dispersion assessment applications and emergency response systems. In the present study observations from these mesonet stations and wind observations from QuickSCAT satellite are assimilated in the MM5 atmospheric model to examine their impact on the predicted meteorological quantities and plume dispersion in a mesoscale range in the southeast coast of India. Model simulations are conducted for four cases in different seasons to study the impact of observational assimilation on the model performance. The performance of the model is also tested by conducting a few separate sensitivity experiments with two PBL and surface physics options. 2. Methodology 2.1. Description of the meteorological model The Pennsylvania State University/National Centers for Atmospheric Research (PSU/NCAR) mesoscale model MM5 is used in the present study to generate the meteorological fields in dispersion simulations. The MM5 model has Arakawa-B horizontal grid staggering, terrain-following sigma vertical coordinate, a second-order leapfrog time integration scheme, nesting of multiple domains, and has a number of parameterization schemes for atmospheric physical processes. In the present study the model is configured with four nested domains with 36, 12, 4 and 1.33 km horizontal resolution respectively. A total of 32 vertical levels are used in all the three domains. The outer domain has 90  90 grids, the second domain has 112  112 grids, the third domain has 142  142 grids and the fourth domain has 157  157 grids, the fine mesh covers the southeast coastal Kalpakkam region and its neighboring area (Fig. 1). The inner domains 2, 3 and 4 are two-way interactive. An explicit cloud microphysics parameterization option (Dudhia, 1989) is used to predict the grid scale cloud and rain water mixing ratio and cloud water vapour. On the model outer grids (domains 1, 2) the Grell (Grell et al., 1991) cumulus parameterization scheme is used to account for the large-scale convective processes. The atmospheric radiation schemes (Dudhia, 1989) and RRTM (Mlawer et al., 1997) are used to represent the shortwave and long wave processes that interact with the atmosphere, cloud and precipitation and with the surface. A number of parameterization schemes for PBL and land-surface physics are available in MM5. The land-surface and PBL parameterizations are influential for the simulation of winds, turbulence and other state variables in the lower atmosphere where dispersion and transport of pollutants occurs. In this study the Blackadar high resolution scheme (BK) (Blackadar, 1976; Zhang and Anthes, 1982),

2847

medium range forecast (MRF) non-local PBL turbulence diffusion scheme (Hong and Pan, 1996), Eta MelloreYamada (MY) level 2.5 TKE scheme (Mellor and Yamada, 1982; Janjic, 1996, 2002), GaynoeSeaman (GS) (Shafran et al., 2000), Burk and Thomson (BT) (Burk and Thomson, 1989), and Asymmetric convective model (ACM) (Pleim and Xiu, 1995) are used for PBL turbulence. The Blackadar is a high resolution PBL scheme with 5 layers. The surface fluxes of heat and moisture are computed based on standard similarity theory, where the friction velocity is derived from the wind speed and stability functions. Four stability categories viz., stable, mechanically induced turbulence, unstable (forced convection) and unstable (free convection) are considered separately, which are derived form Bulk Richardson number. The MRF scheme is a non-local first order scheme in which the vertical transfers are dependent on the bulk characteristics of the PBL and includes counter gradient transports of temperature and moisture arising from large-scale eddies. The eddy diffusivity coefficient for momentum is a function of the friction velocity and the PBL height, while those for temperature and moisture are computed using a Prandtl number relationship. The friction velocity and the surface exchange coefficients for heat, moisture and momentum are calculated through the surface layer similarity theory. The MYJ scheme includes a prognostic equation for turbulent kinetic energy (TKE), a level 2.5 turbulence closure approximation to determine eddy transfer coefficients and uses local vertical mixing within PBL. The ACM is a combination of the high resolution Blackadar model and an eddy diffusion model. It computes eddy diffusion in the stable conditions, both local and non-local transport in unstable conditions. The land-surface models (LSMs) use atmospheric information from the surface layer scheme together with the landsurface properties (defined by land use, soil type etc.) to compute eddy heat and moisture transports in the PBL. Four well-suited LSM schemes available in MM5 are chosen in the study. These are the 5layer soil thermal diffusion (SOIL) model (Dudhia, 1996), the Noah land-surface model (NOAH) (Chen and Dudhia, 2001) and the PleimeXiu (PX) LSM (Xiu and Pleim, 2001). The 5-layer soil model solves the thermal diffusivity equation with 5 soil layers. The energy budget includes radiation, sensible and latent heat fluxes. It treats the snow-cover, soil moisture fixed with a land use and season dependent constant value. The Noah LSM treats explicit soil and vegetation effects. It uses the time dependent soil fields and uses a 4-layer soil temperature and moisture model with canopy moisture and snow-cover prediction. The PleimeXiu LSM includes a 2-layer force-restore soil temperature and moisture model and considers evapotranspiration, soil evaporation, and evaporation from wet canopies. The PX LSM is coupled to the ACM PBL. The BT scheme includes a force-restore surface temperature prediction scheme. The BK and GS schemes are used with SOIL scheme. For the control and assimilation runs the MRF PBL along with the SOIL model are used. A set of eight experiments are conducted separately to test the sensitivity of the model to different PBL and surface schemes. The sensitivity experiments are denoted as MRFSOIL, MRFNOAH, MYSOIL, MYNOAH, BKSOIL, BT, GSMSOIL, ACMPX respectively. The options used in the model are given in Table 1. 2.2. Meteorological observations A dense mesonet is installed by the Indian Space Research Organisation (ISRO) in India to improve short-range weather prediction (Rao, 2008). This mesonet consists of a network of about 300 AWS distributed allover India and provide real-time observations on surface weather parameters of wind speed and wind direction, air temperature and relative humidity. In the present study these AWS observations, the upper-air radiosonde

2848

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 1. Modelling domains used in MM5 in the study.

observations from India Meteorological Department (IMD) are used along with 2-d wind data over the Bay of Bengal region from QSCAT satellite for assimilation study. The first guess initial conditions for the MM5 model are obtained from the National Centers for Environmental Prediction (NCEP) Final Analysis (FNL) data. Locations of a few observations sites in the domains (3,4) are shown in Fig. 2. Data from the stations Gadanki, Arakonam, Gumminipundi, and Kanchipuram are used for assimilation while the data from Tirupati, Puttur, Kalpakkam, Taramani, Pulicatnagar, and SHAR Sriharikota is used for comparison.

