Monitoring heavy metal concentrations in turbid rivers: Can fixed frequency sampling regimes accurately determine criteria exceedance frequencies, distribution statistics and temporal trends?

The accurate determination of water quality criteria exceedance frequencies, distribution statistics (e.g., mean, median and percentile concentrations) and temporal trends in constituent concentrations is critical to effective water resources management. Here we examine the effect of sampling regime on the accuracy of trace metal concentration, water quality criteria exceedance, and trend statistics at three sites in a turbid and highly dynamic river (mean 1 standard deviation total suspended solids [TSS] concentration = 56 +/- 148 mg L-1, 278 +/- 777 mg L-1 and 521 +/- 909 mg L-1). Daily TSS data from the Red Deer River (RDR) in Alberta, Canada were used to generate a 10-year baseline data set of total Pb, Hg, Cu and Cd concentrations based on linear regression relationships. The baseline data was then sub-sampled to create fixed frequency (3-day, 7-day, 14- day, 30-day, 60-day and 90-day) and flow augmented (30-day + Q >= 90th percentile) regimes. Precision increased with increased sampling frequency for all statistics over both annual and decadal time scales. However, annual statistical estimates exhibited consistently poorer precision than estimates summarized over 10 years. For estimates of annual mean and 90th percentile concentrations, precision decreased as the variation in daily constituent concentrations in the baseline data set for each year increased. Estimates of median concentrations were generally more precise than the mean or 90th percentile, while estimates of criteria exceedance had particularly poor precision and exhibited systemic bias when the frequency of exceedance in the baseline data was low (i.e., < 10%). In terms of bias, estimates of mean, median and 90th percentile concentrations generally exhibited little to no systemic bias. Flow augmented sampling had similar or better precision than 14-day fixed frequency sampling (which had a similar sampling effort; i.e., n = 152-153) but resulted in large positive bias (median % error = 65-729%) for concentration and exceedance statistics. Considerable variation in estimates of trend statistics were observed when fixed frequency sampling was employed. Importantly, at a monthly frequency, significant trends (p < 0.1) were detected when a trend in the baseline data did not exist. Finally, based on a 20% error threshold, the application of fixed frequency sampling regimes (3-day, 7-day, 14-day and 30-day) failed to accurately estimate metal concentrations and criteria exceedances. Our research highlights the importance of considering the uncertainty associated with fundamental concentration statistics when designing and/or interpreting data from water quality monitoring networks in turbid river systems.