 
|
Project On-line Deliverables: D02.3User Manuals (Final)(basic models)
 EXECUTIVE SUMMARYThis document describes the basic models, their application range and their restrictions. It gives an overview of the selected and developed models, their data requirements, the output choices and their limitations. The model system design is based on the user requirements and their consequences. In particular, a number of basic constraints must be fulfilled. The core part of the model system consists of three main parts:
All submodules use their own set of parameters. The developed interfaces between the modules guarantee a quick and error-free communication between different kinds of models. Monte Carlo techniques are used to determine the impact of the data uncertainties to the final model results.
Table of Contents1 Introduction2 The Release Models3 The Wind Field Generator4 The Dispersion Model5 ReferencesA Appendix1 IntroductionFor emergency planning and especially for emergency management purposes, the system has to provide a result or a recommendation in a fraction of real time. This requires fast data management as well as a fast execution of the underlying models. To date, simple screening models have been used for these purposes. These models are very fast but very limited in their application because they normally assume a flat terrain and do not allow complex wind fields as an input. Also, although the quality of the input parameters is often very questionable, they are considered as known. The HITERM approach uses advanced models in conjunction with uncertainty analysis methods. In general, the execution of this system in an appropriate resolution and time requires the use of high performance hardware platforms. However, smaller companies as well as authorities or fire brigades often do not have access to such systems. To make the system more flexible, the code of the individual software packages was parallelized using different parallelizing strategies according to its inherent programming structure. The simulation software was designed to run optimally on a cluster of workstations using PVM. Alternatively, other hardware platforms or single workstations are supported which do not guarantee sufficiently small execution time for emergency management calculations but which are nevertheless appropriate for training and analyzing tasks. The HITERM model system is able to compute the dispersion of toxins from stationary or mobile sources. It considers different potential release types to estimate the release rate of the toxin. Source strength models for emergency management tasks suffer from a high degree of uncertainty regarding the input parameter set. Depending on the sensitivity of the source function, this often causes unrealistic concentration patterns. To allow a more realistic determination of the release, a Monte Carlo method is used to construct a probability function of the source strength for different times of the release duration. With this method, all uncertain parameters are varied in a selected range. The time dependent source term probability function gives the data serving as an input for the Lagrangian dispersion model. The user must define the uncertainty band for all these parameters. The parallel implementation of the source term models uses a generalized mask to run different types of source term models. A complex 3D wind field generator takes into account atmospheric stability, slope flows and kinematic effects in complex terrain to construct the flow in rather different model domains. A particle dispersion model determines the concentration pattern of the hazardous substance, especially near the surface. 2 The Release Models
The model system for the computation of the source strength
consists of different individual submodules with regard to the release
type. The HITERM system explicitly includes:
If the source strength is known, the release models are no longer necessary or perform very simply However, in case of an accident very often only some geometric quantities (e.g. of a spill) are known. Most of the parameters needed for the source strength estimation are very uncertain. To take these uncertainties into account, for the evaporation pool release type a Monte Carlo simulation is used to determine a source strength distribution over time instead of using a single number for the source strength. Key parameters of these time dependent distributions are used as an input for the dispersion model. Such key parameters can be, for example, the emission rate at the mean and at the 95th percentile of the probability functions. Although explosion or burning may also be important release types, their explicit handling in the system is impossible for practical reasons (no data, expensive computational burden). With a high level of probability, an immediate total release of the substance can be assumed for explosions. For release types due to burning the determination of the source strength and the properties of the released substances may be very uncertain. However, this is a basic problem and not merely a problem of the modelling tools. For fire type release, the applied methodology is similar to buoyant jet