Internal workplan for EPA Alternate Test Procedure approval of the Virridy Lume, structured as two parallel tracks: Track A — Colorado freshwater, beginning with a Boulder Creek limited-use ATP (Reg 93 / 303(d) compliance monitoring), expanding to Colorado state-wide, then nationwide freshwater; Track B — coastal applications via ASBPA, independent of the Colorado work, targeting enterococci at ocean beaches.
This plan details Virridy’s path to EPA Alternate Test Procedure (ATP) approval for the Lume sensor, establishing tryptophan-like fluorescence (TLF) coupled with machine learning as a recognized analytical method under 40 CFR Part 136 for continuous microbial water quality monitoring. The plan runs as two independent tracks:
| Track A — Colorado Freshwater | |
| Partner | City of Boulder Utilities · CDPHE |
| Initial Focus | Colorado Regulation 93 / 303(d) compliance monitoring — Boulder Creek, 6 monitoring sites (E. coli geometric mean threshold: 126 CFU/100 mL) |
| Reference Method | IDEXX Colilert-18 (E. coli, freshwater) |
| State Regulator | Colorado Department of Public Health and Environment (CDPHE) |
| Federal Regulator | EPA Region 8, Denver |
| Track A Pathway | Phase 1: Boulder Creek limited-use (40 CFR 136.5) → Phase 2: Colorado state-wide → Phase 3: Nationwide freshwater (40 CFR 136.4) |
| Track B — Coastal (Parallel / Independent) | |
| Partner | American Shore & Beach Preservation Association (ASBPA) — coastal/marine sites, other states |
| Target Analyte | Enterococci (marine/coastal water) |
| Reference Method | IDEXX Enterolert |
| Track B Pathway | Coastal limited-use ATP → Nationwide coastal ATP (coordinates with Track A for combined 40 CFR 136.4 submission) |
| Existing Data | |
| Performance | 856 paired observations vs. Colilert. R² = 0.67, kappa 0.82–0.84. See Comparability Report |
The Lume has 856 paired observations against the IDEXX Colilert-18 reference method—the same EPA-approved Part 136 method used by the City of Boulder for weekly recreational water quality reporting. All datasets (field, laboratory, and drinking water) use Colilert as the consistent reference, providing a uniform basis for comparability. The existing data demonstrates strong agreement: R² = 0.67 and 7% MAPE on continuous regression, 92–95% balanced accuracy on categorical classification (kappa 0.82–0.84), and a conservative error direction that favors public health protection.
| Evaluation | Key Metric | Value | n |
|---|---|---|---|
| Continuous regression (Boulder Creek) | R² / MAPE | 0.67 / 7.12% | 38 |
| 3-class categorical (<10, 10–100, >100 MPN) | Bal. Acc / Kappa | 95% / 0.84 | 334 |
| Binary (threshold = 1 CFU/100 mL) | Accuracy / Kappa | 91% / 0.82 | 361 |
| Binary (threshold = 10 CFU/100 mL) | Accuracy / Kappa | 92% / 0.84 | 361 |
| Chlorine residual detection (binary) | Accuracy / Kappa | 85% / 0.70 | 66 |
For complete analysis including raw paired data, per-class performance, confusion matrices, chlorination effects, and method precision, see the Lume–Colilert Comparability Report. For interactive performance charts and the full publication list, see Research & Performance.
Two parallel tracks. Track A (Colorado freshwater): regulatory engagement → Boulder Creek field study → method documentation → statistical analysis → limited-use ATP (Boulder Creek) → Colorado state-wide expansion → nationwide freshwater ATP. Track B (Coastal / ASBPA): runs independently in other states, targeting enterococci at ocean beaches, and feeds the nationwide coastal ATP submission. Quarters shown are relative to project kickoff (Q1 = Months 1–3).
