Routine monitoring networks provide a valuable source of data on environmental conditions, which can be used in integrated assessments.  Their ability truly to represent the full range of environmental conditions (or human exposures to them) is nevertheless limited, for the environment itself is highly variable, over both space and time, while monitoring is costly and technically constrained, so networks are limited both in their extent and what they measure.  These limitations need to be borne in mind whenever monitoring data are used for integrated assessments - whether as a basis for exposure assessment in their own right, as inputs to models, or as a basis for model validation.  By the same token, considerable care is needed in designing monitoring systems or measurement campaigns for the purpose of assessment. 

Estimating representivity

Determining the representivity of monitoring networks is not easy.  On the one hand, the concept of representivity is somewhat vague (and there is no agreement even about what to call it!), so that relevant measures are not always obvious (see references below).  Simple statistical measures are available to estimate sample sizes.  All such analyses, however, depend on assumptions about the underlying statistical distributions of the properties being measured - and these can only be deduced from the data provided by the existing monitoring network.  Inadequacies in the sample design of this network inevitably means that these assumptions are uncertain.  Unless additional monitoring can be carried out, at a sampling density sufficient to detect any unseen regularities or local variabilities in the environment, therefore, the estimates of sample requirements are likely to be unreliable (and in many cases to under-estimate the true sampling density that is required, or to define the most effective sampling configuration. 

Nor, in the case of assessment, is the the need usually just to minimise the standard errors of the estimates.  Instead, monitoring may have to ensure that:

• hotspots and vulnerable population sub-groups are correctly identified;
• uncertainties are reasonably equal across different parts of the study area and population;
• a strong correlation exists between predicted and actual conditions across the whole study area/population;
• monitoring costs are kept within necessary limits. 

This implies the use of a number of different measures, reflecting different design criteria, to estimate representivity.  Optimising them all is rarely possible, for the different criteria are to some extent in conflict, so trades off have to be made between them.  Representivity is not an absolute, therefore, but depends on the needs of the study, and can only be judged in terms of context.

An illustration of many of these considerations, and an example of how different network designs can affect estimates of population exposures to environmental pollution is given in the attached document, on the representivity of air pollution monitoring networks.  This was developed through a simulation exercise, using the SIENA urban simulator.   A detailed report can be downloaded below. 

Tools for sampling design

A number of tools to help design monitoring systems and sampling strategies exist, including the Visual Sample Plan (VSP), developed by Battelle Pacific Northwest National Laboratory.  This provides an interactive and highly flexible, map-based system for estimating sample size and selecting between different sampling configurations, for a wide range of different user-defined purposes, including estimation of the sample mean, trend detection and exceedance analysis.   

File Attachments: 

Representativeness of monitoring networks.pdf