Sampling, Not Lab Error, Drives Uncertainty in Fluoride Monitoring of Plants

When regulators want to know whether industrial fluoride emissions are harming vegetation near aluminum smelters, they rely on measurements of fluoride concentrations in grass, birch leaves, and other plants. Those numbers are compared against permit limits and species-specific damage thresholds, and the decisions that follow can carry real ecological and economic weight. But a new study from Iceland delivers an uncomfortable message for anyone who treats a laboratory result as the whole truth: the act of collecting the plant sample, not the chemistry performed on it, is often the dominant source of uncertainty in the final figure. The work, published in Environmental Monitoring and Assessment by E. Antonsson, E. I. Eyjólfsdóttir, and K. Gíslason, provides one of the clearest demonstrations yet that environmental monitoring programs systematically underestimate how uncertain their vegetation data really are.

Fluoride is a natural component of the environment, released by volcanic activity and weathering of rocks, but aluminum smelting is a major anthropogenic source of airborne fluorides. Plants accumulate fluoride from the air on and in their foliage, which makes vegetation a convenient biomonitor for industrial emissions. The stakes extend beyond botany: grazing sheep and horses can ingest fluoride-laden forage, and chronic exposure leads to dental and skeletal fluorosis in livestock. Icelandic researchers have tracked fluoride accumulation in sheep and horses near the country’s smelters for more than a decade, and vegetation measurements underpin those assessments. If the measurement itself is unreliable, every downstream comparison to a limit value inherits that unreliability.

The technical framework behind the study comes from a body of metrological thinking that has gained traction over the past two decades. Under the Guide to the Expression of Uncertainty in Measurement, laboratories routinely quantify analytical uncertainty through repeated measurements of reference materials and replicate digestions. But since the late 1990s, researchers led by Michael Ramsey and colleagues have argued that sampling is part of the measurement process, and that its variability must be estimated empirically rather than assumed away. The Eurachem guide on measurement uncertainty arising from sampling, now in its second edition, codifies this approach, and the Icelandic team applied it in a field setting where it matters most.

Their method was the duplicate design, a structured sampling experiment that separates the two error sources. At each of a set of predetermined monitoring locations near aluminum smelters in Iceland, field teams collected duplicate vegetation samples following a standardized protocol, taking two independent samples at the same target. Each sample was then split or processed so that two separate chemical analyses could be performed on each. This creates a balanced hierarchy: differences between duplicate samples at the same location reflect sampling variability, while differences between duplicate analyses of the same sample reflect analytical variability. A nested analysis of variance, using the RANOVA2 software from the Royal Society of Chemistry’s Analytical Methods Committee, partitions the total variance into these components.

The vegetation studied comprised three species with different roles in the monitoring program: grass, which serves as a proxy for livestock forage; downy birch, a native woody species; and rowan, another tree species used as an indicator. Fluoride determination was carried out with ion selective electrodes, a well-established potentiometric technique in which the electrode potential of a prepared sample solution is converted to fluoride concentration through calibration. The technique is sensitive and widely used, but like any analytical method it carries its own uncertainty from weighing, ashing, alkali fusion or extraction, calibration standards, and electrode drift.

The results were striking. For grass and downy birch, sampling dominated the uncertainty budget outright, contributing more variance than the entire analytical procedure. For rowan, sampling and analytical uncertainties were of comparable magnitude. Expressed as relative expanded measurement uncertainty, the standard metric that combines both components and scales them to the measured concentration, the total uncertainty reached 43 percent for grass, 45 percent for downy birch, and 13 percent for rowan. In other words, for grass and birch, the true fluoride content of the vegetation could plausibly lie nearly half below or above the reported value. That margin dwarfs the few percent typically quoted by laboratories when they report analytical quality control data alone.

The implications for regulatory interpretation are immediate. Environmental permits for smelters specify fluoride limit values in vegetation, and exceedances can trigger operational changes or enforcement action. If a measured concentration sits close to a limit, a 45 percent expanded uncertainty means the compliance decision is genuinely ambiguous. Conversely, a monitoring program that wants to detect a genuine downward trend in emissions after abatement measures must contend with sampling noise that could mask or mimic such a trend. The authors emphasize that uncertainty estimates based solely on chemical analysis severely underestimate the total measurement uncertainty, a conclusion that applies broadly to vegetation biomonitoring programs worldwide, not just to Iceland.

Why would sampling be so variable? Vegetation is an inherently heterogeneous medium. Fluoride deposition varies with leaf age, position in the canopy, distance from the emission source, local wind patterns, and wash-off by rain. Two handfuls of grass taken a meter apart can differ substantially in fluoride content simply because of micro-scale variation in deposition and growth. Woody species like rowan, with larger and more uniform leaves, may integrate deposition more evenly, which is consistent with the lower relative uncertainty observed for that species. The duplicate method captures all of this real-world heterogeneity, which is precisely its strength: it measures the uncertainty of the entire measurement process as it is actually practiced, in the field, by real sampling teams.

The study also connects to a broader movement in analytical science. The theory of sampling, developed by Kim Esbensen and colleagues, has long called for integration between sampling science and classical measurement uncertainty estimation, and studies like this one put that principle into quantitative practice. Similar duplicate-based assessments have been applied to moss biomonitoring of heavy metals and to radioactivity measurements in environmental samples, with comparable findings that sampling often rivals or exceeds analysis as an uncertainty source. What the Icelandic work adds is species-specific numbers for a fluoride monitoring program tied directly to regulatory thresholds, giving practitioners concrete benchmarks for the first time.

For monitoring agencies, the practical lessons are clear. First, any vegetation monitoring program that reports analytical uncertainty alone is presenting an optimistic picture of its data quality, and programs should budget for periodic duplicate sampling campaigns to estimate the full uncertainty. Second, where sampling dominates, investing in better sampling design, more composite subsampling, or tighter field protocols will buy more accuracy than upgrading the laboratory. Third, when comparing measured fluoride concentrations to permit limits or damage thresholds, regulators should account for the expanded uncertainty, treating borderline results with appropriate caution. The authors frame their approach as a reliable way to incorporate sampling contributions into measurement uncertainty assessment in environmental monitoring using vegetation, and the numbers they report suggest that doing so is not optional. A monitoring result is only as good as the sample behind it, and in fluoride biomonitoring, the sample is where most of the doubt lives.

Subject of Research: Measurement uncertainty from sampling and chemical analysis in fluoride determination in vegetation near aluminum smelters

Article Title: Measurement uncertainty of fluoride determination in vegetation: contributions from sampling and analysis

Article References: Antonsson, E., Eyjólfsdóttir, E. I., & Gíslason, K. (2026). Measurement uncertainty of fluoride determination in vegetation: contributions from sampling and analysis. Environmental Monitoring and Assessment, 198(11), Article 1165. https://doi.org/10.1007/s10661-026-15981-z

Image Credits: AI Generated

DOI: 10.1007/s10661-026-15981-z

Keywords: fluoride, vegetation, measurement uncertainty, sampling, aluminum smelters, ion selective electrode, environmental monitoring, biomonitoring, Iceland, duplicate method, grass, downy birch