| Workstream | Q1 | Q2 | Q3 | Q4 | Q5 | Q6 | Q7 | Q8 | Q9 | Q10 |
|---|---|---|---|---|---|---|---|---|---|---|
| Track A — Colorado Freshwater (E. coli / Colilert) | ||||||||||
| 1. Regulatory Engagement | ||||||||||
| 2. Boulder Creek Field Study | ||||||||||
| 3. Method Documentation | ||||||||||
| 4. Statistical Analysis | ||||||||||
| 5. Limited-Use ATP — Boulder Creek | ||||||||||
| 5b. Colorado State-Wide Expansion | ||||||||||
| Track B — Coastal / ASBPA (Enterococci / Enterolert, parallel & independent) | ||||||||||
| 6. Coastal Beach Validation (ASBPA) | ||||||||||
| 7. Nationwide Recreational Water ATP | ||||||||||
| Key Milestones | ||||||||||
Active Planned Future Milestone
Ongoing coordination with CDPHE and EPA Region 8 throughout the project. Front-loaded in Q1–Q2 to align study design with regulatory expectations before committing to the full field deployment. The regulatory context is Colorado Regulation 93—CDPHE’s 303(d) program for impaired waters—under which Boulder Creek carries E. coli impairment. The initial ATP application will be scoped to support Reg 93 compliance monitoring at Boulder’s six monitoring points.
| Task | Timing | Details |
|---|---|---|
| Introductory briefing to CDPHE Water Quality Control Division | Month 1 | Present the Lume technology, existing Boulder Creek validation data (research page), and ATP intent. Identify CDPHE staff lead for ongoing coordination. |
| Review Colorado Regulation 93 / 303(d) requirements for Boulder Creek | Month 1–2 | Confirm the Reg 93 E. coli impairment listing and TMDL threshold (126 CFU/100 mL geometric mean) for Boulder Creek. Identify Boulder’s existing monitoring obligations at the 6 monitoring points and determine how continuous Lume data maps to Reg 93 compliance reporting requirements. |
| CDPHE feedback on study design | Month 2–3 | Share proposed data collection protocol (see Section 10). Get CDPHE input on sample counts, seasonal coverage, and any Colorado-specific requirements. |
| Task | Timing | Details |
|---|---|---|
| Pre-submission meeting request to EPA Region 8 | Month 1 | Contact EPA Region 8 Water Technical Unit in Denver. Request formal pre-submission consultation per ATP program guidance. |
| Prepare briefing package | Month 1–2 | Package includes: technology summary, published validation data, proposed method scope (E. coli in freshwater and enterococci in marine water via TLF-ML for recreational water monitoring), proposed study design, draft QA/QC framework, and specific questions for EPA (see below). |
| Pre-submission meeting | Month 3–4 | Formal meeting with EPA Region 8. Present briefing package, receive guidance on study design, statistical requirements, and any Region 8-specific considerations. |
| Parallel consultation with EPA OST (headquarters) | Month 3–4 | Contact Lemuel Walker at EPA Office of Science and Technology (walker.lemuel@epa.gov, 202-566-1077). OST handles nationwide ATP review; early engagement ensures Region 8 and OST expectations are aligned. |
How should continuous sensor readings map to discrete compliance determinations? Options: instantaneous readings at defined intervals, rolling averages, time-weighted composites. This is the fundamental framing question.
Does EPA consider the ML component a “black box” concern? How should model versioning, retraining, and site-specific calibration be documented in the method? What transparency is required?
Confirm that TLF-ML is classified as a “new method” (new determinative technique) rather than a “modified method.” This determines which validation protocol applies.
Can the existing Boulder Creek paired dataset (see research) supplement the formal validation study, or must all regulatory data be collected under the finalized protocol?
For limited-use approval at Boulder’s sites, is single-operator (Virridy + Boulder) data sufficient, or does EPA Region 8 require independent laboratory operation even for facility-specific approval?
Does CDPHE impose additional requirements beyond federal ATP? Are there Colorado water quality standards for E. coli that differ from federal thresholds?
The core data collection effort. Deploy Lume sensors at all 6 of the City of Boulder Utilities’ recreational water monitoring locations on Boulder Creek. Collect continuous TLF data paired with Boulder’s existing weekly Colilert E. coli grab samples.