2.3. Model initialization Numerical experiments are conducted using the MM5 model for 4 cases in 21e23 May’08, 24e26 Jan’09, 15e17 Apr’09 and 22e24 Sep’09 which are selected based on the availability of observations. For the cases selected in Jan, Apr, and Sep’09 GPS Sonde field meteorological experiment was conducted at Kalpakkam. The first run for each case is a control run in which a static initialization of the MM5 is performed using the National Centers for Environmental Prediction (NCEP) 6 hourly final analysis (FNL) data. The FNL

Table 1 Details of the model domains and physics used in the MM5 model. Dynamics

Primitive equation, non-hydrostatic

Vertical resolution

32 sigma levels (1.00, 0.995, 0.99, 0.985, 0.980, 0.975, 0.97, 0.965, 0.96, 0.55, 0.95, 0.94, 0.93, 0.91, 0.89, 0.85, 0.80, 0.75, 0.70, 0.65, 0.60, 0.55, 0.50, 0.45, 0.40, 0.35, 0.30, 0.25, 0.20, 0.15, 0.10, 0.05, 0.0) Domain1 Domain2 Domain3 Domain4 36 km 12 km 4 km 1.33 km 90  90 112  112 142  142 157  157 62.61e95.37 E; 0.80e30.15N 73.69e86.40 E; 6.49e18.60N 77.04e82.31 E; 10.22e15.33N 78.88e80.80 E; 11.81e13.69N Dudhia scheme for shortwave and RRTM for longwave processes NCEP FNL analysis data Grell scheme on the outer grids domain1, domain2 Simple ice (SI) scheme CTL, FDDA runs e MRF PBL, 5-layer soil model Sensitivity runs e MRF PBL, 5-layer soil model (MRFSOIL) MRF PBL, Noah LSM(MRFNOAH) Eta PBL, 5-layer soil model (MYJSOIL) Eta PBL, Noah LSM (MYJNOAH) Blackadar, 5-layer soil model (BKSOIL) BurkeThomson (BT), GaynoeSeaman, 5-layer soil model (GSSOIL), Asymmetric Convective Model, Pleim-Xiu LSM (ACMPX)

Domains Horizontal resolution Grid points Domains of integration Radiation Sea surface temperature Convection Explicit moisture PBL turbulence Surface processes

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 2. Locations of the AWS stations used in model comparison (shown in circles) and receptor points used in Hysplit model (shown in boxes).

data is interpolated to the model grids on mandatory and supplementary pressure levels. The model starts at 00UTC 21 May, 00UTC 24 Jan, 00UTC 15 Apr and 00UTC Sep 22 in the respective cases and is integrated for 48 h. In the second experiment for each case the model is dynamically initialized in which data are assimilated into the model for a 12 h pre-forecast period. Analysis nudging method is used in this work. The observations from AWS, QuickSCAT winds and upper-air radiosonde are merged with the first guess fields using standard Cressman objective analysis procedure for the model domains 1 and 2. The resulting improved analysis is used for analysis nudging during a 12-h pre-forecast initialization period in each case. In the analysis nudging experiment surface and 3d analysis nudging is performed for winds, temperature and humidity over the model first and second domains. The values for nudging coefficients are specified as 2.50e4 for wind speed and direction, 1.0e4 for temperature and 1.0e5 for mixing ratio. These are the default values for nudging available with the code and correspond to a time scale of 1 h. Four more experiments are conducted with two different PBL and LSM options as given in Table 1 to test the model physics for the prediction of surface fields.

2.4. Dispersion model Plume dispersion is simulated separately with meteorological fields from control and FDDA runs using the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Draxler and Hess, 2004). The dispersion computation consists of a) particle transport by the mean wind b) a turbulent transport component and c) computation of pollutant concentration. The turbulent dispersion is computed by adding a random component to the mean advection velocity in each of the three-dimensional wind component directions. The horizontal mixing is computed from short-range isotropic similarity method and the vertical mixing

2849

from similarity theory where the inverse Obukov length (1/L) is estimated from surface fluxes (Kantha and Clayson, 2000). The meteorological fields needed in the model are u, v, w (horizontal, vertical wind components), temperature, height/pressure, surface pressure and the optional fields are moisture, vertical motion and turbulent kinetic energy. The meteorological profiles from the MM5 model at each horizontal grid point are linearly interpolated to the dispersion model terrain-following vertical grid. The dispersion simulation is done over a range of 150 km around the source. A horizontal grid of 1.5  1.5 with grid spacing of 0.01  0.01 (roughly 1 km  1 km) and with six vertical levels 10, 25, 50, 100, 500, 1000 m above ground level (AGL) is considered in the calculation with HYSPLIT dispersion model. Ground level concentrations are computed as averages for the model level from 0 to 10 m AGL. The source term in the present study is taken for a postulated accident condition at the coastal site for an upcoming Prototype Fast Breeder Reactor (PFBR) for which impact assessment in the mesoscale distance range is considered. A hypothetical source of I-131 with assumed release strength 5.167E þ 04 Bq/h at the ground level for PFBR location is used in the HYSPLIT model. The source strength is taken from core disruptive accident calculations for PFBR for ground level releases (Srinivas and Venkatesan, 2005). The dispersion is simulated for the case 21e23 May’08. A total of 5000 particles are released per one emission cycle over the release duration of 48 h starting at 12UTC 21 May’08. The pollutant concentrations are estimated as integrated mass of individual particles as they pass over the concentration grid which is a matrix of cells, each with a volume defined by its dimensions. 3. Results 3.1. Simulated vertical profiles The vertical distribution of different meteorological quantities from control and nudging experiments for Kalpakkam and Chennai locations is analysed and presented to examine the performance deviations in the two model cases. Comparisons are made with GPS Sonde profiles for Kalpakkam for morning condition and radiosonde profiles for Chennai station for both morning and evening conditions. The profile comparisons are made for Kalpakkam station for 00UTC 16 Apr’09 (Fig. 3) and for Chennai station for 12UTC 22 May (Fig. 4) respectively. From these plots it is noted that the vertical structure of the atmospheric variables is relatively better simulated in the FDDA experiment at least in the surface layer as can be seen from the model temperature distribution at different times. The profiles for Kalpakkam during the stable conditions at 00UTC 16 Apr’09 indicate a ground level inversion up to 150 m and a neutrally stable layer from 150 m to 600 m which are well captured by the FDDA run. The vertical variation in humidity, wind speed and wind direction are better depicted in the FDDA run than in the control run as seen from comparison with high resolution GPS sonde measurements. The potential temperature profiles from observations for the morning condition (at 06 LT/ 00UTC) for Chennai (not shown) indicate a stable atmosphere up to 100 m height and an unstable layer from 100 m to 500 m above and a stable atmosphere aloft. The temperature profile from the FDDA experiment is closer to the observations and gives the vertical variations. The temperature profile for the evening hours at Chennai indicate a shallow well mixed layer up to 550 m and a stable layer aloft in the observations. The temperature profile in the control run indicates a stable layer in the lower atmosphere while the FDDA run simulated a well mixed layer in the first 550 m and closer to observations. PBL height is estimated from the radiosonde profile of Chennai for 12UTC May’08 using Richardson number (Rib) with a threshold value of 0.3. The estimated PBL

