Definition
The PEMS Drift Detection Index (PDDI) measures the percentage divergence between a PEMS-predicted emissions value and a reference-measured emissions value at the same operating condition and the same time.
Plain language first.
The Measurement Gap
Marco is the environmental compliance manager at a 180,000 bpd refinery in the Gulf Coast. It is October. His FCCU PEMS has been reporting NOx in the high 30s ppm all year: well under the 60 ppm Title V permit limit. The last RATA was nine months ago, and the model passed within the required ±20% relative accuracy at that test. Quarterly Cylinder Gas Audits since then have all been green. The PEMS dashboard says everything is fine.
The annual RATA is scheduled for next week. Marco's CEMS contractor sets up the reference analyzers, runs the parallel comparison over three nine-run sequences, and hands him the data: actual stack NOx averaged 58 ppm during the test, while PEMS predicted 39 ppm. The 19 ppm gap is a 33% relative error. The model has failed RATA. Worse, the gap has almost certainly been there for months: the catalyst on the FCC SCR has been aging through the year, NH3 injection has been creeping up to compensate, and the PEMS model was trained on a younger catalyst. Marco does not know exactly when actual emissions first exceeded 60 ppm. He has to report a permit exceedance and he cannot bound how long it lasted.
This is not a story about a broken sensor. It is a story about a model that drifted away from reality and the operational system around it that did not notice. And the legacy metrics in use at the refinery cannot see it.
Methane emissions intensity is the wrong scale. It rolls up facility-level methane against production volume: useful for ESG reporting and corporate disclosure, useless for spotting drift on a single emission source. Flaring intensity has the same scale problem and a different combustion source: the FCCU stack is not a flare. The most direct compliance metric in use today is "PEMS uptime": what percentage of the operating hours the model produced a number rather than going offline. PEMS uptime can be 99.9% while the number it produced was wrong by 33%. The metric reports availability, not accuracy.
The quarterly CGA is supposed to catch this, and sometimes does. But a CGA tests the analyzer (or the model, depending on the program) against a known cylinder gas at a single concentration point. It does not detect a drift that affects the model's mapping of process inputs to emissions: only a drift that affects the calibration constant. A PEMS model that increasingly mispredicts emissions during high-load operation but predicts correctly at the low-load conditions CGAs typically use will pass every CGA in the calendar year and still fail the next RATA.
The measurement gap is precise. PEMS is a regulatory instrument that depends on a model staying calibrated to physical reality. The model can drift on a timescale faster than the RATA cycle, in a direction that compliance metrics cannot detect, and the only way to know is to compare to a reference measurement on a cadence shorter than annual. The question PDDI proposes to answer is the one Marco could not ask in March: are the PEMS-reported emissions still accurate for regulatory compliance between RATA cycles?
Formula
The PEMS Drift Detection Index (PDDI) measures the percentage divergence between a PEMS-predicted emissions value and a reference-measured emissions value at the same operating condition and the same time.
Plain language first. Take the PEMS number for a pollutant: NOx, SOx, CO, whatever the source emits. Take a reference measurement of the same pollutant at the same moment. The absolute percentage difference is PDDI. Track its rate of change over time as PDDI_trend. The reference can be CEMS (if you have one running alongside PEMS), a RATA result, or a portable analyzer reading. The point of the metric is simple: how far has the model drifted from physical reality, and how fast.
Notation:
PDDI = |PEMS_predicted − Reference_measured| ÷ Reference_measured × 100
PDDI_trend = d(PDDI) ÷ dt
Where:
- PEMS_predicted = the PEMS model output at the time of the reference measurement (units appropriate to the pollutant, typically ppm or lb/MMBtu)
- Reference_measured = the value from the reference instrument or compliance test at the same time and operating condition
- t = time since the last calibration event (months)
The absolute-value bars in the numerator are intentional. Drift in either direction matters operationally: over-prediction creates an unnecessary regulatory burden, under-prediction creates undetected exceedance. Both move you toward a RATA failure and both should trigger investigation. The denominator is the reference measurement because that is the physical truth the model is meant to approximate.
Edge cases worth knowing.
Unreliable reference measurement is the first concern. If the portable analyzer is the reference and it has not been audited against a certified gas in the last quarter, PDDI is measuring two sources of error and you cannot separate them. The reference must be more trustworthy than the PEMS, or the metric is noise.
Different pollutants drift differently. The PEMS model for NOx and the PEMS model for SOx on the same stack are typically independent: they use different input variables and have different sensitivities to process changes. Compute PDDI per pollutant per source. Aggregating across pollutants washes out the signal you need.
Some drift is process-driven, not model-driven. Catalyst deactivation in an SCR, soot buildup on a heater, fuel composition changes. These all shift the actual emissions independent of the model. PDDI rising because the catalyst aged is still PDDI rising (the regulatory exposure is the same) but the response is different. Drift caused by process change requires a process correction (catalyst replacement, fuel adjustment) followed by recalibration. Drift caused by model staleness requires only recalibration. Investigate the cause before acting on the metric.
Proposed formula, pending validation pilots.