Transcript PM Monitoring Network Design Ideas

PM Monitoring Network Design Tool
and Resources
Planning Document from 1996
dfdshg
Background
•
•
•
•
The promulgation of the proposed PM2.5 air quality standard will require the
creation of a suitable PM2.5 monitoring network. Proper network design
requires substantial resources including experience, data and tools. In many
state and local agencies such resources are not available but there is
considerable knowledge about the local conditions.
The current thinking is that the general guidance and tools for the new network
design is to be prepared by the federal EPA, while the detailed design,
implementation and evaluation of the network is to be conducted primarily by
state or local teams.(Frank, 1996).
The design and evaluation of monitoring networks has many subtle features.
Some aspects of network design can use the methods of science. Other aspects
are a craft, requiring experience, craftsmanship and tools. Yet other parts of
network design are literally art in that they are dependent on the intuition of
the designer.
The goal of this project is to aid the science and the craft of network design
including design rationale, methods, tools and to provide the necessary
information resources.
Network Assessment/Design
• Air quality monitoring network design is about
layout and operation such that it best characterizes
AQ for the specific network purpose.
• At the minimum, the best design needs to:
– 1. Reduce the uncertainty of concentration estimate
for the unmonitored data domain
– 2. Incorporate and weigh the conditions imposed by
the network purpose.
Optimal Network Design
• An inherent problem of optimal network design lies in the following
paradox:
– In order to design a network to ‘best’ characterize the pollutant, it is
necessary to know the pollutant patterns.
– If the pollutant patterns are indeed known, what is the purpose of
designing a network
• In reality, air quality monitoring network design is somewhat less
paradoxical.
– Usually there are some monitoring data available as a guidance for full
network design.
– The goal may be simply to rearrange the network to better serve the needs.
– The purpose of the design may also be to reduce the network, so the actual
AQ pattern may be known reasonably well.
Network Design Theory(1)
•
•
•
The theory of network design is largely an optimization problem. The goal of the
optimization is to minimize the uncertainty of the (purpose-weighed) exposure
estimates at the unmonitored data domain. Cost may also be a factor.
Below is a simple model for the evaluation of network performance toward a specific
purpose.
The basic assumption is that the contribution of a monitoring site can be measured by
the combination (product) of two independent functions:
– one is related to the pollutant exposure and
– a relevancy-function determined by the purpose(s) of the network
G(x,y,t) = E(c(x,y,t))* R(x,y,t)
G(x,y,t)- relevance -weighed exposure function applicable throughout the data domain
E(c(x,y,t)) - pollutant exposure function which depends on c(x,y,t)
c(x,y,t) - ambient concentration as a function of space and time
R(x,y,t)- relevancy function which is to be defined by the network purpose (is there a better name for this?)
The reason for modifying the pollutant exposure function E with the relevancy
function R is to assure that irrelevant data receive less weight, while highly
relevant data receive proper consideration. For instance, for human exposure
monitoring, sites that have low population density are less relevant.
Uncertainty of Network Data Value
• Throughout the data space, the relevance-weighed exposure function, G, has an
uncertainty since both the exposure and the relevancy factor have their
respective uncertainties, UE and UR at the unmonitored location and times. (Or
for that matter even at the monitored domain.). One can then define UG as the
uncertainty of G.
• UG(x,y,t) = f [UE (c(x,y,t))* UR (x,y,t)]
• The function, f, could be the standard error-propagation function, i.e. the square
root of the sum of the error squares.
• In the past, this apparent paradox was resolved by network designers by basing
the network layout primarily on the relevancy function, R. In the case of
SLAMS for example, the monitoring sites were placed roughly in proportion to
the population density. [Is this how it went?].
Network Performance Measures
•
•
•
•
•
•
•
At this time there is no generally accepted definition of network performance,
however there are several statistical criteria that can be used to define
performance based on existing monitoring data. These include:
1.
Direct cross-validation with data (selectively removing sites)
2.
Maximum information (Caselton and Zidek, 1984)
3.
Minimum relative uncertainty (Venkatram, 1988)
4.
Minimum variogram estimate (Warrick and Myers, 1987)
A common characteristic of these algorithms is that each provides a measure
of uncertainty for concentration estimates based on monitoring data, hence
they are suitable for the evaluation of network performance.
A detailed exploration of these methods is currently underway at CAPITA by
Stefan Falke as part of his PhD dissertation. Inputs from the entire project
team would be helpful for this task.[Done in 1999]
Computer Aided Network Design Tools
• Ideally, network design should be guided by objective methods using
the set goals, available data, and clear algorithmic methods. It is,
however, clear that the network design has to incorporate many factors
that do not lend themselves to analytical modeling. Such factors may
include site availability and accessibility, costs, existing long-term
monitoring data, etc. The incorporation of all these factors in the
design process is best accomplished by designers who are familiar with
the local circumstances. For this reason it is desirable to develop
design methodology and a set of tools and data resources and to
support the local designers.
• It is assumed that the following aspects of network design are given to
a designer: the goal of the network; The geographic area of the
network coverage; Monitoring data for air quality and related
variables; Geography and other contextual information.
Network Design Tool Specification
 