2850

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 3. Vertical profiles of potential temperature (A), relative humidity (B), wind speed (C), wind direction (D) at Kalpakkam station from the control and FDDA experiment along with GPS Sonde observations for 00UTC 16 Apr’ 08.

height is 576 m while the model value is 462 m from control run and 622 m from FDDA run. The vertical profile of wind speed for the morning time at 06 LT/ 00UTC indicates strong winds (4e10 m s1) up to a height of 600 m and calm winds between 600 m to 1250 m. It is noted that the control and FDDA runs underpredict wind up to 1000 m and then over predict aloft. The vertical trends in wind could be simulated by the FDDA run. For Chennai location the control run compares vertically at all levels while the winds from the FDDA run compare well up to 500 m and are under predicted further upwards (not shown). In the evening time for Chennai the wind in the FDDA run is overpredicted upto 500 m and agrees thereafter with observations (Fig. 4). The winds from the control run for the evening hours are overpredicted at all levels. The wind direction for the April case at Kalpakkam for morning conditions is southwesterly up to 1200 m and easterly aloft. The FDDA run has given this trend very closely. For Chennai location for the May case, the radiosonde observations indicate westerly and northwesterly winds in the lower 1 km region of the atmosphere and southerly winds up to 4 km. Winds from the control run show a deviation of about 30e40 degrees in the wind direction at lower levels while the FDDA compares well with observations. Humidity profiles at 00UTC 16 Apr’09 for Kalpakkam indicate the humidity falls continuously with height and the FDDA run brings out the vertical tendency matching with observations. For Chennai location at 00UTC 22 May (06 LT) the profiles from the FDDA experiment are more closer to the observations than in the control experiment. For the afternoon condition at Chennai profiles from FDDA experiment follow the

vertical trend but large deviation is noted in the lower levels and the values are more closer to observations above 500 m. In the control run the vertical profile of RH for the evening hours deviates from observations. 3.2. Diurnal trends in the simulated meteorological fields The diurnal evolution of the surface meteorological variables from the control, FDDA experiments is examined for a few locations in the domain 4 (fine grid) and domain 3 and compared with observations from automated weather stations for the corresponding grids. The plots of surface air temperature for a few locations at Tirupathi, Puttur, Kalpakkam, Taramani, Pulicatnagar, SHAR Sriharikota, Kancheepuram, Gumminipundi, and Gadanki are shown in Figs. 5e8 for air temperature, relative humidity, wind speed and wind direction respectively. For the surface air temperature (at 2 m), it is noted that the values from FDDA case show closer agreement with observations than from the control run. The surface air temperature is under predicted in the control experiment for most stations and the predictions from FDDA are relatively closer to the observations. For the relative humidity at 2 m height the FDDA experiment gives closer values to the observations than the control run. For most stations the RH is slightly under predicted by the FDDA experiment while it is overpredicted by the control experiment (Fig. 6). The deviation in the simulated and observed relative humidity values is more in the case of the stations Gadanki and Kancheepuram. Note that the wind observations from AWS stations correspond to 4 m level, hence the plots show large

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

2851

Fig. 4. Vertical profiles of potential temperature (A), relative humidity (B), wind speed (C), wind direction (D) at Kalpakkam location from the control and FDDA experiment along with Radiosonde observations for 12UTC May 08.

deviations in the predicted values (Fig. 7). The control experiment gives large errors in wind speed for Tirupathi, Kalpakkam, Pulicotnagar, Gumminipundi, SHAR Sriharikota and the error is relatively more for Sriharikota which is located in the coarse domain 3. Overvall, the FDDA experiment gives better agreement for wind speed. For the wind direction at 10 m level the control experiment gives large differences in the wind directions at most stations and the FDDA experiment gives very close values in terms of the trends and the magnitude at most stations excepting SHAR Sriharikota, and Pulicotnagar (Fig. 8) located in domain 3. Thus overall, the time series of various surface level parameters is better simulated in the experiment with nudging. 3.3. Spatial variations of simulated meteorological fields Spatial wind field for the model initial time i.e., 12UTC 21 May 2009 from the control and FDDA experiments is shown in Fig. 9. The surface level wind field from the control experiment is westerly in the northwestern parts and southerly over most of the domain (Fig. 9A). Over the oceanic region the wind field is southerly. In the FDDA run the initial wind field is south-southwesterly over much of the domain including the oceanic region where the QuickSCAT wind data is utilised in the assimilation (Fig. 9B). Adjacent to the coast the wind blows to southerly in southern parts and southeasterly in the northern parts. Significant wind divergence is noted just above middle of the domain where Tirupati hills are located. Unlike in the case of control run the divergence is localized and spread to a wider extent in the FDDA case. At 925 hPa level the flow field in the control experiment is southwesterly over most of the domain excepting the northwestern parts where it is westerly