| Task | Timing | Details |
|---|---|---|
| Formalize MOU with City of Boulder Utilities | Month 1 | Define roles, data sharing, access to compliance sites, sampling coordination, and liability. Boulder continues their existing Colilert program unchanged; Virridy adds sensor deployment. |
| Site survey of all 6 compliance locations | Month 1–2 | Visit each site. Assess: mounting infrastructure, power availability, cellular/data connectivity, flow characteristics, security, and seasonal access constraints (ice, high water). |
| Sensor procurement & pre-deployment calibration | Month 1–2 | Prepare 8 Lume sensors (6 primary + 2 spares/rotation for maintenance). Factory calibration with tryptophan standards. Document sensor serial numbers, firmware versions, and initial calibration data. |
| Install sensors at all 6 sites | Month 2–3 | Staggered installation over 2–4 weeks. At each site: mount sensor, verify data transmission, confirm measurement interval (15-min default), photograph installation, record GPS coordinates. Begin continuous logging. |
| Validate data pipeline | Month 3 | Confirm all 6 sensors are transmitting reliably to the Lume data platform. Set up automated data quality alerts (signal dropout, out-of-range readings, battery/power status). |
| Task | Frequency | Details |
|---|---|---|
| Continuous Lume TLF measurement | Every 15 min | Automated. Each reading: TLF intensity (ppb), turbidity (NTU), temperature (°C), ML model E. coli estimate (CFU/100 mL), categorical classification, model confidence score. |
| Paired Colilert grab sample (Boulder’s program) | Weekly | Boulder staff collect grab samples at each recreational water monitoring site per their existing reporting schedule. Log exact sample collection time to match against nearest Lume reading. |
| Enhanced sampling during events | As needed | Increase Colilert grab frequency during high-flow events, storm runoff, and any known upstream contamination events. Target: 2–3x weekly during events to increase paired observations at high concentrations. |
| Field QC: duplicate sensor deployment | Monthly rotation | Deploy a second Lume sensor at one site (rotating) for 1 week per month. Generates sensor-to-sensor precision data (relative percent difference between paired sensors). |
| Field QC: blank verification | Monthly | Submerge sensor in DI water before and after each maintenance visit. Verify TLF reading is below MDL. Document any drift. |
| Sensor maintenance | Bi-weekly | Clean optical window, check mounting integrity, download local backup data, verify calibration against tryptophan standard. Log all maintenance activities with timestamps. |
| Data review & QA flagging | Weekly | Review incoming data for each site. Flag and document: fouling events, power interruptions, sensor malfunctions, outliers. Apply data quality flags per QA/QC protocol. |
| Metric | Target | Rationale |
|---|---|---|
| Study duration | 12–18 months minimum | Must span all four seasons to capture temperature, flow, and biological variability. Two full spring runoff cycles preferred. |
| Paired observations per site | ≥52 (weekly × 12 months) | 52 is the baseline from weekly Colilert. Enhanced event sampling should push this to 70–100+ per site. |
| Total paired observations (all sites) | ≥400–600 | 6 sites × 70–100 each. Substantially exceeds typical EPA multi-lab requirements. |
| Concentration range | <1 to >1,000 CFU/100 mL | Must cover the full range relevant to recreational water quality criteria. High concentrations during spring runoff and storm events; low during baseflow. |
| Continuous data completeness | ≥90% uptime per sensor | Account for maintenance windows, fouling, and equipment failures. Two spare sensors provide coverage during repairs. |
| Duplicate sensor precision data | ≥12 one-week deployments | One per month across different sites = reproducibility data across conditions. |
Write the complete analytical method in EPA standard format. Begins after EPA pre-submission meeting (WS 1) confirms requirements. Draft in parallel with data collection (WS 2).
| § | Section | Content |
|---|---|---|
| 1 | Scope and Application | Target analytes: E. coli in freshwater recreational water (rivers, lakes, beaches) and enterococci in marine/coastal recreational water. Applicable concentration range (MDL to upper reporting limit). Applicable water matrices. Limitations and exclusions. |
| 2 | Summary of Method | TLF measurement at 275/340 nm excitation/emission. Multi-parameter correction (turbidity, temperature). ML model inference (gradient-boosted decision trees). Output: continuous E. coli concentration estimate and categorical risk classification. |
| 3 | Definitions | TLF, tryptophan-like fluorescence, ppb tryptophan equivalents, categorical classification bins, calibration period, universal model, site-calibrated model, model version. |
| 4 | Interferences | Dissolved organic carbon (DOC), humic-like fluorescence (HLF), turbidity > [threshold TBD] NTU, temperature effects, biofouling, optical window contamination. Reference: known limitations on research page. |