The design tool should allow the display of existing air quality data on
spatial maps and time charts.
 
The spatial display on maps should allow the superposition of multiple
data layers to provide spatial context. These should include major emission
point sources, population density, topography, major roads, rivers etc.
 
The tool should allow the placement of network locations on the map
including an updated display of the estimated change of information value
contributed by each network configuration.
• The resulting tool should allow the designers to lay out and explore potential
network configurations while at the same time showing the information value
measure of the network. Incorporation of multiple potential value criteria may
be desirable.
Illustration of the Network Design Tool
•
•
An illustration of the tool’s appearance is given in Figure 1. It consists of a
Map view, Time view, and Information view. The Map view, consisting of
multiple data layers, would allow the addition, removal, and dragging of
stations from the new monitoring network. As stations are added, the
information view would automatically display the current information value
metric as a feedback to the designer.
The specification and evaluation of the tools should involve the project team.
The implementation of these tools is to be the responsibility of CAPITA.
Data and Knowledge Resources
•
•
•
•
Existing AQ data. The main data set in support of new network design consists of the
existing monitoring data for the region. This data set is used to establish the pattern of
exposure as well as to estimate the estimation of information value added by each
station. The air quality data should be augmented by other air quality related parameters.
These may include emission fields for point and area sources, surrogate variables (e.G.
PM10 as a surrogate for PM25), possibly wind rose data, etc.
Relevancy function data. It is presumed that the major relevancy function will be related
to population density. Zip code level population data need to be prepared for the
information value analysis.
Geographic context. Further context for the network design may be added by
geographic data on topography, major roads, rivers, and lakes. Political boundaries at
zip code, county, state, and national level would also help the design process.
Knowledge resources. The knowledge resources in support of local designers would
include tutorials on design rational, design procedure guidelines, and network evaluation
procedures.[John, how do you like your main task to be tucked away in this corner
here!]
Network Assessment through a Community
• The PM25 network design process across the country can be
aided by interaction among the designers and data analysts. A
community of network designers, with common interests,
could share a stable forum to show their designs, (e.g.
through Web pages); receive feedback, to learn from others,
and if needed, to act collectively. Such a community can be a
Web-based ‘virtual’ community, that transcends the
boundaries of states, and their brick-and-mortar institutions.
• The Web-based network designer’s forum is not a necessity
to the project, but it is exposed here for consideration and
comments by the project team. CAPITA is interested in this
aspect of the project in the context of SMARTS
(http://capita.wustl.edu/smarts/smarts3.htm).
References
•
•
•
•
•
Caselton W.F. and Zidek J.V. (1984) Optimal monitoring network designs. Stat.
Prob. Lett. 2, 223-227
Frank N.H. (1996) Regulatory Monitoring Strategy for Revised PM NAAQS: A
Blueprint for a New National Monitoring Program for Particulate Matter. EPAOAQPS. Draft Feb 8, 1996, Do not Cite or Quote.
Program for Particulate Matter. EPA-OAQPS. Draft Feb 8, 1996, Do not Cite or
Quote.
Venkatram, A. (1988) On the use of kriging in the spatial analysis of acid
precipitation data. Atmospheric Environment 22, 1963-1975.
Warrick A. W. and Myers D.E. (1987) Optimization of sampling locations for
variogram calculations. Wat. Resour. Res. 23, 496-500