(Fig. 9C). In the FDDA experiment the flow field at 925 hPa level is westerly in the central and northern parts and is less strong than in the control experiment (Fig. 9D). The diverging wind pattern in the northwestern portion extends vertically to 925 hPa level to some extent in the FDDA run while it does not appear in the control run. Wind flow at 850 hPa level is significantly southwesterly in the control run in the entire domain while it is weak westerly in the FDDA experiment. The control run shows higher surface air temperature by 2e4  C over a wider land region than in the FDDA run (not shown). Thus a few differences are noted in model initial winds, temperature and humidity more clearly at the lowest level in the control and FDDA experiments. Simulated surface wind field at 06UTC 22 May 2008 from the 4th domain (Fig. 10A,B) indicates the flow is northerly in the central and southern parts and is southwesterly in the northern parts of the domain. The wind is calm in the central parts of the domain in the control experiment. In the FDDA experiment the surface wind field is more organized and the flow is northwesterly over the land portion of the domain. The wind is stronger in the land portions of the domain in the FDDA experiment while it is stronger over the oceanic region in the control experiment. The wind field at 925 hPa level (Fig. 10C,D) follows a westerly and northeasterly flow pattern in the control experiment while it is largely westerly in the experiment with FDDA. Thus, there are significant differences in direction and strength of the wind field in the two experiments, which reduced at higher levels indicating the impact of assimilation of surface observations in the model. Simulated surface air temperature at 06UTC 22 May in the control and FDDA runs (Fig. 11A,B) shows a 2e3 degrees higher temperature especially in the land portions of the eastern parts of the

2852

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 5. Diurnal trends of the simulated surface air temperature (at 2 m) from 12UTC 21 to 18UTC 23 May from control, FDDA experiments along with observations for (A) Tirupathi, (B) Puttur, (C) Kalpakkam, (D) Taramani, (E) Pulicatnagar and (F) Sriharikota.

domain in the FDDA experiment than in the control run. The surface relative humidity is higher over a large area in the control run than in FDDA run except at the coast where the FDDA shows higher humidity (not shown). The simulated PBL height corresponding to 06UTC 22 May’08 (Fig. 11C,D) indicates that it is relatively lower adjacent to the coast and gradually increases in the inland portions to the western parts in the control run. The PBL height simulated in the control experiment is about 600 m in the coastal parts while it is about 1200e1800 m in the inland regions. The PBL height in the experiment with FDDA is higher,

about 1600 m near the coast and increases to 2400e3000 m in the western portions of the domain. 3.4. Results from sensitivity experiments From the previous discussions of the model comparisons it is clear that the surface temperatures are not properly simulated. Model results show errors of 2e4  C in the surface temperature indicating cold bias. The results from the sensitivity runs for the case 24e26 Jan’09 are examined to identify appropriate choice for

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

2853

Fig. 6. Diurnal trends of the simulated Relative Humidity (%) (at 2 m) from 12UTC 21 to 18UTC 23 May from control, FDDA experiments along with observations for (A) Tirupathi, (B) Puttur, (C) Kalpakkam, (D) Taramani, (E) Pulicatnagar and (F) Sriharikota.

the simulation of surface variables and lower atmospheric winds. The diurnal evolution of surface temperature, humidity at 2 m and wind speed and wind direction at 10 m are shown in Fig. 12 for the location Kalpakkam from the fine grid domain. The observations from meteorological tower at Kalpakkam station are considered for model comparison as the tower observations are found more reliable than the AWS data. It is seen that the daytime temperature is generally under predicted in all the eight sensitivity simulations. The daytime temperature is under predicted by 6  C (GSM, BT), 5  C

(ACM), 4  C (BK), 3  C (MRFSOIL, MYJNOAH, MYJSOIL) and 2  C (MRFNOAH) in the experiments MYSOIL, MRFSOIL, MYNOAH and MRFNOAH respectively indicating cold bias in the model. The nighttime minimum temperature is overpredicted by 3  C (MRFNOAH), 2  C (BK, MYJSOIL, ACM, MRFSOIL) and 1  C (MYJNOAH, BT, GSM) indicating cold bias. The absolute errors in simulated daytime temperature are minimum with MYJNOAH followed by MRFSOIL and MYJSOIL. The errors are higher in all the other experiments. The errors in minimum temperature are found

2854

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 7. Diurnal trends of the simulated wind speed (m s1) (at 10 m) from 12UTC 21 to 18UTC 23 May from control, FDDA experiments along with observations for (A) Tirupathi, (B) Gumminipundi, (C) Kalpakkam, (D) Pulicatnagar, (E) Sriharikota and (F) Gadanki.

relatively lower in the cases MYJNOAH, MRFSOIL, GSM, BT and are relatively higher in the other experiments. As seen from the simulated diurnal temperature wave in different cases the experiments MYJNOAH, MRFSOIL are found to give better simulation for temperature. The model tends to overestimate the surface humidity in the daytime as well as in the nighttime in all the experiments except for MRFNOAH where the model has always under predicted the humidity. The absolute differences in relative humidity are about 5e8% in the cases BK, MYJNOAH, GS, ACM, BT, 10e15% in the cases MRFSOIL, MYJSOIL, BK and 20% in the case of

MRFNOAH. The nearest comparisons are given by MYJNOAH. As seen from the diurnal humidity trends the BK, MYJNOAH and MRFSOIL schemes produced nearest comparisons. The MRFNOAH has given worst prediction for humidity with largest errors. All the experiments have overestimated the wind speed at 10 m, except for MRFNOAH which has shown little underestimation during nighttime. The absolute differences in wind speed are about 0.5e1 m s1 for MRFSOIL, 2 m s1 for MYJSOIL, 3 m s1 for MYJNOAH, 4e5 m s1 for BK, BT, ACM and GS schemes respectively. MRFSOIL has produced the best simulation for wind speed both in diurnal trend

