Desktop version

Home arrow Engineering

  • Increase font
  • Decrease font

<<   CONTENTS   >>

Urban and regional greenhouse gas emissions and observing networks

Cities and metropolitan areas aggregate populations and energy usage and are major contributors to a nation's total GHG emissions inventory. In 2018, it was estimated that approximately 55% of the world’s populations live in urban areas and megacities (cities with populations >10 million) with urban populations projected to reach 68% by 2050 (UN-DESA, 2018). Further concentration of human activities and concomitant emissions are anticipated. Measurement systems that better quantify urban emissions are likely to become critical as a means of quantifying the effectiveness of carbon management efforts.

Prior to к 2010, most atmospheric greenhouse gas measurement capabilities were focused on continental- and global-scale observations and analyses. A notable U.S. exception is the research effort in Salt Lake City (Strong et ah, 2011: Lin, 2018) that began in the mid-2000s. In recent years, urban greenhouse gas observing networks have been established in six U.S. cities: Salt Lake City, Utah; Boston, Massachusetts (Sargent, 2019); Indianapolis, Indiana (Elementa. 2018); Los Angeles, California (Yadav,

2019); Oakland and the Bay Ar ea ofCalifomia (Cohen, 2019); and the Baltimore. Marylaud-Washington, D.C. region of the Northeast Corridor (Martin, 2019). These observing networks, and the research efforts they support, differ from one to the next in the number of observing nodes, GHGs observed (although

CO, and methane are the most prevalent) and in the range of emission estimation methods brought to bear, although most if not all utilize versions of Bayesian inference methods. These efforts are aimed at developing and demonstrating emissions quantification capabilities using combinations of bottom-up and top-down techniques. The effort in the Oakland/San Francisco Bay Area (Kim, 2018) is focused on both air quality and greenhouse gas measurements, using a network of low to moderate cost/low to moderate performance sensors in a relatively high-density observing strategy, with ~ 50 observing nodes, at this writing. The Boston project is focused on whole-city emissions estimates. Three of these cities form NIST’s Urban Greenhouse Gas Measurements Test Bed System, Indianapolis (the Indianapolis Flux Experiment, INFLUX), the LA Megacity Carbon Project, and the Northeast Corridor. The Baltimore/ Washington (the NEC/BW project) is the beginning of this effort that is planned to reach to the Boston area given sufficient resources. All use surface-based observing networks of designs specific to the urban character of the locale. Aircraft-based whole-city observations and methods of analysis are also used. Most employ high-sensitivity GHG instrumentation and standards referred to mole fraction standards disseminated by WMO’s Central Calibration Laboratory (NOAA’s GMD) in order to ensure consistency in measurement data globally. Emissions modelling provides a means of transforming bottom-up emissions data and estimates at spatial and temporal scales consistent with the variety of atmospheric observation and analysis methods applied (Gurney, 2011). These transformations of inventory reporting information combined with atmospheric observation and analysis, described subsequently, are a means of moving toward an as yet unrealized reference framework supporting quantitative comparison of emissions measurement and estimation results for a range of independent emissions quantification methods.

As with global emissions dynamics simulation, urban GHG mole fraction observational data is combined with atmospheric transport simulation and statistical optimization methods in order to infer source/sink properties of greenhouse and other longer-lived trace gases. Aircraft observations are often used in a mass-balance analysis to provide whole-city emissions estimates. Figure 4 is a cartoon showing the major elements of an urban region that includes residential, commercial, and industrial buildings, power plants, transportation and communications systems, and geo-political boundaries. Situated within the region is a communications tower-based, two-segment GHG concentration observing network. One segment is located within the more densely urbanized section of the city and the other located along its boundaries. The wind moves GHGs into and through the urban and regional domain. Incoming atmospheric mole fractions are modified by emission and removal processes in and around the city.

Elements of the atmospheric inversion method of locating and estimating the flow of greenhouse gases to and from the atmosphere in an urban setting

Figure 4. Elements of the atmospheric inversion method of locating and estimating the flow of greenhouse gases to and from the atmosphere in an urban setting. The NWP model establishes the three-dunensional gridded domain depicted by the red

(surface/honzontal) and green (vertical) meshes.

Incoming and outgoing GHG concentrations are measured by the outer segment nodes of the observing network. Its inner segment is more sensitive to GHG fluxes within the city. NWP models simulate atmospheric motions within its spatial and temporal grid system, an essential element for inferring the location and magnitude of source emissions and uptake. Mole fraction values shown in Figure 4 are those typically found in moderately-sized U.S. cities. The atmosphere entering the domain is laden with CO, resulting from partial mixing of emissions upwind. Typically, incoming air concentrations vary due to differing upwind emissions, causing ever-changing incoming values dependent upon wind speed and direction and temporal change of upstream GHG source and sink fluxes. Emission enhancements are differences between observed incoming concentrations and those observed by network nodes. Estimates of incoming mole fraction loading may also use other sources, e.g., global or continental-scale models, for incoming mole fraction estimation, although direct measurement is likely to give more realistic estimates across the domain’s entiy plane. The outgoing flux from a city increases in mole fraction by a few to several tens of pmol/mol depending on atmospheric dynamics and varying emissions and uptake from the parts of the urban area with significant vegetation.

