Systematic Sampling for Suspended Sediment

Because of high costs or complex logistics, scientific populations cannot be measured entirely and must be sampled. Accepted scientific practice holds that sample selection be based on statistical principles to assure objectivity when estimating totals and variances. Probability sampling--obtaining samples with known probabilities--is the only method that assures these results. However, probability sampling is seldom combined with appropriate estimators to determine suspended sediment loads. Many current load-estimating methods, therefore, have unknown bias and variation making the estimates questionable. Suspended sediment loads are often estimated by sampling concentration at fixed intervals. This type of sampling is-promoted by the widespread use of pumping samplers which can be set to sample at regular intervals. Sampling intensity is sometimes increased during periods of high water discharge. Randomly started systematic samples are probability samples, and estimates of totals from such samples are unbiased. These estimates tend to have low variance, but the variance cannot be estimated, and is not always reduced by increasing sample size. Systematic sampling of concentration distributes data evenly over time, so that most measurements are collected during low flows, and few during the brief high-flow periods when most sediment is transported. Systematic sampling for estimating suspended sediment loads is investigated for a "complete" sediment record from the Mad River in northern California. The "true" total for the 31-day period is compared to expected estimates from systematic samples without random starts. Systematic sample variance is compared to three other finite sampling schemes and to estimates using the simple random sampling variance formula. The effects of changing systematic sample size are also studied. If systematic sampling is used to estimate suspended sediment loads, the limitations of the method should be realized and correct estimating formulas used. The best use for systematic "sampling" is to define the sampled population for further sampling by more efficient finite sampling schemes.