| 5 | Safety | No hazardous reagents. Standard field safety for in-water sensor deployment. Electrical safety for powered sensor housing. |
| 6 | Equipment and Supplies | Lume sensor specifications (optical, physical, electrical). Mounting hardware. Data transmission module. Tryptophan calibration standards. DI water for blanks. Maintenance tools (cleaning supplies, spare optical windows). |
| 7 | Reagents and Standards | Tryptophan stock solution preparation. Calibration standard concentrations. Standard storage and shelf life. DI water specification for blanks. |
| 8 | Sample Collection | In-situ measurement: no sample collection, preservation, or transport required. Define sensor positioning relative to water surface, minimum submersion depth, flow velocity requirements. For paired reference samples: defer to existing Part 136 sample collection methods. |
| 9 | Calibration | Factory calibration procedure. Field calibration verification (tryptophan standard check, blank verification). Site-specific ML model calibration: minimum training dataset size, calibration data requirements, model acceptance criteria. Calibration frequency. Recalibration triggers. |
| 10 | Quality Control | Initial demonstration of capability (IDC). Ongoing QC: field blanks, calibration verification, duplicate sensor deployments, data completeness requirements. Corrective actions for QC failures. Data quality flags and acceptance criteria. |
| 11 | Procedure | Step-by-step: site selection criteria, sensor deployment, initialization sequence, measurement interval configuration, data transmission verification, routine maintenance, data retrieval, decommissioning. |
| 12 | Data Analysis | ML model application: input features, model version, inference procedure. Output: E. coli concentration (CFU/100 mL), categorical classification, confidence interval. Reporting: time-stamped results, averaging period for compliance reporting, data completeness requirements. |
| 13 | Method Performance | MDL, precision (RSD from duplicate sensors), accuracy (bias vs. Colilert), comparability statistics. Matrix-specific performance tables. Reference to Boulder Creek validation data and published literature. |
| 14 | Waste Management | No waste generated during routine operation. Disposal of calibration standards per local regulations. Sensor end-of-life: electronic waste recycling. |
| Task | Timing | Details |
|---|---|---|
| Obtain EPA method template / formatting requirements | Month 3–4 | From pre-submission meeting or EPA guidance documents. Ensure format matches what OST expects for review. |
| Draft Sections 1–8 (instrument and procedure) | Month 4–6 | These sections are largely defined by the existing sensor design. Can be drafted before validation data is complete. |
| Draft Sections 9–10 (calibration and QC) | Month 6–9 | Requires input from field QC data (WS 2). Define acceptance criteria based on observed sensor performance. |
| Draft Sections 11–14 (procedure, analysis, performance) | Month 9–15 | Section 13 (performance) requires complete validation data from WS 2 and statistical analysis from WS 4. |
| Internal review and revision | Month 15–16 | Technical review by Virridy team. External review by independent method documentation expert if budget allows. |
| CDPHE / EPA Region 8 review draft | Month 16–17 | Share draft with regulators for informal feedback before formal submission. |
Formal statistical analyses required for the ATP submission, following EPA’s validation protocols and Appendix H comparability procedures.
| Task | Details |
|---|---|
| MDL determination per 40 CFR Part 136 Appendix B | Minimum 7 replicate measurements of a low-level standard (near expected detection limit). Calculate MDL = t-value × standard deviation. Perform in both laboratory (controlled tryptophan solution) and field (low-concentration site) conditions. |
| E. coli MDL (derived) | Translate TLF MDL to E. coli concentration MDL via the ML model. Report as CFU/100 mL. Compare against existing literature value of ~10 CFU/100 mL (research page). |
| Task | Details |
|---|---|
| Sensor-to-sensor reproducibility | From duplicate sensor deployments (WS 2, monthly rotation): calculate relative percent difference (RPD), relative standard deviation (RSD), and coefficient of variation (CV) across all paired sensor observations. |
| Temporal precision | Assess variability of sensor readings during stable conditions (e.g., low-flow baseflow periods). Characterize instrument noise vs. real environmental signal. |
| Cross-site precision | Compare model performance metrics across all 6 Boulder Creek sites. Assess whether site-specific calibration is required or whether a universal Boulder Creek model is sufficient. |
| Analysis | Details |
|---|---|
| Continuous regression | Lume E. coli estimate vs. Colilert result for all paired observations. Report: R², RMSE (raw and log-transformed), MAPE, slope and intercept of regression line, prediction intervals. Existing Boulder Creek performance: R² = 0.67, MAPE = 7%. |
