Validate, Qualify, and Release Lake Water-Quality Data: From Raw Record to Decision Dataset

Before reviewing values, identify the decision, data user, applicable QAPP or equivalent plan, methods, acceptance criteria, reporting pathway, and the person authorized to accept, qualify, reject, or release data. A dataset adequate for education or equipment troubleshooting may be inadequate for regulatory reporting, public-health communication, loading analysis, or treatment verification.

EPA quality guidance separates planned requirements, verification of completeness and conformance, validation of usability, and assessment against the decision need. Record which functions one reviewer may perform and which require independent or agency review.

Sources: [1], [2], [3]

1. 01

   Acquire originals

   Collect field records, logger files, calibration records, station/depth definitions, labels, custody and shipping records, laboratory reports and electronic deliverables, telemetry logs, maintenance notes, photographs, and governing method/QAPP versions.

2. 02

   Create a manifest

   Assign stable file and record IDs; record source, creator, acquired time, checksum or preservation evidence, retention, access classification, and whether each item is original, copy, or superseding record.

3. 03

   Lock the raw layer

   Make the original package read-only under the project retention system. Never overwrite a logger file, laboratory deliverable, signed field sheet, or original photograph with a cleaned version.

4. 04

   Open an issue log

   Record missing, duplicate, conflicting, corrupt, or late items without repairing them silently. Assign an owner, due date, impact, and resolution evidence.

Sources: [1], [6], [7]

Normalization is not validation. Each conversion must retain the original value and unit, the exact rule, and required context such as UTC offset or depth reference. A missing offset must remain missing rather than being guessed from station location.

Derived products should be reproducible from stable inputs. If a chart excludes flagged intervals, retain those records in the source dataset and record the explicit filter and release version.

Sources: [1], [6], [8]

• Project, event, station, sample, profile, instrument, sensor, accession, method, and result IDs are unique and cross-walked
• Collection, preservation, shipment, receipt, preparation, analysis, and telemetry times include time zone or UTC basis
• Coordinates include datum, accuracy/source, and station version; moved stations are not relabeled as unchanged
• Depth retains as-collected and canonical units, reference point, total depth, and bottom clearance where applicable
• Analyte, matrix, fraction, preparation, analytical method/version, unit, and wet/dry basis are explicit
• Instrument/sensor identity and configuration link to the correct calibration and maintenance record
• Conflicting metadata remain open findings until resolved with evidence

Sources: [1], [2], [5]

Compare pre/post checks, calibration standards and expiration, cleaning and fouling observations, clocks, stabilization, depth/station context, power and telemetry gaps, maintenance, and raw sensor behavior with project-specific correction and acceptance rules. Use related parameters as evidence, not permission to delete an unusual observation.

QARTOD-style range, spike, rate-of-change, flatline, climatology, or location tests require documented thresholds and test versions. Preserve each test result. Later manual review may add a separate conclusion; it should not erase the real-time flag history.

Sources: [5], [8], [6]

1. 01

   Close custody

   Match field IDs, container count, matrix/fraction, preservation, analyses, transfers, cooler condition where required, receipt, rejection notes, and accession IDs.

2. 02

   Confirm the deliverable

   Reconcile electronic and signed reports, counts, methods, dates, units, detection/reporting limits, dilutions, reanalyses, and case narrative.

3. 03

   Evaluate QC by the plan

   Review applicable blanks, duplicates, spikes, standards, surrogates, control samples, calibration, and completeness against exact project/method criteria, not universal limits borrowed from another method.

4. 04

   Propagate findings

   Link each QC finding to every affected result and retain qualifier, reason, reviewer, date, and fitness for the named decision.

Sources: [1], [2]

Keep the laboratory or data-system qualifier, detection condition, limit type/value/unit, dilution, and narrative with the result. EPA WQX guidance recommends an explicit detection condition and applicable detection-limit metadata for censored observations.

Do not replace nondetects with zero, one-half a limit, or another constant in the released observation field. If an approved analysis estimates censored values, place estimates in a derived product with method, assumptions, varying limits, sensitivity analysis, and code/version.

Sources: [4], [3], [2]

1. 01

   Flag

   Create a finding without changing the raw value. Name affected records, test or observation, severity, detection time, and reviewer.

2. 02

   Investigate

   Gather field, calibration, maintenance, custody, laboratory, telemetry, and related-parameter evidence. Record alternatives and unresolved uncertainty.

3. 03

   Decide

   Apply the project requirement and assign accepted, qualified, rejected, or pending-for-use with a reason and intended-use scope.

4. 04

   Correct or derive

   If an approved correction is warranted, create a linked value with rule, calculation, reviewer, and date. Never replace the source value.

5. 05

   Close and audit

   Record resolution, affected releases, corrective/preventive action, communication, and whether similar records require review.

Sources: [2], [5], [8]

• Release ID/version, review period, intended use, audience, owner, and approving authority recorded
• Dictionary defines stations, methods, matrix/fraction, units, qualifiers, flags, missing values, censored results, and derived fields
• Raw manifest, transformations, code/formula versions, QC findings, decisions, and exclusions are traceable
• Completeness, representativeness, precision/bias evidence, sensitivity, deviations, and unresolved issues evaluated against DQOs
• Charts identify provisional/qualified data, gaps, method/station changes, and release version
• Summary states what data support, do not support, and who controls health, compliance, or management decisions
• Archive, permissions, retention, correction contact, and superseded-release handling documented

Sources: [1], [3], [6]

When an error affects released data, identify records and products, reopen review, preserve the prior release, create a corrected version, document cause and decision, reapprove, and notify known users according to risk. USGS procedures exemplify unapproving, revising, marking, documenting, and reapproving rather than editing history invisibly.

Public communication should distinguish provisional, accepted, qualified, rejected, revised, and unknown data. A quality flag does not diagnose cause, declare water safe, or authorize treatment. Link claims to the exact release and retain null, adverse, or inconclusive results.

• Prior release retained and marked superseded
• Affected records, products, decisions, users, and date range identified
• Correction/reanalysis rule, reviewer, evidence, and approval recorded
• Revision note explains what changed, why, impact, and whether conclusions changed
• Dashboards, exports, reports, and recipients updated consistently
• Root cause and preventive action tracked when systematic

Sources: [7], [6], [2]