
A Guardian report dated 14 November 2026 describes Indigenous students retracing the route of the Lewis and Clark expedition, with a new wildlife survey carried out along the way. The summary recalls that in 1804 Thomas Jefferson sent Meriwether Lewis, William Clark and the Corps of Discovery into the land west of the Mississippi, and that the president's primary aim was political. The headline offers a phrase that every monitoring program should take seriously: “We can't protect what we cannot see.” (Indigenous students retrace Lewis and Clark's historic journey, by Ashley Stimpson, The Guardian.)
For a water utility, the equivalent problem is not wildlife but coverage. A sampling network is a set of points. Each station produces a result for one location, one set of analytes, and the days it was visited. Everything between stations, and everything that happened between visits, is inferred. Most of the time that inference is reasonable. It fails in predictable places: tributaries that enter between the last station and the intake, storm runoff that arrives between scheduled grabs, and contaminants that are not on the lab panel at all.
Three states that get collapsed into one
A monitoring record usually contains three different outcomes, and reports often merge them. A location can be sampled with the analyte detected. It can be sampled with the analyte not detected above the reporting limit. Or it can be unsampled, which leaves no result at all. Only the first two are measurements. The third is an absence of data, and an absence of data says nothing about whether conditions were clean. A map that shades unsampled reaches the same way it shades non-detect reaches will look more complete than the record supports.
The distinction matters because a lab report does not carry it on its own. A “not detected” line on a certificate describes the sample in the bottle, measured against the method's limit on the day it was run. It does not describe the reach upstream, the week before, or the analytes left off the request. Writing these limits into the data record, rather than into a footnote, is the inexpensive part of the job.
Coverage has a cost curve
More stations and more frequent visits reduce the inference gap, but they raise sampling, transport, analytical, and staffing costs at a rate that is not linear. Storm sampling in particular requires staff available on short notice, and that availability is hard to buy in advance. Fewer stations push more of the risk onto modeling and assumptions, which are harder to audit later. Neither choice is wrong by default. The problem appears when the network was designed once, years ago, and nobody has rechecked which reaches now carry the most weight in the decisions the utility makes.
What to write down
A coverage register is a short document, and most of it can be built from existing records. For each source reach or intake approach, list the stations that inform it, the sampling frequency, the analyte groups tested, the date of the last result, and the events the schedule is known to miss. Then mark each gap as either a known blind spot or an open question. Review the register when land use changes, when a discharge permit is issued or closed, or when a new lab method changes what can be detected.
The Lewis and Clark story is a reminder that measurement programs are often older than the questions they now have to answer. The students are conducting a survey, and that survey will show what it can show. A water sampling plan works the same way. Its results are valid for the points it reached and silent about everything else, and the utility's obligation is to say where that silence begins.