The Gold Standard · Ecology
Make the decision path visible
A recommissioned site is often calibrated under the conditions available at restart. Those conditions may cover ordinary flow, moderate temperature or one biological season, while the monitoring purpose depends on floods, droughts, winter conditions or short-lived ecological events. Applying the restart calibration outside its evidence range is a transfer claim.
Good transfer design states the source conditions, target conditions and mechanisms that could change the relationship. It gathers independent measurements across seasons and event ranges, distinguishes real ecological change from measurement drift and refuses to hide unsupported extrapolation inside a smooth time series.
Choose the route closest to your decision
Chapter contents
Chapters 1–3
Chapters 4–6
Chapters 7–9
Chapters 10–12
CHAPTER 1 OF 12 · BOUND THE CALIBRATION
1. Write the source calibration domain
Record the sensor, installation, datum, preparation or rating model, independent standards, dates and environmental conditions used to calibrate the restarted site.
Describe the observed range, not just its minimum and maximum. A few measurements at one edge provide different evidence from dense coverage across the range.
Keep commissioning adjustments separate from the accepted model. The final calibration needs a version, effective date and identifiable supporting observations.
CHAPTER 2 OF 12 · BOUND THE CALIBRATION
2. Define the target conditions that matter
Map the monitoring decisions to seasons, event types and operating ranges. A streamgage may need high-flow validation; a water-quality sensor may need warm, low-flow and fouling conditions; an acoustic station may need seasonal soundscapes.
Name combinations, not only single variables. Temperature and concentration, flow and sediment, or humidity and signal attenuation can interact.
Prioritise conditions by consequence and frequency. Rare extremes can still deserve strong evidence when they drive safety or ecosystem decisions.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 3 OF 12 · BOUND THE CALIBRATION
3. Identify transfer mechanisms
Calibration can fail through sensor response, mounting, environmental control, biofouling, matrix composition, algorithm behaviour, time synchronisation or a changed relation between the measured proxy and ecological quantity.
A stable laboratory calibration does not guarantee stable field conversion. The surrounding channel, habitat or sample matrix may change.
Write competing mechanisms before inspecting anomalies. This makes field checks more informative and reduces after-the-fact storytelling.
| Element | Specify |
|---|---|
| Source domain | Conditions used for calibration |
| Target domain | Seasons and event ranges needed |
| Mechanism | Why the relation could change |
| Evidence | Independent checks required |
CHAPTER 4 OF 12 · DESIGN TRANSFER EVIDENCE
4. Design seasonal validation
Schedule independent checks across expected seasonal states, but also use trigger-based sampling when temperature, vegetation, flow or biological activity crosses important conditions.
Repeat some reference measurements under comparable conditions so seasonal effects are not confused with new operators or procedures.
If a season is missed, label the transfer unvalidated. Absence of an alert during that period is not evidence that the calibration held.
CHAPTER 5 OF 12 · DESIGN TRANSFER EVIDENCE
5. Capture event ranges safely
Events may be brief, dangerous and hard to sample. Use redundant sensors, remote standards, autosamplers or planned post-event surveys where suitable.
Define safe operating limits and do not send staff into hazardous conditions merely to complete a calibration table. Evidence design must include safety and permissions.
Record hysteresis and timing. The relation on the rising limb of an event may differ from the falling limb even at the same measured level.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 6 OF 12 · DESIGN TRANSFER EVIDENCE
6. Work a calibration-transfer case
A stream site restarts in the dry season. Twelve discharge measurements validate the new rating from 0.2 to 1.8 metres, but the historical decision range extends to 4 metres.
During the first monsoon, remote data reach 3.2 metres. The team marks high-range discharge provisional, obtains safe post-event surveys and four high-flow measurements, and finds that sediment changed the control above 2.4 metres. A segmented rating replaces the original extrapolation, while observations below 1.8 metres remain linked to the restart calibration.