release, taking into account the especially broad range of uncertainty for the model input parameter. For accidental release models, it is advisable to perform a worst case analysis to ensure that most of the cases are covered. Under accidental conditions where no or insufficient information is available to run a more precise source strength estimation, the system resorts to total release assumptions. The input for the release models is formed by a set of substance specific parameters and a set of release specific parameters such as release type, release and meteorological conditions. An example of the substance specific data set for the evaporation pool release type is formed by: MOLECULAR WEIGHT [G/MOL] 53.06 HEAT OF VAPORIZATION [J/MOL] 35500. BOILING POINT [C] 78. GAS DENSITY OF SOURCE [KG/M**3] 1.9 LIQUID DENSITY OF SOURCE [G/M**3] 8.06e+5 SATURATION PRESSURE CONST. A 1978. SATURATION PRESSURE CONST. B -27. SATURATION PRESSURE CONST. C -27. DIFFUSIVITY [M**2/S] 0.077e-4 THERMAL COND. OF LIQUID [J/MSK] 0.159 whereas
The kind of input data depend on the release type. For jet and buoyant gas release, only a few parameters for the gas phase are used. An effective source location is computed in the case of jet or buoyant gas release (Briggs 1984). This module is implemented as a part of the Lagrangian Dispersion Model for this release type. An example for this data set is given by: RELEASE VELOCITY [M/S] 20.4 STACK DIAMETER [M] 2.5 GAS TEMPERATURE [K] 500. FLOW RATE [M3/S] 600. whereas
Release and meteorological parameters can resemble the
following:
RELEASE TYPE 1 GAMMA [K/M] -0.016 STABILITY CLASS 3 MIXING HEIGHT [M] 800. WIND SPEED [M/S] 5. WIND DIRECTION [DEG] 30. MEASUREMENT HEIGHT [M/S] 10. ROUGHNESS LENGTH [M] 0.5 SURFACE TEMPERATURE [K] 279. SOURCE HEIGHT [M] 20. CLOUD [%] 20. LONGITUDE [DEG] 15.3 LATITUDE [DEG] 52.1 DATE [DD.MM.YYYY] 20.10.1997 DEPTH [M] 2.5 FLOW RATE [M3/S] 300. RELEASE DURATION [S] 1500. RELEASE VELOCITY [M/S] 0. SOURCE TEMPERATURE [K] 350. whereas
Especially the input parameters for the evaporation poole release are often very inaccurate in the case of an accident. A Monte Carlo frame has been implemented to take these uncertainties into account (HITERM deliverable D05). The user can select upper and lower limits for the uncertain quantities: WIND_SPEED 2.5 MIN_WIND_SPEED 2. MAX_WIND_SPEED 3. SURFACE_TEMPERATURE 303. MIN_SURFACE_TEMPERATURE 300. MAX_SURFACE_TEMPERATURE 305. SPILL_DURATION 3600. MIN_SPILL_DURATION 3200. MAX_SPILL_DURATION 4000. DEPTH 0.01 MIN_DEPTH 0.005 MAX_DEPTH 0.015 TOTAL_VOL 52.68 MIN_TOTAL_VOL 48. MAX_TOTAL_VOL 56. The output of the release module in a Monte Carlo frame is a time dependent source strength probability function, and the 95th percentile of this function is sufficient for a worst case emission scenario of evaporating pool release. For further details, please refer to D05. 3 The Wind Field GeneratorOne of the most important features of an atmospheric emergency simulation system is the computation of a realistic wind field in a complex terrain. The core part of the chosen wind field model is based on the Diagnostic Wind Model (DWM, Douglas 1990) which generates gridded wind fields at a specified time. The DWM adjusts the domain-scale mean wind for terrain effects (kinematic effects, such as lifting and acceleration of the airflow over terrain obstacles, as well as thermodynamically generated slope flows). It performs a divergence minimization to ensure mass conservation. The original code of the DWM has been mainly modified with respect to
In its current installation, the meteorological and regional input data are clearly separated from the technical model parameters which should not be touched by nonexperienced users. These parameters have been carefully selected by test runs or by using recommended default values (Douglas 1990). The default input data set for the technical model values is given in Table 2. The parameter set can be changed by editing the file diapar.in which can be found in the DWM home directory. The original code of the DWM includes two different approaches for estimation of the first-guess wind field: objective analysis of measured quantities or use of a very simple vertical profile. Objective data analysis performs well in the case of a dense net of observations. This is very unlikely in the case of the HITERM application range. An objective analysis with very few stations leads to a rather poor initial field and has therefore been drooped. All parameters and submodules previously needed to run this part of the model have been removed from the code. The other possible initialization of the vertical wind profile in the original DWM was rather crude, the user selecting either a constant wind profile or a profile given by a power law. In both cases, the initial wind profile has nothing to do with the concrete meteorological situation. The new incorporated submodule determines the vertical wind profile as a function of the stability class or /and of the temperature lapse rate and mixing height as well as dependent of the aerodynamic properties of the underlying surface. It follows the parameterization of the vertical wind profile suggested by Ermak (Ermak 1991).
