Compared with what its setting predicts
Watch Site 1 finds fewer pollution-sensitive animals than its setting predicts, and the gap appears to be widening (p = 0.055, about 11 percentage points a decade).
A model learned from every other site in the country what a survey should find given its season, survey form, watershed, recent rain, water temperature, flow and team. It never sees the year, so change over time stays in the gap between what a site finds and what it should find.
At Site 1 the gap appears to be widening (p = 0.055): the loss of sensitive animals is not explained by season, recent rain, water temperature, flow or team. Site 2 shows no clear trend (p = 0.35), and Site 1's overall score percentile shows no clear trend (p = 0.22).
With p between 0.05 and 0.1 this is a signal to watch rather than proof; a few more years of surveys, ideally in the same season each year, would settle it.
Each new survey can be checked against its expected range as soon as it is entered. Two of Site 1's last three surveys fell below the range, including the most recent on 2025-09-21.
| Date | Site | Form | Sensitive share found | Expected range | Status |
|---|---|---|---|---|---|
| 2026-05-17 | Site 2 | VA SOS rocky bottom | 44% | 23%–46% | within range |
| 2025-09-21 | Site 1 | VA SOS rocky bottom | 18% | 26%–50% | below expected |
| 2024-09-15 | Site 2 | VA SOS rocky bottom | 30% | 21%–44% | within range |
| 2024-09-15 | Site 2 | SOS Biological | 30% | 21%–44% | within range |
| 2024-07-14 | Site 2 | SOS Biological | 50% | 25%–48% | above expected |
| 2024-05-26 | Site 2 | SOS Biological | 57% | 22%–45% | above expected |
How this was worked out
Gradient-boosted trees, five folds grouped by site; residual trend by the seasonal Kendall test.