5.2.1 Atmospheric inversion modeling of urban and regional greenhouse gas fluxes

Bayesian inference methods are applied to estimation of GHG fluxes and their dynamics across global, continental, regional, or urban modeling domains. The observed quantity is the GHG concentration. Statistical optimization methods are based on Bayes’ concept that the probability associated with an event, or initial data set, can be updated by additional information. Stated more formally, Bayesian inference is a method of statistical inference based upon Bayes’ theorem,8 where it is assumed that the probability for a hypothesis can be modified or updated as more evidence or information becomes available. In the case of evidence-based GHG flux estimation, an initial estimate, termed a prior, is updated by additional observations in order to inform or refine the prior estimate with more information and construct a posterior flux estimate. In estimating urban source and sink fluxes, the hypothesis is based upon emissions inventory data for the region of interest. In some cases, an analysis may begin with a so- called “flat prior” that might be derived from a whole-city emission estimate which is then sub-divided equally among the surface grid cells of the NWP applied domain, as illustrated in Figure 4.

Figure 5 illustrates the major components of the Bayesian inference approach as applied to atmospheric observation and analysis:

Atmospheric inversion analysis frame work diagram illustrating the major components of flux estimation

Figure 5. Atmospheric inversion analysis frame work diagram illustrating the major components of flux estimation.

Although Bayes’ theorem originated in the early 19th century, its implementation is, in most cases, computationally intensive. Only with the advent of high capacity computing capabilities has it become widely used in science, technology, and some social science fields.

  • 1. The initial, or prior, flux estimate for the domain over the time of the analysis,
  • 2. An NWP and a transport/dispersion model for simulating atmospheric transport dynamics,
  • 3. Simulation of network node observations based on a flux prior coupled with atmospheric transport,
  • 4. The observed mole fraction data, and
  • 5. Statistical optimization.

NWP simulation of atmospheric dynamics, particularly those within the planetary boundary layer (PBL), where emissions first begin mixing with the atmosphere, coupled with Lagrangian backwardtime dispersion modelling provides a means to simulate mole fraction observations. Another approach is the development of a sensitivity function, or adjoint, by the NWP model itself using its atmospheric turbulence and dispersion models. The latter approach is rarely implemented due to the complexities of forming the adjoint.

With sensitivity or influence functions, Greenhouse gas mole fraction signals can then be simulated by combining prior emissions information with the sensitivity, or influence function, to obtain a modelled observed signal (orange, upper, data trace). These can then be compared with the GHG mole fraction observations (purple, lower, data trace). Optimization of the simulated data by adjusting prior emissions values relative to the measured data yields an updated mole fraction data set for each network node along with estimates of fluxes at each grid cell in the domain. In essence, mole fraction measurements conflated with atmospheric dynamics simulation and initial emissions estimates inform an evidence- based measurements and analysis outside of traditional ones.

The next few sections discuss the major features of each of these atmospheric analysis method components used widely for greenhouse gas flux estimations from the global to the local scale and based on a variety of atmospheric observation methods and initial flux prior estimates. The prior flux estimate—Emissions modelling

Emissions data sources contributing to inventory compilation cover broad ranges of temporal and spatial scales. The purpose of emissions models is to transform these data using emission-activity factor methods to scales consistent with the NWP scales driving atmospheric inversion analysis. Development of these transformations has been underway and in use since the mid-1980s (Marland, 1985; Anders, 1996; Gtuney, 2003; Gregg, 2009) and used as prior estimates or to compare with various other types of atmospheric analyses. Initial efforts were focused on anthropogenic emissions from fossil fuel combustion, particularly by power plants in the U.S. that are Continuous Emissions Monitoring System-equipped and coal-fired as these are some of the largest emissions sources.

Emissions modelling methods har e been extended to deal with other emission sectors relying on a range of data sources. Because CO, emissions are closely related to the fossil fuel combustion process itself in many of these processes, emissions flow directly to the atmosphere without a practical means of direct measurement. Rather, some kind of indirect analyses involving proxy parameters associated with the originating processes are needed for determining activity and emission factor values. Here, a general description of proxy parameter uses, and the underlying operational information needed in order to estimate vehicular emissions will be used as an example of emissions estimation from a complex emissions sector, transportation emissions. Similarly, complicated analyses are applicable to other sectors and their accompanying proxies and operational conditions.