The parallelization methodology can be found in deliverable
D02.1 Installation (Parallel environment). The overall execution time for
the DWM for a 100x100x15 domain in the original version was 210.4 s on
a SUN U2/2170Cr2, the optimized and parallelized version executed on a
cluster of 6 heterogeneous SUNs with considerable load took only 12.8s
for an identical set of input data and parameter settings. The following
steps are performed:
The diagnostic wind model is based on a digital elevation map of the model domain. A grid based mean elevation for every grid for the desired resolution must be provided by the customer. A typical application range of the model is from medium scale to regional scale (1 km x 1 km up to 200 km x 200 km), covering vertical levels up to the mixing height. The horizontal resolution depends on the given data (recommended 1D horizontal grid size is 1/100 of the domain length, e.g. for a 10 km x 10 km model domain an appropriate grid size is 100 m x 100 m). The vertical resolution is computed based on the stability (mixing height) and usually rises from a few meters at the bottom to some hundred meters at the top of the model domain. The number of vertical layers used also depends on the mixing height, e.g. for stable conditions and low mixing heights, the model uses 14 layers whereas it takes 20 layers for atmospherically unstable conditions. In contrast to the original DWM model, the vertical stratification of the levels is no longer user-definable but internally computed in order to avoid wrong or inadequate selection of the vertical model structure by a user in the case of an accident. The model is not able to resolve very local wind structures (e.g. lee waves at a building) but it provides a mean wind field for the area of interest. This is an example of the input data set for the wind
field model:
BERLIN DATENSATZ NX 49 NY 53 DX [M] 100. DY [M] 100. GAMMA [K/M] -0.016 STABILITY CLASS 3 MIXING HEIGHT [M] 800. WIND SPEED [M/S] 5. WIND DIRECTION [DEG] 30. MEASUREMENT HEIGHT [M] 10. ROUGHNESS LENGTH [M] 0.5 SURFACE TEMPERATURE [K] 297. whereby NX is the number of grid elements in East direction, NY the number of grids in the North direction, DX, DY the grid space [m] in x,y-direction (horizontal resolution), GAMMA is the vertical lapse rate [K/m] below the mixing height, STABILITY CLASS refers to one of six stability classes, MIXING HEIGHT defines the top of mixed layer, WIND SPEED [m/s] and WIND DIRECTION [DEG] are wind observations at MEASUREMENT HEIGHT [m] above the surface. Data availability regarding the lapse rate at the moment of a hypothetical accident seems to be very questionable. For this reason, an experienced user has the option of selecting an appropriate stability class, or he can enter a lapse rate. The procedure for selecting a mixing height is similar. If it is known from measurements, it can be specified. Otherwise a mixing height compatible with the chosen stability class is assigned. Table 3 lists default lapse rates and mixing heights for the six stability classes. If the user selected value for STABILITY CLASS is less than or equal to zero, a typical stability class will be determined using the lapse rate GAMMA; otherwise a typical lapse rate GAMMA for the given stability class will be chosen. If an user selects a mixing height equal to or less than 100 m, a typical mixing height value for the selected stability class will be assigned; otherwise the given mixing height will be use in the calculation.
In addition to the parameter set which has to be specified for every new run, a second set of parameters defines internal model parameters such as accuracy, the parametrization type or strength. This option is only valid for an experienced user of the system and for training purposes. The model output is a terrain and atmospheric stability-adjusted
3D wind field with its appropriate stability parameters. If there are no
known stability parameters, some advice can be given following Pasquill's
suggestions (Pasquill 1961).