| Categorical classification | At regulatory thresholds (e.g., 126, 235, 410 CFU/100 mL for E. coli recreational criteria; plus any Boulder-specific permit limits): confusion matrices, accuracy, balanced accuracy, Cohen’s kappa, sensitivity, specificity, positive predictive value, negative predictive value. |
| Bias assessment | Bland-Altman analysis (difference vs. mean) to assess systematic bias across the concentration range. Paired t-test or Wilcoxon signed-rank test for overall method bias. |
| Matrix-specific performance | Stratify all analyses by: site, season (winter/spring/summer/fall), flow condition (baseflow/event), concentration range (low/medium/high), turbidity range, and temperature range. |
| Equivalence testing | Two one-sided t-test (TOST) or EPA-specified equivalence procedure to formally demonstrate that the Lume method produces results statistically equivalent to Colilert within defined bounds. |
| Task | Details |
|---|---|
| Model architecture description | Gradient-boosted decision tree. Document: number of trees, max depth, learning rate, feature set (TLF, turbidity, temperature, derived features). Gradient-boosted trees are inherently interpretable (feature importance, partial dependence plots). |
| Feature importance analysis | SHAP values or permutation importance for all input features. Demonstrates that TLF is the primary driver with turbidity/temperature as corrections—not a black box. |
| Model versioning protocol | Define: how model versions are tracked, when retraining is required, minimum retraining dataset size, performance acceptance criteria for new model versions, rollback procedures. |
| Universal vs. site-calibrated performance | Report performance of: (a) universal model (no site-specific training), (b) site-calibrated model (with Boulder Creek training data). Existing data shows 75%+ out-of-the-box accuracy, >94% with calibration. |
Facility-specific ATP approval under 40 CFR 136.5, submitted by the City of Boulder to EPA Region 8 for Colorado Regulation 93 / 303(d) compliance monitoring on Boulder Creek. This is the near-term regulatory milestone: a site-specific approval that allows Boulder to use Lume data in fulfillment of its Reg 93 impaired-waterbody monitoring obligations, with the TMDL-driven threshold of 126 CFU/100 mL geometric mean as the key compliance metric.
| Task | Timing | Details |
|---|---|---|
| Compile Boulder Creek paired dataset | Month 13–14 | After 12 months of data collection. Merge Lume continuous data with Boulder’s Colilert results. Apply QA/QC flags. Generate site-specific performance statistics. |
| Prepare site-specific comparability report | Month 14–15 | Statistical analysis from WS 4 tailored to Boulder’s specific sites and permit conditions. Include seasonal breakdown, flow-condition breakdown, and performance at Boulder’s permit limit thresholds. |
| Draft limited-use application | Month 15–16 | Application per 40 CFR 136.5 requirements. Includes: method description, comparability data, QA/QC protocol, proposed reporting framework (how continuous data will be used for recreational water quality monitoring). |
| City of Boulder internal review & approval | Month 16–17 | Boulder Utilities reviews and approves the application. Boulder is the applicant under 136.5—Virridy provides technical support. |
| CDPHE pre-submission review | Month 16–17 | Share draft with CDPHE for informal feedback before formal submission to EPA Region 8. CDPHE support strengthens the application. |
| Task | Timing | Details |
|---|---|---|
| Submit to EPA Region 8 | Month 17 | City of Boulder submits the limited-use ATP application to the EPA Region 8 Administrator. |
| Respond to EPA information requests | Month 17–20 | Anticipate questions and requests for additional data. Have supplementary analyses pre-computed (e.g., alternative statistical approaches, additional matrix-specific breakdowns). |
| EPA Region 8 approval | Month 18–21 | If approved, EPA issues a limited-use ATP approval letter. Boulder can begin using Lume data for recreational water quality monitoring on Boulder Creek. |
| Task | Details |
|---|---|
| Compliance reporting protocol | Define how Boulder reports Lume data for recreational water quality monitoring. Agree on: averaging period, reporting frequency, data completeness thresholds, and backup procedure (revert to Colilert if sensor data completeness falls below threshold). |
| Parallel monitoring period | Continue Colilert sampling in parallel with Lume for an initial period (6–12 months) after limited-use approval. Builds confidence and provides ongoing comparability data. |
| Performance monitoring & reporting | Quarterly performance reports to EPA Region 8 and CDPHE during the parallel monitoring period. Document ongoing accuracy, precision, and any issues. |
After the Boulder Creek limited-use ATP is approved, expand the Lume’s recognized use to additional Colorado waterbodies listed under Reg 93 / 303(d). The goal is to build a body of multi-site, multi-condition evidence that supports both CDPHE statewide recognition and the nationwide freshwater ATP submission.
