TetraSoft Atlas

How CO Reporting Quality Works

Methodology reference for the anomaly detection shown in this panel

Data Source

All figures come from Colorado's Oil and Natural Gas Atmospheric Emissions Inventory Report (ONGAEIR), submitted annually to CDPHE/COGCC. Facilities report CH₄, CO, NOx, VOC, and CO₂e across 18 equipment categories. TetraSoft ingests each year's raw submissions as they become available and assigns each facility a stable site identifier, so the same physical location can be tracked across years even when its facility ID changes.

What the Analysis Does

For each facility and reporting year, CH₄ values are measured against the full population of facilities in the same equipment category, year, and site type. The score used for comparison is a modified z-score.

A standard z-score measures distance from the mean in units of standard deviation. That is sensitive to extreme values: a few very large reporters can inflate the baseline and make everyone else look normal. The modified version replaces the mean and standard deviation with the median and the median absolute deviation (MAD). The MAD is the typical gap between individual values and the group median. Because it is calculated from the middle of the distribution, a handful of outliers cannot shift the scale. A score of 0 means the record sits exactly at the group median. A score of 10 means it is ten MAD-units above it.

Not all equipment categories get the same treatment. Categories with fundamentally different operating patterns use different detection logic, described in the tier section below.

Warning vs. Critical

Warning: score above 3.5 and CH₄ above the category minimum floor. The record is statistically unusual. Worth a second look, but not far enough out to call definitively wrong.

Critical: score above 7.0 and CH₄ above the category critical floor (varies by tier, described below). At that level the record is well outside its peer group in both statistical and absolute terms. Critical records from Tier 1 categories are removed from the Adjusted CH₄ trend.

Equipment Category Tiers

Not all 18 ONGAEIR equipment categories behave the same way statistically. Treating them identically would produce misleading flags. The analysis uses three tiers.

Tier 1: Standard peer comparison

Tanks, Separators, Stationary Engines or Turbines, NR Internal Combustion Engines, Pneumatic Controllers, Amine Units, Combustion Control Devices, and all categories not listed in Tiers 2 or 3.

Each facility is scored against same-category, same-year peers in the same site type. Critical threshold: z-score above 7.0 and CH₄ above 5 times the category's 99th percentile. These are the categories where volumes are large enough that a genuine outlier represents a real data quality question. Only Tier 1 critical records are subtracted from the Adjusted trend line.

Tier 2: Peer comparison with absolute floor

Heaters or Boilers (5 MT floor), Loadout (5 MT), Well Bradenhead (5 MT), Pneumatic Pumps (5 MT), Dehys (5 MT), Flowback or Completions (10 MT), Pipeline (10 MT), Drill Mud (2 MT).

Same z-score logic as Tier 1, with one addition: critical severity also requires CH₄ above a fixed floor. These categories have very tight distributions. A heater reporting 2 MT can score a z-score above 40 simply because most heaters report under 0.1 MT. That statistical signal is real, but 2 MT is not the kind of emission that warrants an audit review. Below the floor, severity stays at Warning regardless of z-score. Tier 2 critical flags appear in the table but are not subtracted from the Adjusted trend.

Tier 3: Site-historical spike detection

Fugitives, Venting or Blowdowns, II.G Well Maintenance.

Peer comparison does not apply here. These categories exist to capture non-routine operational events: a fugitive emission, a series of planned blowdowns, maintenance activity. A midstream operator running many blowdowns in a year is supposed to report high values. Scoring them against a peer average would flag accurate reporting as anomalous.

Each site is compared against its own history instead. A year is flagged when the site's total for that category exceeds three times its own prior peak. Sites with no prior history are flagged at Warning if the value exceeds 10 MT. The goal is to catch something genuinely new at a specific site, not to penalize consistent high-volume operators.

Tier 3 flags never reduce the Adjusted trend. These are real emissions.

The Adjusted CH₄ Trend

The trend chart shows two lines. Raw is the total reported CH₄ across all CO ONGAEIR facilities for that year. Adjusted is the same total after removing critical Tier 1 records (SEVERE_OUTLIER flags in Tanks, Separators, Stationary Engines, and similar high-volume categories).

The gap between the lines shows how much of the statewide number comes from a small number of extreme records. A wide gap does not mean those records are wrong. Some are genuine large emission events. The Adjusted line answers one specific question: what does the trend look like if we set the statistical extremes aside?

Peer Groups and Site Types

Colorado ONGAEIR facilities report under four site types: Production (wellpads, the largest group at around 19,000 sites), Midstream (compressor stations, processing plants), Pre-Production (drilling sites), and UIC (injection wells). A wellpad and a compressor station handle very different operations, and treating them as one population would produce comparisons with no useful signal. Each site is scored against others in the same site type, equipment category, and year. When fewer than ten facilities exist for a given combination, the comparison expands to the full statewide population.

A note on Colorado's emissions reporting

Colorado ONGAEIR is among the most detailed oil and gas emissions reporting frameworks in the country. CDPHE/COGCC have built a dataset that other states have looked to as a model, and that reflects years of sustained effort from both the agency and the operator community.

TetraSoft's founders have worked with this dataset since its early years. That history is what makes this analysis possible, and it is also why we read the flags carefully. The patterns surfaced here are statistical. They are starting points for review, not findings. A flagged record may turn out to be a legitimate large-emission event reported exactly as required. Others may reflect reporting conventions or edge cases in submitting software. We flag them so you can look, not so you can conclude.

This tool is for informational use only. Nothing here constitutes a regulatory finding, compliance determination, or audit result. Verify any flagged record against the original ONGAEIR submission before drawing conclusions.