Measuring stack emissions directly with an analyzer is accurate but expensive to install and maintain, so operators have long looked for a way to know their emissions without a probe in every stack. A predictive emissions monitoring system, or PEMS, is that alternative: instead of sampling the gas, it calculates emissions from the process conditions using a model. This guide explains how a PEMS predicts emissions from engine and process parameters, what inputs it needs from the control system, and how it is validated and audited so it can serve as an approved monitoring method in place of a hardware analyzer.
Predictive Emissions Monitoring System (PEMS) in one line: A predictive emissions monitoring system, or PEMS, estimates a source's emissions by feeding process parameters such as fuel flow, load, temperatures, and pressures into an empirical or statistical model that has been correlated to actual measured emissions, rather than sampling the exhaust gas directly. It is used as a lower-cost alternative to a hardware continuous emissions monitor, most commonly for nitrogen oxides from engines and turbines. To be accepted for compliance, a PEMS must be validated against a reference method and periodically audited, much like a CEMS.
A PEMS rests on a simple observation: the emissions from a combustion source are not random but are governed by how the source is operating. For an engine or turbine, the rate of nitrogen oxide formation depends strongly on things like load, fuel flow, combustion temperatures, manifold pressures, and ambient conditions, all of which are already measured or measurable. If the relationship between those operating parameters and the resulting emissions can be captured in a model, then reading the parameters lets the system predict the emissions without ever sampling the gas.
Building that model is the core of a PEMS. During development, the source is run across its operating range while a reference measurement records actual emissions, and the paired data is used to fit a model, which may be a physics-based relationship, an empirical statistical correlation, or a trained data-driven model, that reproduces the measured emissions from the input parameters. Once fitted and validated, the model runs continuously on the live parameter inputs, outputting a predicted emission rate that stands in for a direct measurement.
The appeal is largely economic and practical. A hardware CEMS needs an analyzer, a sample conditioning system, and ongoing calibration gases and maintenance, all of which add cost and can themselves be a source of downtime. A PEMS needs no probe in the stack and no sample handling, so it avoids much of that burden, and because it derives its output from parameters the control system is already tracking, it can be highly available as long as those inputs are healthy.
Because a PEMS predicts from operating conditions, its accuracy depends entirely on getting good, timely values for the parameters its model uses. Those inputs come from the same instrumentation and control system that runs the equipment: fuel flow measurements, engine or turbine load, exhaust and combustion temperatures, intake and manifold pressures, and often ambient temperature and humidity. The exact set depends on the model and the source, but the common thread is that the PEMS is a consumer of process data rather than a separate sensing system.
This dependence cuts both ways. On one hand it means a PEMS integrates naturally with existing controls and adds little new hardware. On the other, it means the quality of the emissions prediction is only as good as the quality of the input signals, so a drifting temperature sensor or a failed flow meter does not just affect operations, it corrupts the emissions estimate. A well-designed PEMS therefore includes checks on its inputs, so that when a required parameter is missing or out of range the system flags the condition rather than silently producing a bad prediction.
The system also needs to know when the source is in a state its model was not built to cover, such as startup, shutdown, or an unusual operating mode, because a model fitted to normal operation may not predict those transients well. Handling these conditions correctly, whether by substituting conservative values or by flagging the period, is part of what makes a PEMS defensible, and it underscores that a PEMS is not just a model but a system wrapped around a model that manages the model's limits.
Regulators do not accept a PEMS on the strength of its design; they require it to prove itself against reality, and the proof is the same relative accuracy test audit used for a hardware CEMS. A reference method measures actual emissions while the PEMS predicts, and the two are compared to confirm the model agrees with the truth closely enough. Just as with a CEMS, the PEMS must also be re-audited periodically to confirm it has not drifted out of agreement as the equipment ages or conditions change.
Ongoing quality assurance for a PEMS focuses on the inputs and the model together. Because the prediction depends on sensor signals, the sensors themselves must remain calibrated and healthy, and the system must track its own data availability so that periods when it could not predict validly are accounted for. A PEMS that meets its accuracy criteria on audit but frequently loses its input signals is no more compliant than a CEMS that is accurate but often offline.
This is where connecting a PEMS to a broader monitoring platform helps. A cloud SCADA platform such as Merobix already gathers the fuel flow, load, temperature, and pressure signals a PEMS relies on, and it can retain the predicted emissions alongside them, so the audit trail, the input history, and the emissions record all sit together. For an operator using PEMS across multiple engines or sites, having every unit's inputs, predictions, and audit status in one place makes it far easier to keep each PEMS in the accurate, high-availability condition its approval as a monitoring method requires.
A CEMS measures emissions directly by sampling the stack gas with an analyzer, while a PEMS predicts emissions from process parameters using a model, without sampling the gas. A PEMS avoids the analyzer, sample handling, and calibration gases a CEMS needs, which lowers cost and maintenance, but its accuracy depends on the quality of the input signals and the validity of the model.
A PEMS uses process parameters that govern emissions, typically fuel flow, engine or turbine load, combustion and exhaust temperatures, manifold and intake pressures, and often ambient conditions. These come from the same instrumentation and control system that runs the equipment, so a PEMS is a consumer of process data rather than a separate sensing system. Its prediction is only as good as those input signals.
A PEMS is validated the same way a hardware CEMS is, through a relative accuracy test audit that compares its predictions against a certified reference method measuring actual emissions. If the two agree within the allowed limit, the PEMS is accepted, and it must be re-audited periodically to confirm it has not drifted. It also has to track its own data availability and keep its input sensors calibrated.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.