| Task | Timing | Details |
|---|---|---|
| Identify target Colorado watersheds | Month 18–20 | Select 2–4 additional Colorado 303(d)-listed waterbodies with active E. coli impairments and willing utility/agency partners. Geographic and hydrologic diversity strengthens the broader ATP submission (e.g., mountain streams, plains rivers, reservoirs). |
| Submit additional 40 CFR 136.5 applications | Month 21–24 | Each new site submits its own facility-specific ATP application to EPA Region 8, using the Boulder Creek method documentation and comparability report as the technical foundation. Site-specific performance data supplements the base case. |
| Task | Timing | Details |
|---|---|---|
| Briefing to CDPHE on multi-site performance | Month 22–24 | Present aggregated Colorado performance data across all approved sites. Propose CDPHE formally recognize the Lume method for E. coli monitoring on Colorado Reg 93 / 303(d) waterbodies, independent of site-by-site EPA limited-use approvals. |
| CDPHE state guidance or policy adoption | Month 24+ | Work with CDPHE to issue guidance that streamlines future Reg 93 monitoring approvals for the Lume across Colorado. Provides a state-level endorsement that supports the nationwide EPA submission. |
A separate, independent track from the Colorado freshwater work, run in partnership with the American Shore & Beach Preservation Association (ASBPA) at ocean beach sites in other states. This track targets enterococci—the EPA-recommended indicator for marine and coastal recreational water—paired against Enterolert as the reference method. It proceeds on its own timeline and is not a prerequisite for the Colorado limited-use or state-wide approvals; instead, it builds toward a separate coastal limited-use ATP and ultimately a combined nationwide submission under 40 CFR 136.4.
| Task | Timing | Details |
|---|---|---|
| Formalize partnership with ASBPA | Month 15–16 | Define scope, site access, data sharing, and coordination. ASBPA provides access to monitored ocean beach sites and existing enterococci sampling programs. |
| Select 2–4 coastal beach sites | Month 16–17 | Target geographic diversity: Atlantic, Gulf, and/or Pacific coast. Sites should have existing Enterolert monitoring programs to maximize paired data collection efficiency. |
| Deploy Lume sensors at coastal sites | Month 17–18 | Install sensors at selected ocean beach monitoring locations. Adapt deployment protocols for saltwater/tidal conditions (biofouling, salinity interference characterization). |
| Task | Timing | Details |
|---|---|---|
| Collect paired Lume TLF + Enterolert data | Month 18–24 | Continuous Lume readings paired with Enterolert (enterococci) grab samples from existing beach monitoring programs. Target: 6–12 months per site, covering seasonal variation and storm events. |
| Characterize marine-specific interferences | Month 18–24 | Assess salinity, algal fluorescence, and tidal cycle effects on TLF signal in marine water. Document any marine-specific calibration requirements. |
| Enterococci model development | Month 21–24 | Develop and validate ML model for enterococci estimation in marine water. May require marine-specific training data or transfer learning from the freshwater E. coli model. |
| Task | Details |
|---|---|
| Enterococci regression & classification | Same statistical framework as freshwater E. coli analysis (WS 4), but with enterococci thresholds: 35 CFU/100 mL (EPA Beach Action Value), 70 CFU/100 mL (statistical threshold value). Compare against Enterolert and EPA Method 1600. |
| Cross-matrix performance summary | Unified report covering both freshwater (E. coli / Colilert) and marine (enterococci / Enterolert) validation. Demonstrates the Lume’s dual-indicator capability across the full range of recreational water types. |
Full ATP application to EPA’s Office of Science and Technology for incorporation into 40 CFR Part 136 via the Methods Update Rule process. Both tracks feed this submission: Track A provides freshwater E. coli / Colilert data from Boulder Creek and additional Colorado sites; Track B provides marine enterococci / Enterolert data from ASBPA coastal sites. Together they deliver dual-indicator validation covering both freshwater and coastal recreational water types. Note that Track A (freshwater) may reach the nationwide submission threshold before Track B; a freshwater-only nationwide application can be submitted first, with a subsequent coastal amendment.
Nationwide approval requires multi-laboratory validation. The Colorado state-wide expansion (WS 5b) provides multiple independent Colorado sites for freshwater data. ASBPA coastal sites provide independent marine validation. Additional sites outside Colorado further broaden geographic coverage.
