Tech TALK Odor Control Dispersion Modeling: Feeding the Data Monster B y A d a m R o s e , PE , G I SP, PMP, A l a n P l u m m e r A s s o c i a t e s , I n c . ( APA I ) Dallas, Texas Abstract Dispersion modeling has been used to support wastewater treatment and collection projects for a number of years. Generating an atmospheric dispersion model of an odor source requires a great deal of input data: numerous field samples, structure geometry and processed terrain data to name a few. This input data is typically processed (via a dispersion model) through a high volume of historical meteorological data that then produces a large variety of output scenarios. These scenarios aim to present the broad range of atmospheric conditions that may occur at any given time, given the broad range of input conditions that may also exist. Clearly, this is a data-intensive process. For some practical reasons, data collection, as well as model calibration, have historically been secondary to the development of the model and results. This article discusses ways to amass more data, using new and non-traditional methods, for the collection and calibration of dispersion models. Background The topic of atmospheric dispersion modeling is very broad. The following sections aim to break down the topic into a few core data requirements, describe some traditional means and methods of data collection, and then discuss new features that may be helpful in the ongoing field of dispersion modeling. The goal of this article is to begin the discussion of how technologies prevalent in other fields may become useful support tools for model development. Many of the technologies discussed in this article exist in some form in other sectors of society, even in our daily lives. 32 july/august 2012 Model Building Needs Atmospheric dispersion models are a good decision support tool for odor control planning, design and testing projects. A valuable output of these models is an estimate of the concentration of a given chemical at a given time, location, duration and initial/liquid or solid phase condition. The number of different combinations of input variables, which generate just a single output value, is extensive (Table 1). The data needs of the model correspond directly to the number and type of required inputs, and can also be quite extensive. In many instances the Input Variable Type Sampling Requirements Note Odor Source Source Concentration Gas-phase samples of odor generating locations in the plant or system. a Varies based on liquid phase concentration, gas-liquid boundary conditions, temperature, wastewater characteristics, etc. Defined by regulatory requirements or project experience. b These typically only measured if being used as part of calibration. Temperature Atmospheric Stability Relative Humidity Wind Speed/ Direction Cloud Cover Varies based on risk aversion. Time Step Time Duration Elevation Terrain Grid Size Reflectivity Roughness Defined by project requirements. c Stochastic models usually run for one year to test for statistical significance. Deterministic models run over a short time period (1-6 hours) to test specific conditions. Defined by project requirements. d Higher resolution produces more results but requires more processing time. Typically defined by land use May also be further refined by direction from source. Table 1 – Selected Data Needs for Dispersion Modeling Each unit process may be highly spatially and temporally variable. Typically sampling attempts to detect the worst case release condition (independent of atmospheric dispersion conditions). b Previously generated datasets are typically used due to quality control standardization. c A longer time step duration will lead to lower average dispersed concentrations due to meteorological variability. Note that many modeling applications do not directly measure time steps shorter than 60 minutes – they infer concentrations based on power law assumptions. The duration will correspond to the type of meteorological data. d Most models produce results based on Cartesian or polar grids. The trade off in small grid size (keeping total coverage area the same) is in computing time: even modern computers may require several hours to generate a single, complex model run of several hundred receptors. a Click HERE to return to table of contents