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

2855

Fig. 8. Diurnal trends of the simulated wind direction (at 10 m) from 12UTC 21 to 18UTC 23 May from control, FDDA experiments along with observations for (A) Kalpakkam, (B) Taramani, (C) Pulicatnagar, (D) Sriharikota, (E) Kancheepuram and (F) Gadanki.

and magnitude. For wind direction at 10 m level experiments MRFNOH, ACMPX, GSSOIL produced large errors followed by BT, BT, MRFSOIL, BKSOIL, and MYJNOAH experiments. The MYJSOIL, MRFSOIL and BKSOIL provide close comparisons to observed wind direction. Vertical profiles of simulated variables along with radiosonde observations are shown in Fig. 13 for Chennai station corresponding to the daytime at 12UTC 25 Jan’09 to examine the veridical tendency of different variables simulated in different cases. It is seen that all the experiments produce profiles consistent with convective well mixed conditions during the daytime.

However the mixed layer height is differently simulated in different experiments. The observations indicate a mixed layer development up to 900 m. Simulated potential temperature distribution indicate a mixed layer of 300 m, 500 m, 600 m, 700 m, 750 m in the cases GSSOIL, BT, ACMPX, MRFNOAH, MYJNOAH respectively. The MYJSOIL, MRFSOIL and BKSOIL all produced a mixed layer of height 800 m comparing closely to observations. The vertical distribution of temperature indicates all the experiments underpredict temperature by 2e3  C and the nearest comparison is provided by MRFNOAH and BKSOIL respectively. Similarly the vertical trends in

2856

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 9. Model initial surface wind field at 12UTC 21 May’08. Left panel corresponds to the control run and right panel corresponds to the experiment with FDDA. Top and bottom panels correspond to the surface and 925 hPa respectively.

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

2857

Fig. 10. Simulated wind field from the 4th domain at 06UTC 22 May’09. Left and right panels correspond to the control and FDDA runs respectively. Top and bottom panels correspond to the surface and 925 hPa respectively.

humidity are well produced by the BKSOIL, MYJSOIL and MRFSOIL respectively. The humidity errors are about 10e20% in different experiments, the highest errors are found between 0.5 and 1.0 km and for MYJSOIL, BKSOIL and MRFSOIL respectively. The variation of

wind speed with height is almost similarly simulated by all the experiments except BKSOIL which produced the nearest profile in the lower levels (below 600 m). There is an error of roughly 2e3 m s1 in the wind speed in all the experiments excepting

2858

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 11. Simulated surface air temperature ( C) and PBL height (m) at 06UTC 22 May’09. Left and right panels correspond to the control and FDDA runs respectively. Top and bottom panels correspond to the temperature and PBL height respectively.

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

BKSOIL. The experiments GSSOIL, MYJNOAH, ACMPX, MRFNOAH, MYJSOIL, MRFSOIL have given larger errors in the layer from surface to 1 km. The BKSOIL produced the nearest comparison of wind speed profile followed by the MRFNOAH and BT schemes. The vertical trends in the wind direction are well simulated by MRFSOIL, BKSOIL and MYJSOIL with minimum deviation (w10 ) in wind direction. The MRFNOAH produces large errors in wind direction upto 50 . Though the wind directional difference is very small in the surface layer in the case of GSSOIL, ACMPX, MYJNOAH, BT they resulted in large errors vertically upwards. Thus the comparisons of diurnal variations and vertical profiles indicate that the MRF PBL scheme performed better than the other PBL schemes for most variables. For the surface variables and their diurnal trends the simple soil thermal diffusion scheme performed better than the Noah LSM and PleimeXiu LSM especially for the wind speed and direction. Considering all the eight model experiments with different combinations of available PBL and surface schemes the overall performance of the model is found to be better in two experiments i.e., MRFSOIL and BKSOIL where the 5-layer soil thermal diffusion scheme is used for surface temperature prediction. Both Blackadar and MRF are non-local turbulence diffusion schemes and produced minimum errors for most variables at the surface as well as at different vertical levels. 3.5. Statistical trends of various simulated meteorological fields The model performance in the control and FDDA experiments for all the simulation periods is quantitatively evaluated by

2859

calculating several statistical metrics following the standard procedures outlined in several works (e.g., Willmott, 1982, Gilliam et al., 2006, Zhang et al., 2006). The parameters of temperature, humidity, wind speed, wind direction and geopotential heights at the surface and the upper-air levels at 1000, 850, 700 hPa are considered for computing the statistical indices. The statistical metrics in the present analysis include Pearson correlation coefficient (r), Spearman rank correlation coefficient (s), Mean Error (ME), Multiplicative Bias (BIAS), Mean absolute error (MAE) and Root Mean Square Error (RMSE). The observations for evaluating the statistical metrics are obtained from the national meteorological observations consisting surface and upper atmospheric data for a number of parameters collected by IMD and globally archived by the NCEP, USA. This data consists of about 25 upper-air observations and 100 surface observations over the Indian region. From the composite skill scores of various parameters over the 48 h forecast period for various simulation cases in 2008 and 2009 (Table 2), it is seen that the correlations r, s decrease from the surface to the upper atmospheric levels for the temperature, relative humidity, geopotential height and wind direction in both the control and FDDA experiments. For the wind speed the correlations are higher at 1000, 850 and 700 hPa levels in both the cases. The magnitude of the correlations for all the parameters is relatively higher in the FDDA experiment than in the control run. For the air temperature while the BIAS, RMSE are comparable in both the control and FDDA experiments, the ME, MAE are lower in the FDDA experiment than in the control experiment. For the geopotential height all the statistical metrics ME, MAE, BIAS and RMSE are slightly higher in

Fig. 12. Diurnal trends of the surface temperature (A,B), humidity (C,D) at 2 m, wind speed (E,F) and wind direction (G,H) at 10 m from 00UTC 24 to 00UTC 26 Jan’09 from sensitivity experiments along with observations. Left and right panels correspond to Gadanki and Kalpakkam respectively.

