This report has been prepared in response to the Acid Mine Drainage Task Force’s
request for a review of the effectiveness of the monitoring programs at existing mines,
and the design of optimum monitoring programs for the B.C. context. Although
conceived as a statistical exercise using existing data sets, the exercise immediately
foundered due to the inadequacy of the available data. This report documents our
conclusion that the existing fixed-frequency data sets are suited only for the description
of very long-term trends; accurate estimates of mean concentrations, loads and peak
values require different sampling methods. Having very little data to work with and an
obvious need for education regarding monitoring design, the emphasis of the project
shifted to writing a mini-text on monitoring design for ARD sites.
Section 1 begins with an examination of the monitoring methods currently used in Waste
Management permits for mines with acid drainage potential. The unreplicated fixed frequency
samples are shown to be inaccurate in estimating mean concentrations and
completely inadequate to indicate peak values and short-term fluctuations. Alternate
methods of monitoring are reviewed from the monitoring and statistical literature, each
with its own advantages and disadvantages. Different monitoring goals (e.g. detecting
long-term trends, accurately measuring excursions) are discussed with reference to the
different monitoring methods available. The point is made that no monitoring program
can be optimized statistically without clearly stated goals: a program that efficiently
measures monthly means would not also efficiently catch peak values. Rather than
burden industry with monitoring programs that attempt to measure all possible variations
for all possible contingencies, it is recommended that the Acid Mine Drainage Task
Force engage in an ‘Environmental Audit’ process to determine the specific goals of
monitoring for each site. This discussion leads to the first and most important
recommendation in the report: to critically examine the information needed for
management at each mine: accuracy, threshold concentrations, time lags, cost constraints
and risks for each ARD component. Monitoring results should be ‘defensible’, both in
the scientific and enforcement senses.
Chapter 2 is a review of basic sampling statistics as they are applied to water quality
data. The problems of dealing with rapidly fluctuating values are emphasized. The
technique of performing a preliminary sampling study of a site is described. Preliminary
studies determine the variances in different components of a site, and thus permit the
calculation of predicted accuracies of different sample sizes, selection of optimum strata,
and the allocation of future samples to optimize sampling efficiency. The lack of proper
preliminary sampling at any of the B.C. mines examined in this study made it impossible
to perform one of the initial goals of this project, which was to design optimum
monitoring methods for specific sites. Sampling design requires measures of variance,
which are lacking in unreplicated fixed-frequency data.
Understanding the process of the generation and release of ARD helps to focus a
monitoring program on critical time periods. Chapter 3 illustrates how the process
affects water quality sampling, with an emphasis on seasonal and flow-related effects.
The critical importance of good flow data at AKD sites is emphasized.
Chapter 4 is an exploration of the best monitoring data set available; a year’s worth of
almost daily data from a coastal mine. Day-to-day variations in concentration are high
and greatly exceed the analytical error of the mine’s environmental lab; i.e. the speed
with which a sample can be analyzed many be more important for getting an accurate
reading than the usual ‘quality assurance’ concerns of laboratory technique. Daily data
are compared with the monthly official monitoring record to illustrate the short-comings
of monthly sampling in a rapidly fluctuating system. Three different monitoring
schedules are designed for this mine to suit three different monitoring goals: peak
values, mean values and loads. For example, the error of the estimated annual zinc load
could be decreased by more than 60% by taking 6 additional samples (18 instead of 12).
This improvement is accomplished by allocating the samples according to the observed
seasonal variance pattern instead of fixed monthly intervals.
Chapters 5 and 6 contain general guidelines for the monitoring of untreated mine water
and monitoring in the receiving environment. This discussion was limited to generalities
because there were no data sets available that supported proper monitoring design or
even a rigorous determination of general confidence intervals or accuracy. The use of
experimental design to ensure that proposed field studies (both regular monitoring and
special studies) are more likely to have conclusive and useful results is very strongly
recommended. Section 6.6.3 illustrates what can happen when more effort is put into
trying to sample ‘everything’ rather than carefully identifying the information goals of the
monitoring program.
A brief discussion of biological monitoring as an alternative to water quality monitoring
is the main topic of Section 7. Biological samples integrate water quality over time, and
thus contain much more information than an accurate measure of an ephemeral quantity
such as dissolved concentrations. Any discussion of optimum water quality monitoring
would be incomplete if it did not point out the value of biological monitoring.
The theme of this document is that improved statistical meihodology for monitoring rests
on defining the information needed for good management. Too much emphasis has
been put on laboratory analysis techniques and on trying to apply statistics to squeeze
something out of existing data sets; not enough emphasis has gone into answering hard
questions about how defensible the monitoring data is. What degree of certainty is
needed on estimates? Does the data alert us when an environmental risk threshold has
been breached? Is it available in time to permit useful management responses? What
could we do better if we had the information? These are not statistical questions, but
they are of the greatest priority in optimizing ARD monitoring.