Since vehicles change location continuously and are not equipped with direct emissions measurement, emissions models have been developed primarily for air quality emissions. These also pror ide CO, emissions values (EPA-NVFEL. 2019) that rely on proxies, such as vehicle type and operating parameters, to estimate activity/emission factor values coupled with spatiotemporal information. Emissions maps that include time varying behavior, such as that of Figure 6, can then be made (Gately, 2015). Vehicle miles travelled is a primary activity proxy. Emission factor proxies are somewhat more complex as they must accoimt for engine emission characteristics that vary with engine, vehicle type, and load. Vehicle fleet models and engine types and then emissions properties are characterized as a function of load and speed by both the U.S. EPA’s National Vehicle and Fuel Emissions Laboratory' (NVFEL. 2019) and auto manufacturers using extensive engine and vehicle type testing to obtain both activity and emission factor

Oil-road CO, emissions

Figure 6. Oil-road CO, emissions: Coterminous U.S. with selected urban areas at 1 bn resolution (Insets) maps showing details of metro areas siurounding Seattle (A), Los Angeles (B), Houston (C), Atlanta (D), and Boston (E). Image from

(Gately, 2015).

data. Engine load, and therefore emitted amounts of GHGs and air quality gases, is vehicle speed, weight, and road characteristic dependent. Road properties, primarily vertical incline, influence vehicle engine load, as does vehicle speed. To facilitate the function of these and similar models, road information is given in the form of segments of known length with additional descriptive parameters related to vehicle loading. These properties (location and road segment length) and load requirements to maintain a given speed are among the parameters needed in order to provide spatiotemporal vehicle emissions data. Road type and location, miles travelled on each type, average speed, vehicle type and similar information can be obtained from U.S. statistical data collected and disseminated by local, state, or national transportation agencies. Similar rationale and methods are applied to train, marine, and plane emissions. Simulating atmospheric transport dynamics

Inversion analyses require the use of dispersion models coupled with NWP models, e.g., the Weather Research and Forecasting model (Skamarock, 2008) or those of the European Centre for Medium- Range Weather Forecasts (ECMWF, 2019), to simulate the transport and dispersion of trace gases in the atmosphere. Typically, NWP simulations output atmospheric dynamics and related information with spatiotemporal resolutions of 1 km2 and larger and at sub-hourly scales in urban domains. In addition to using fundamental physical principals, NWP models rely on meteorological observations taken both at individual observing stations on Earth’s surface and from both surface and satellite-based radar observations, to update simulations and forecasts of atmospheric dynamics (Kalnay, 2003). For North America, an alternative to NWP simulations for a given domain are meteorological data products from the U.S. National Weather Sendee, such as the North American Mesoscale Forecast System9 (NAM). NAM data is available at a range of spatiotemporal scales, the smallest currently being 12 km, beginning in 2003 and 2004. Similar meteorological data products are available from Europe and other parts of the world. Influence functions, footprints, and simulated emissions—Linking source/sink location with observations

A means is needed to relate contributions to obseived mole fraction signals from source or sink fluxes upwind of the observation location, whether the observation is from a surface-based network, a satellite 51

track giving path-integrated mole fraction observations, or an aircraft flight path. Identifying these source/ sink locations and estimating then flux magnitude at an observing node location is a two-step process utilizing wind fields throughout the computational domain. Approaches based on Eulerian or Lagrangian dynamical methods are used to determine influence functions, or footprints, linking network observing nodes to flux locations.

Lagrangian methods take the view of a reference frame attached to marker particles that move through the atmosphere under the influence of wind fields prov ided by Eulerian NWP simulation data. These are principally three-dimensional wind speed and direction, turbulence intensity, and thermodynamic atmospheric state parameters defining the varying meteorological conditions encountered during of a marker's trajectory. In simulations mu backward in time, marker particles are released from a receptor, an observing location, to determine the particles’ trajectories through the domain during the time of the simulation. For example, an air mass with straight-line winds of 5 m/s (~ 11 miles/hr) transits a domain of 200 km (я 125 miles) extent in approximately 11 hours. It is not unusual for modelling computations to span several times this transit time as meteorological conditions, such as shifting wind direction and speed, may cause marker particles to remain within the computational domain significantly longer. Typically, several hundred to thousands of particles may be released, generally uniformly, over the period of one to several hours. The resulting collection of trajectories allows determination of marker residence time in grid cells of the atmospheric column above each surface grid cell in the domain (Lin, 2012; Lin. 2003). The assumption that each atmospheric column exists in a vertically well-mixed state below a specified altitude, i.e., the GHG mole fraction below that altitude is constant, allows association of residence time with surface or selected grid cell emission and uptake locations. Common practice is to use PBL height or one half that value as the height below which the well-mixed assumption holds. Residence times are normalized to the total time and number of markers to develop the fr action of GHG mole fraction enhancement that source/sink grid cells contribute to, or influence, mole fraction signals. Influence functions have units of mole fr action per unit flux, e.g.. pmol/mol/(mol/(m--s)). Convolving influence functions with prior flux estimates provide simulated mole fraction signals at an observing location, as illustrated by the orange data trace shown in Figure 5.

Eulerian or forward-time methods, e.g., WRF-CHEM (Grell, 2005; Fast, 2006) of describing particle motions use a spatially fixed reference frame to move particles within the computational domain. Influence functions are determined in the same general way, i.e., through particle release and trajectory tracking. However, markers travel forward in time and must be released from all potential flux origination locations, tracing out trajectories that may or may not intercept receptors. These require multiple simulations to link flux origination locations with those of receptors, have much higher computational costs, and are, therefore, rarely used for influence function determination.

<<   CONTENTS   >>

Related topics