2860

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 13. Vertical profiles of potential temperature (A), relative humidity (B), wind speed (C), wind direction (D) at Chennai station from sensitivity experiments along with Radiosonde observations for 12UTC 25 Jan’ 09.

Table 2 Composite skill scores of MM5 predictions from the 4th domain for the Control and FDDA runs. Parameter

Level

Control run

FDDA run

r

s

ME

BIAS

MAE

RMSE

r

s

ME

BIAS

t

Surface 1000 850 700

0.683 0.790 0.744 0.530

0.601 0.591 0.705 0.664

1.844 1.409 2.320 1.167

0.994 0.995 0.992 0.996

2.235 2.114 2.551 1.463

2.658 2.334 2.912 1.797

0.762 0.893 0.857 0.708

0.691 0.900 0.910 0.667

1.675 1.605 1.586 0.938

0.994 0.995 0.995 0.997

MAE 2.060 1.656 2.116 1.219

2.445 1.945 2.360 1.561

h

1000 850 700

0.825 0.584 0.539

0.857 0.521 0.544

6.582 11.465 8.518

1.484 1.008 1.003

11.087 13.723 16.761

13.905 15.131 18.738

0.757 0.823 0.524

0.734 0.821 0.517

6.365 9.642 10.823

1.269 1.007 1.004

8.847 11.935 17.805

10.484 13.028 18.733

UV

Surface 1000 850 700

0.450 0.725 0.634 0.716

0.407 0.634 0.568 0.617

1.194 1.892 0.646 0.684

1.928 2.036 1.181 1.144

2.174 2.370 2.144 2.819

2.528 2.607 2.563 3.380

0.555 0.729 0.688 0.832

0.532 0.500 0.643 0.773

1.364 1.636 0.253 0.598

2.025 1.995 1.108 1.129

1.801 1.735 2.643 2.171

2.166 1.902 3.022 2.494

Dir

Surface 1000 850 700

0.457 0.736 0.599 0.742

0.485 0.582 0.538 0.648

0.802 0.102 0.136 0.829

1.545 8.348 1.368 0.214

1.854 2.382 2.230 3.404

2.540 2.685 2.751 4.025

0.550 0.821 0.673 0.763

0.542 0.813 0.575 0.770

0.720 0.422 0.676 0.236

1.450 2.185 2.565 1.948

1.458 2.082 2.552 2.638

2.433 2.283 2.972 3.164

RH

Surface 1000 850 700

0.563 0.770 0.715 0.335

0.563 0.701 0.633 0.453

3.251 2.364 3.959 4.510

1.059 0.958 1.139 1.108

9.403 10.114 16.186 21.953

11.775 11.104 19.507 25.659

0.655 0.840 0.725 0.420

0.619 0.717 0.772 0.550

4.444 0.819 2.579 2.586

1.072 0.990 1.125 1.200

8.689 9.788 14.271 14.654

10.641 10.946 16.115 17.122

t e temperature, h e geopotential height, uv e wind speed, Dir e wind direction, RH e relative humidity. r e Pearson Correlation coefficient, s e Spearman Correlation coefficient, ME e mean error, BIAS e multiplicative bias, RMSE e root mean square error.

RMSE

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

the control experiment. For the wind speed and relative humidity all the error metrics are lower in the FDDA experiment than in the control experiment. For the wind direction substantial improvement is found in FDDA experiment in the error metrics at all levels except at 850 hPa. Thus overall for most of the parameters the FDDA experiment gives better correlation with observations and has minimum error coefficients.

3.6. Results from dispersion simulations The impact of the meteorological model results with and without assimilation of observations on flow trajectories and plume concentration are examined from the HYSPLIT results. 3.6.1. Simulated trajectory Air flow trajectories are considered from two hypothetical source locations in the fine domain at 12.56 N, 79.0 E in the western side of the domain and at 12.56 N 80.16 E in the eastern side just at the coast. The starting time of the release of the trajectories is 12UTC, 21 May’08 and the trajectories are calculated for 36 h. The forward trajectories (Fig. 14) for the source located at the coast (line in red color) are almost identical in both the cases of meteorological fields. For the source located inland the forward trajectories in the control and FDDA cases are different. In the simulation with flow field obtained from the MM5 control run, the trajectory of the air parcels curves initially to the east and then to south. In the simulation with flow field obtained from MM5 FDDA experiment the trajectory though initially is directed to the east, makes a loop and then turns to the southeast. Clearly the recurving of the trajectory in the FDDA experiment is because of the diurnal variation of the flow field at the coast and development of sea-land breeze flow in the FDDA experiment. Trajectories in the western part of the domain reflect the differences in the wind field simulation in the nudging experiment. In the FDDA run the wind flow in the lower atmospheric region is seen to the east, northeast direction and then northwest and westward directions in the central and northern parts of the MM5 fine domain. Thus the flow field from the control run without data assimilation is likely to lead to significant errors in the trajectory of the pollutants released from

2861

the inland location while the FDDA run corrects this error in the trajectory to some extent. 3.6.2. Plume distribution pattern The concentration contours from HYSPLIT from the ground level (0e10 m layer) analyzed and presented. The orientation and distribution of the plume in the control and FDDA cases is significantly different at different time steps (Fig. 15). At 20 UTC 21 May’08 the plume from control run is oriented to the southeast direction while the plume from FDDA run is oriented to the east. The plume concentration is about an order higher in the FDDA run than with control run which may be because of relatively weaker winds simulated in FDDA run. At 06UTC 22 May 08 while the plume in the control run is still in the southeast direction, the plume simulated in the FDDA run makes incidence on to the land due to the setting up of sea breeze flow. In the subsequent time intervals the plume for FDDA is completely over the land while that calculated for control run stayed over the oceanic region. The simulated plume in the case of FDDA at 06UTC and 12UTC, 22 May indicates wide distribution or spread due to enhanced atmospheric turbulence resulting from the incidence of sea breeze flow and consequent change in the direction of wind flow. Thus the plume position is different in the two simulations for corresponding times and this variation may be attributed to the variation in the simulated wind field in the two cases as can be seen from the plume distribution at different times. The plots of the pollutant plumes from the two simulations at different times indicate a change in the concentration levels. For instance the plot for 21 UTC May 08 gives a higher concentration in the downwind distances up to 50 km for the simulation with MM5 FDDA fields than in the simulation with MM5 control fields. The plot for 06UTC 22 May 08 gives a higher concentration in the downwind distances up to 20 km for the simulation with MM5 control fields than in the simulation with MM5 FDDA fields. The differences in the concentration distribution may be attributed to the differences in the plume position and also to the variation in the respective meteorological fields from the two simulations. The changes in concentration filed from the two cases of simulation are illustrated in Fig. 16 through the time series of concentration for a set of six imaginary receptor points each 2 km apart from the source location in the northwest direction. In the