| Task | Details |
|---|---|
| Recruit 2–3 additional validation sites | Recreational water sites in different states/EPA regions. Ideal: one inland lake/reservoir, one Great Lakes beach, one different climatic zone. Each site operates the Lume independently following the written method (WS 3). ASBPA coastal sites (WS 6) serve as additional independent validation. |
| Technology transfer & training | Train independent operators on sensor deployment, maintenance, calibration, and data management per the method documentation. Operators must be able to run the method without Virridy involvement. |
| Independent validation data collection | Each additional site collects 6–12 months of paired data (Lume + approved Part 136 method). Follows same protocol as Boulder Creek study. |
| Component | Source |
|---|---|
| Complete method documentation | WS 3 |
| Boulder Creek freshwater validation report (6 sites, 12–18 months) — Track A Phase 1 | WS 2 + WS 4 |
| Colorado state-wide multi-site validation reports — Track A Phase 2 | WS 5b |
| Coastal beach validation report (Enterolert-paired data) — Track B | WS 6 |
| Multi-laboratory validation reports | WS 5b + WS 7.1 |
| Statistical comparability analysis (Appendix H) | WS 4 |
| MDL study report | WS 4.1 |
| ML model documentation & interpretability analysis (E. coli + enterococci) | WS 4.4 + WS 6 |
| QA/QC protocol and field QC results | WS 2 + WS 3 + WS 6 |
| Limited-use approval letter and post-approval performance data | WS 5 |
| Published peer-reviewed literature | Research page |
| Task | Details |
|---|---|
| Submit to EPA OST | Complete application package to the Office of Science and Technology, Office of Water. |
| EPA review | OST reviews the application against validation protocols. May request additional data, clarification, or supplementary analyses. Timeline: 6–18 months (variable). |
| Methods Update Rule | If accepted, EPA proposes including the Lume method in the next Methods Update Rule (MUR). Notice-and-comment rulemaking: Federal Register notice, public comment period (typically 60 days), EPA response to comments, final rule. |
| Post-approval support | Respond to public comments during rulemaking. Provide supplementary data if requested. Support EPA in developing any method-specific guidance documents. |
The City of Boulder Utilities operates 6 recreational water monitoring locations on Boulder Creek. Each site will receive a Lume sensor deployment. Specific site details to be confirmed during the site survey (WS 2, Task 2).
| Site | Description | Expected Characteristics |
|---|---|---|
| Site 1 | Upstream reference / background | Low E. coli baseline. Provides upstream context and natural background TLF levels. Important for establishing the lower bound of the sensor’s operating range. |
| Site 2 | Above WWTP discharge | Mixed urban/natural signal. Moderate E. coli concentrations influenced by urban runoff and upstream land use. |
| Site 3 | Below WWTP discharge | Critical compliance point. Elevated concentrations during wet weather; low during dry weather with effective treatment. Key site for demonstrating performance at permit limit thresholds. |
| Site 4 | Mid-creek compliance point | Mixed signal: WWTP effluent, urban stormwater, natural attenuation. Variable concentrations across seasons. |
| Site 5 | Downstream compliance point | Integrates all upstream sources. Moderate to low concentrations. Important for demonstrating performance during natural dilution/recovery. |
| Site 6 | Lower Boulder Creek / watershed outlet | Full watershed integration. Variable flow. Seasonal irrigation diversions may affect flow and concentration dynamics. |
Note: Site descriptions are preliminary estimates. Exact locations, names, permit associations, and characteristics will be confirmed in coordination with Boulder Utilities during the site survey phase. The diversity of conditions across the 6 sites is a significant strength of this study design—it effectively provides multi-matrix validation within a single watershed.
Detailed protocol for paired data collection at each Boulder Creek site. Designed to generate EPA-quality comparability data.