4 The Dispersion ModelThe basic concept of Lagrangian models is the observation of individual particles which is why these models are also referred to as Lagrangian particle simulation models. In the case of atmospheric dispersion, the term particle denotes any air pollutant or any buoyant substance in the air. For physical reasons the particles are assumed to have no mass and spatial extension so that they can follow every flow. However, each particle is associated with certain characteristics. Together with the motion of a particle, the modifications of these characteristics will be registered. Gases or evaporating liquids are also represented by particles. Lagrangian models use given wind fields and take into account turbulence effects to predict the pathways of individual particles or air volumes for each time step. For this reason these models are also referred to as trajectory models. The definite form of a Lagrangian model is mainly determined by the chosen scale, affecting for example, the type of turbulence simulated. Particles or air volumes, respectively, may be released from any number of locations. The type of source, e.g. point or line source, has no influence. In contrast to Gaussian models, Lagrangian trajectory models are appropriate for the description of dispersion in complex meteorological situations and/or structured orography. Data input is provided by the wind field generator, the release model and from data sets. All data are based on a Cartesian coordinate system. The input from the wind field model comprises meteorological data and general information about the simulation area (file dwm.out):
The last two parameters are needed to calculate the effective stack height in the case of a jet release (Turner 1994). The parameters HM, ALA, USTAR, and Z0 contribute to determine the standard deviation of turbulent velocity fluctuations and Lagrangian time-scales (see below; Hanna et al. 1982). The orography of the observed terrain is given by a separated data file oro.in. Four types of gas/vapour releases are distinguished:
If an evaporating pool release is observed (release type 1) the value of LNUMCONC gives the information of which sort the values in in the file mit95.out are. The file contains the Monte Carlo values either of the mean or of the 95th percentile probability function or both values. Specific constant model parameters are set in lacon.dat:
The control parameters in latim.dat and lacon.dat are set due to experimentally experienced values, they may be altered by the user. The data input and initialization is followed by the main body of the
model, which is formed by a loop. The loop is controlled by the time step
(DT) for the period of simulation time (NDT) and the upper bound of emitted
particles.
The concentration is calculated for each grid cell by the mass of the particles per volume of the present grid cell. For the visualization an interpolation on the grid cells can be performed by the graphical system but is not provided by the model itself. The structure of the Lagrangian dispersion model makes the model well suited for numerical parallelization. Since for the time periods inbetween the output computation the individual particles can be considered as independent. A parallel implementation offers the possibility to use a large number of particles (the upper bound can be manipulated) and guarantees a reasonable execution time. For further information see D02.2 chapter 7. The following description of the determination of a particle trajectory is the form used in the present model. It represents one of the possible realizations which depend on the underlying scale, available data and intention. The computation of a trajectory is performed in the coordinate system attached to the particle and then the new position of the particle is transfered to the terrain-fixed coordinate system. Both coordinate systems are of Cartesian type. The position of a particle is described by a three dimensional vector
The advective wind is completely determined by the velocity and direction of the wind, while the fluctuation or turbulent component describes the actual fluctuation. The simulation of turbulence is based on the statistical theory of Taylor
for diffusion effects (Taylor 1921), and enhanced by an extension developed
by Obukhov (Obukhov 1959) and Smith (Smith 1968). The fluctuation is simulated
by a Markov sequence of first order: the velocity fluctuation To take into consideration (weakly) non homogeneous turbulences Legg
and Raupach (Legg et al. 1982) introduced an additional term Each model output is a terrain-following concentration field up to the
desired output level.
5 References
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
at a certain time t and changes each time step 
due to the advective wind and turbulence. The location of a particle is
computed by:
denotes the instantaneous velocity at time t and position 
and the fluctuation velocity 
at the position 
indicating
that the particle velocity does not undergo arbitrary changes. Thus, it
is determined by the autocorrelation function
describing the correlation between the present 
.The
random component describes the coincidental effects in diffusion. 
denotes the standard deviation of the velocity fluctuation
for the description of the vertical component of the velocity fluctuation 
which corresponds to the vertical pressure gradient. It implies that particles
are capable of quitting regions with less turbulence. Thus non physical
effects are avoided, such as accumulation of particles which may occure
under the restrictive assumption of homogeneity in an originally well-mixed
profile. The subscript w in the given formula denotes the vertical component
of 
,
respectively.
© Copyright 1995-2002 by: ESS Environmental Software and Services GmbH AUSTRIA