Fig. 14. Forward trajectories of hypothetical sources calculated with the simulated wind fields. Left and right panels correspond to the meteorological fields from control and FDDA experiment respectively. The blue line and red lines indicate the trajectories for the sources located inland and at the coast respectively.

2862

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

Fig. 15. Simulated plume dispersion pattern at different times. Left and right panels indicate the plume dispersion with meteorological fields from the control run and FDDA experiment respectively.

Fig. 16. Time series of the concentration (Bq m3) at six receptor points at intervals of 2 km from the source point in the northwesterly direction.

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

control simulation concentration is found only at the receptor near the source. In the simulation with FDDA fields, except for the receptor near the source (shown as red line) the concentration at all other receptor points fluctuates and the peak level is found between 10 UTC and 16 UTC 22 May. This is because of the plume orientation along the positions of the receptors in the direction of the wind flow at this time in FDDA case. Also as discussed earlier the flow between 10 UTC and 16 UTC 22 May is significantly due to sea breeze and hence most receptors in the plume direction show the peak concentration during this time period. The wind field in the MM5 control run is mostly westerly and alienates the plume to the east or southeast direction over significant period of time. 4. Summary and conclusions In the present study the effect of the assimilation of conventional and satellite based suface observations in a mesoscale model is studied for the simulation of the wind field, mixing height and atmospheric dispersion simulation. The PSU/NCAR mesoscale atmospheric model is used to simulate the wind field, mixing height, temperature and humidity fields in the lower atmosphere. A high resolution (1.33 km) nested domain is used to study the flow and dispersion pattern over the southeast coastal region near Chennai in India. Surface observations from a dense automated weather station network over the land region in southern India and QuickSCAT wind vector data over the oceanic region are used for assimilation in the model. Some improvements in the predictions are noticed in FDDA experiments for various lower atmospheric parameters including the wind field and mixing height with assimilation. The model vertical profiles from FDDA run are closer to observations than the control run indicating a better simulation of model atmosphere in FDDA. The qualitative and quantitative comparisons indicate the FDDA has performed relatively better than the control run indicating the impact of observations in assimilation. The model is found to be sensitive to the surface physics and PBL turbulence parameterization. Numerical experiments conducted with various combinations of available PBL and surface physics showed that two combinations MRF PBL with 5 layer soil model and Blacakdar PBL 5 layer soil model provided relatively better predictions for the temperature, humidity, wind speed and wind direction at the surface as well as in the higher levels. The 5-layer soil model seems to perform better than the more complex Noah LSM for the case under study. The dispersion simulation using a Lagrangian particle model for a hypothetical accidental release condition has given distinctly different plume trajectory, concentration/dose distribution in the cases of control and FDDA MM5 simulations according to the variation in the wind flow and other meteorological quantities in the two cases. The changes in the dispersion simulations are manifested in the simulated trajectory positions, plume direction and concentration distribution pattern. The improvement in the simulated meteorological fields arising from data assimilation is likely to improve the dispersion simulation. A comparison of the predicted plume concentration/dose rate from the two cases of simulations with observations is not available as a hypothetical case is considered in this study. Observations from tracer experiments along with meteorological measurements are planned to validate the dispersion model used in the present study. Further studies with the measured concentration data would be needed to determine the impact of data assimilation on simulated dispersion fields. Acknowledgements Authors thank the NOAA ARL for providing the HYSPLIT Dispersion model used in this study. The QuickSCAT data were obtained