| Parameter | Setting |
|---|---|
| Measurement interval | 15 minutes (96 readings/day) |
| TLF excitation | 275 nm |
| TLF emission | 340 nm |
| Concurrent parameters | Turbidity (NTU), Temperature (°C) |
| ML model output | E. coli estimate (CFU/100 mL), categorical classification, confidence score |
| Data transmission | Cellular (primary), local SD card (backup) |
| Firmware version | Locked for study duration. Document version at deployment. No firmware updates during active data collection without protocol amendment. |
| ML model version | Locked for study duration. Universal model for initial deployment; site-calibrated model evaluated separately after sufficient paired data collected. |
| Parameter | Requirement |
|---|---|
| Sample collection timing | Record exact time of grab sample to the minute. Match against nearest Lume reading (≤15 min). |
| Sample location | Within 2 meters of the Lume sensor. Same depth and cross-section position. |
| Reference method | IDEXX Colilert-18 per Boulder’s existing SOPs (consistent with approved Part 136 method). |
| Sample handling | Per approved Part 136 method requirements: preservation, holding time (≤6 hours for E. coli), transport temperature. |
| Lab analysis | Boulder’s contract laboratory (to be confirmed). Must be certified for E. coli analysis under Colorado certification program. |
| Duplicate grabs | Every 10th paired sample: collect duplicate grab sample for lab duplicate analysis. Provides reference method precision estimate. |
| Field conditions log | At each grab sample event, record: weather (clear/rain/snow), visual water clarity, recent precipitation (past 24–48 hrs), any visible upstream disturbance, flow estimate (if gauge available). |
| Task | Details |
|---|---|
| Central database | All Lume data and paired Colilert results stored in a single, version-controlled database. Each record: site ID, timestamp, Lume TLF reading, turbidity, temperature, ML estimate, Colilert result, data quality flags, field conditions, operator notes. |
| Data quality flags | Automated flags: sensor fouling (TLF drift > [threshold]), power interruption, out-of-range turbidity, temperature outside operating spec. Manual flags: maintenance window, known interference event, sample handling deviation. |
| Chain of custody | Formal chain of custody for all grab samples. Lume sensor data: automated digital record with sensor serial number, firmware version, model version, and tamper-evident timestamping. |
| Archival | Raw data archived in original format. All transformations, QA/QC flag applications, and model inferences documented and reproducible. Minimum 10-year retention. |
| Risk | Impact | Mitigation | |
|---|---|---|---|
| H | EPA classifies continuous monitoring output as incompatible with current permit compliance framework | Fundamental barrier. Limits value of ATP approval if continuous data can’t replace grab-sample-based compliance reporting. | Address in pre-submission meeting (WS 1). Propose specific compliance reporting frameworks. Engage CDPHE early since they manage Colorado’s recreational water monitoring program. If needed, propose ATP for supplemental monitoring initially, with compliance use as future goal. |
| H | EPA requires multi-lab validation for limited-use approval (not just nationwide) | Delays limited-use timeline by 6–12 months. Additional cost and site recruitment needed. | Clarify in pre-submission meeting. If required, accelerate WS 6.1 (multi-lab recruitment) to run in parallel with Boulder data collection rather than sequentially. |
| H | ML model is not accepted as a determinative step in an analytical method | Fundamental challenge to the method framework. EPA may require a non-ML approach. | Emphasize model interpretability (gradient-boosted trees, not deep learning). Provide SHAP analysis and feature importance. Offer model transparency (open-source option). Propose defined model versioning and revalidation protocol. Engage early with EPA on this specific question. |
| M | Boulder Creek doesn’t span the full E. coli concentration range needed | Incomplete validation. EPA may require data at concentrations not naturally occurring at Boulder Creek sites. | Event-based enhanced sampling targets high concentrations (spring runoff, storms). 6 diverse sites increase range coverage. Existing laboratory Colilert data (n=334) already covers <10 to >100 MPN/100 mL. If field gaps remain, additional sites can fill specific concentration bins. |
| M | Sensor fouling or reliability issues reduce data completeness below 90% | Weakens the dataset. EPA may question operational reliability of the method. | Bi-weekly maintenance schedule. Two spare sensors for rapid swap. Automated fouling detection alerts. Robust mounting design for high-flow resilience. 90% is the target; document mitigation strategies for downtime. |
| M | Colilert reference method variability masks Lume performance | Wide Colilert confidence intervals make statistical comparability harder to demonstrate, even if the Lume is performing well. | Document Colilert method variability (literature and from duplicate grab samples). Frame Lume precision as comparable to or better than the reference method’s own precision. The research page notes 14% RPD for TLF duplicates vs. ≥26% for culture-based duplicates. |
| M | City of Boulder organizational changes affect partnership | Loss of champion, budget changes, or priority shifts could delay or derail the study. | Formalize MOU early. Ensure buy-in at management level, not just individual staff. Minimize Boulder’s operational burden (Virridy handles sensor deployment/maintenance; Boulder continues existing Colilert program unchanged). |
| L | EPA rulemaking timeline extends beyond 3–5 years for nationwide approval | Slow path to national market. Does not affect limited-use approvals. | This is expected. Limited-use approval (WS 5) provides regulatory value in the interim. ASBPA coastal validation (WS 6) builds evidence base in parallel. Pursue additional limited-use approvals at other sites to build market while waiting for nationwide rule. |
| L | Seasonal access constraints (ice, flooding) cause data gaps | Missing winter or spring runoff data weakens seasonal coverage. | Design sensor housing for winter deployment (heating element or insulated housing). For spring flooding: secure mounting above flood stage or plan for temporary sensor retrieval with rapid redeployment. |