Changed-context transfer check: at a temperature logger, the target may be a cold-season range rather than high flow. The same method maps source and target support, but the independent standard and failure mechanism change.
| Dimension | Examples |
|---|---|
| Season | Wet, dry, growth, dormancy |
| Range | Low, routine, high, extreme |
| Event phase | Before, rising, peak, falling |
| Environment | Temperature, sediment, fouling, humidity |
CHAPTER 7 OF 12 · UPDATE THE MEASUREMENT MODEL
7. Diagnose seasonal drift and real change
Compare sensor residuals against independent standards and ecological covariates. A seasonal pattern common to several trustworthy sites may be real; a site-specific reference offset suggests measurement change.
Inspect missingness and telemetry. Event-selective data loss can bias the apparent calibration even when observed values are accurate.
Use statistical breakpoints as prompts, not physical explanations. Confirm with inspections, surveys, standards and maintenance records.
CHAPTER 8 OF 12 · UPDATE THE MEASUREMENT MODEL
8. Update ratings without rewriting evidence
When transfer fails, create a new calibration or rating version with an effective interval and supporting data. Preserve original values and released products.
Avoid one blanket offset when the error changes by range, season or event phase. Use segmented or covariate-aware models only when evidence supports them.
Report uncertainty and provisional status with the data. A precise-looking corrected value is misleading when the breakpoint or mechanism remains uncertain.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 9 OF 12 · UPDATE THE MEASUREMENT MODEL
9. Bridge comparable portions of the series
Some ranges may remain comparable even when extremes require a new model. State which conditions are bridged and which require separate treatment.
Recompute dependent summaries only for affected intervals and trace downstream models or sites that borrowed the data.
Do not smooth across genuine ecological transitions. The purpose of a bridge is measurement comparability, not visual continuity at any cost.
| Finding | Action |
|---|---|
| Within supported domain | Use accepted calibration |
| Sparse target evidence | Release provisionally |
| Range-dependent difference | Segment or limit model |
| Fundamental mechanism change | New series or calibration regime |
CHAPTER 10 OF 12 · GOVERN ONGOING USE
10. Set provisional-release rules
Define when data can be released as provisional, what qualification accompanies them and what evidence is needed for approval.
High-impact decisions may require tighter restrictions than exploratory research. Preserve raw observations even when derived values are withheld.
Track how long records remain provisional. Indefinite provisional status can hide a validation backlog and leave users unable to judge fitness.
CHAPTER 11 OF 12 · GOVERN ONGOING USE
11. Use network evidence carefully
Neighbouring sites, redundant instruments and regional models can flag divergence, but they are not automatic truth. Check habitat, catchment, exposure and their own calibration status.
A recommissioned site may anchor estimates elsewhere. When its calibration changes, reassess dependent gaps, trends and alerts.
Publish version and qualification information so downstream users can reproduce which model produced each value.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 12 OF 12 · GOVERN ONGOING USE
12. Know when transfer is unsupported
Do not extrapolate when target conditions lie beyond observed support, mechanisms change fundamentally or independent evidence is too sparse for the consequence.
Options include reporting the direct observation without a derived conversion, restricting the range, adding a distinct seasonal model or beginning a new series.
The final record should state the current calibration domain, unvalidated ranges, event evidence, seasonal checks, version history and next trigger.
| Check | Pass condition |
|---|---|
| Version | Every value maps to a model |
| Range | Unsupported extrapolation labelled |
| Revision | Original release preserved |
| Dependencies | Affected downstream uses reviewed |
Frequently asked questions
Can a restart calibration be used through the whole year?
Only when evidence or a defensible physical model supports the seasonal transfer. Otherwise validate missing conditions explicitly.
Are high-event values invalid until direct sampling occurs?
They may be reported provisionally with clear limits, depending on consequence and available independent evidence.
Does seasonal residual movement prove sensor drift?
No. It may reflect real ecology, environmental control or model mismatch. Triangulate with standards and comparable sites.
When should a new series begin?
When calibration domains cannot be bridged, the measurement mechanism changes fundamentally or uncertainty makes continuity misleading.
Continue through verified routes
Core routes
Ecology subject root · Gold Standard of Education · Gold Standard directory · How X Works hub
Specialist routes
- Recommissioned-site drift surveillance
- Sentinel-network recommissioning
- Sentinel-site network drift
- Reference drift and nonstationary baselines