2863

from NASA. The upper-air observations are obtained from University of Wyoming. Thanks are due to Dr. B. Manikiam and Dr. Kusuma Rao, ISRO for providing AWS observations as part of PRWONAM mesoscale program. The co-authors Mr.Yesubau and Mr.Nagaraju are thankful to ISRO for the grant of junior research fellowships. The authors wish to record grateful thanks to the anonymous reviewers for suggesting improvements in the manuscript. References Blackadar, A.K., 1976. Modeling the nocturnal boundary layer. Preprints. In: Proceedings of the Third Symposium On Atmospheric Turbulence, Diffusion, and Air Quality. American Meteorological Society, Rayeligh, NC, pp. 46e49. Burk, S.D., Thompson, W.T., 1989. A vertically nested regional numerical prediction model with second-order closure physics. Monthly Weather Review 117, 2305e2324. Chen, F., Dudhia, J., 2001. Coupling an advanced land-surface/hydrology model with the Penn State/NCAR MM5 modeling system. Part I: model description and implementation. Monthly Weather Review 129, 569e585. Dickerson, M.H., Knox, J.B., Orphan, R.C., 1979. ARAC Update-1979. Lawrence Livermore Laboratory Report. UCRL-52802, 53 pp. Draxler, R.R., Hess, G.D., 2004. Description of the HYSPLIT_4 Modeling System. NOAA Technical Memorandum ERL ARL-224. Dudhia, J., 1989. Numerical study of convection observed during winter monsoon experiment using a mesoscale two-dimensional model. Journal of Atmospheric Science 46, 3077e3107. Dudhia, J., 1996. A multi-layer soil temperature model for MM5. In: Preprints, Sixth PSU/NCAR Mesonet Model Users’ Workshop. PSU/NCAR, Boulder, CO, pp. 49e50. Grell, G.A., Kuo, Y.H., Pasch, R., 1991. Prognostic evaluation of assumptions used by cumulus parameterizations. Monthly Weather Review 121, 764e787. Grell, G.A., Dudhia, J., Stauffer, D.R., 1995. A Description of the Fifth-generation Penn State/NCAR Mesoscale Model (MM5). Technical Report NCAR/TN-398þSTR. NCAR, Boulder, CO, 122 pp. Hong, S.Y., Pan, H.L., 1996. Nonlocal boundary layer vertical diffusion in a mediumrange forecast model. Monthly Weather Review 124, 2322e2339. Hanna, S.R., 1994. Mesoscale meteorological model evaluation techniques with emphasis on needs of air quality models. In: Pielke, Roger A., Pearce, Robert P. (Eds.), Meteorological Monographs: Mesoscale Modelling of the Atmosphere, vol. 25, No. 47. American Meteorological Society, Boston, Massachusetts 02108. Gilliam, R.C., Hogrefe, C., Rao, S.T., 2006. New methods for evaluating meteorological models used in air quality applications. Atmospheric Environment 40, 5073e5086. Janjic, Z.I., 1996. The surface layer in the NCEP eta model. In: Eleventh Conference on Numerical Weather Prediction, Norfolk, VA, 19e23 August. American Meteorological Society, Boston, MA, pp. 354e355. Janjic, Z.I., 2002. Nonsingular Implementation of the MelloreYamada Level 2.5 Scheme in the NCEP Meso Model. NCEP Office Note, No. 437, 61 pp. Kantha, L.H., Clayson, C.A., 2000. Small Scale Processes in Geophysical Fluid Flows. In: International Geophysics Series, Vol. 67. Academic Press, San Diego, CA, 883 pp. Knox, J.B., Dickerson, M.H., Greenly, G., Gudiksen, P.H., Sullivan, T.J., 1981. The Atmospheric Release Advisory Capability (ARAC): Its Use During and After the Three Mile Island Accident. Lawrence Livermore Laboratory Rep. UCRL-85194. Kovalets, I., Andronopoulos, S., Bartzis, J.G., Gounaris, N., Kushchan, A., 2003. Introduction of data assimilation procedures in the meteorological preprocessor of atmospheric dispersion models used in emergency response systems. Atmospheric Environment 38, 457e467. Rao, K.G., 2008. PRWONAM for Mesoscale Monsoon Research in India and Predictions Over SHAR-KalpakkameBangalore Region. Technology Development for Atmospheric Research & Applications. 193e230. Lagzi, I., Karman, D., Turanyi, T., Tomlin, A.S., Haszpra, L., 2004. Simulation of the dispersion of nuclear contamination using an adaptive Eulerian grid model. Journal of Environmental Radioactivity 75, 59e82. Mellor, G.L., Yamada, T., 1982. Development of a turbulence closure model for geophysical fluid problems. Reviews in Geophysics and Space Physics 20, 851e875. Mlawer, E.J., Taubman, S.J., Brown, P.D., Iacono, M.J., Clough, S.A., 1997. Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. Journal of Geophysical Research 102 (D14), 16663e16682. Pleim, J.E., Xiu, A., 1995. Development and testing of a surface flux and planetary boundary layer model for application in mesoscale models. Journal of Applied Meteorology 34, 16e32. Satomura, T., Kimura, F., Sasaki, H., Yoshikawa, T., Murarji, Y., 1994. Numerical simulation of regional scale dispersion of radioactive pollutants from the accident at the Chernobyl nuclear power plant. Papers in Meteorology and Geophysics 45 (2), 51e63. Seaman, N.L., 2000. Meteorological modeling for air-quality assessments. Atmospheric Environment 34, 2231e2259. Shafran, P.C., Seaman, N.L., Gayno, G.A., 2000. Evaluation of numerical predictions of boundary layer structure during the Lake-Michigan Ozone study. Journal of Applied Meteorology 39, 412e426. Srinivas, C.V., Venkatesan, R., 2005. A simulation study of dispersion of air borne radionuclides from a nuclear power plant under a hypothetical

2864

C.V. Srinivas et al. / Atmospheric Environment 44 (2010) 2846e2864

accidental scenario at a tropical coastal site. Atmospheric Environment 39, 1497e1511. Srinivas, C.V., Venkatesan, R., Muralidharan, N.V., Someshwar Das, Hari Dass, Eswara Kumar, P., 2006. Operational mesoscale atmospheric dispersion prediction using a parallel computing cluster. Journal of Earth System Science 115 (3), 315e332. Stauffer, D.R., Seaman, N., 1990. Use of four-dimensional data assimilation in a limited-area mesoscale model. Part I: experiments with synoptic-scale data. Monthly Weather Review 118, 1250e1277. Stauffer, D.R., Seaman, N.L., Binkowski, F.S., 1991. Use of four-dimensional data assimilation in a limited area mesoscale model. Part II: effects of data assimilation within the planetary boundary layer. Monthly Weather Review 119, 734e754. Stauffer, D.R., Seaman, N., 1994. Multiscale four-dimensional data assimilation. Journal of Applied Meteorology 33, 416e434.

Zheng, D.Q., Leung, J.K.C., Lee, B.Y., Lam, H.Y., 2007. Data assimilation in the atmospheric dispersion model for nuclear accident assessments. Atmospheric Environment 41, 2438e2446. Zhang, Y., Liu, P., Pun, B., Seigneur, C., 2006. A comprehensive performance evaluation of MM5-CMAQ for the summer 1999 southern oxidants study episode e part I: evaluation protocols, databases, and meteorological predictions. Atmospheric Environment 40, 4825e4838. Willmott, C.J., 1982. Some comments on the evaluation of model performance. Bulletin of American Meteorological Society 63, 1309e1369. Xiu, A., Pleim, J.E., 2001. Development of a land surface model part I: application in a mesoscale meteorology model. Journal of Applied Meteorology 40, 192e209. Zhang, D., Anthes, R.A., 1982. A high-resolution model of the planetary boundary layerdsensitivity tests and comparisons with SESAME-79 data. Journal of Applied Meteorology 21, 1594e1609.