Emissions Forecasting for Australian NGER and AASB...

The pattern in Australian reporting cycles is remarkably consistent. August: "we don't have full Scope 3 yet." September: panic. Late October: the auditor lands, opens the working file, and the annual number moves five percent in either direction while the sustainability manager tries to explain what happened. The NGER report gets filed on 31 October. Then the AASB S2 disclosure gets bolted onto the financial report a few months later. And by January, the CFO is asking a question that should have been answered in April: "how did we end up here?"

That question is the reason we built emissions forecasting into Carbonly. Not as a scenario analysis exercise under AASB S2 paragraph 22 (that is climate-scenario resilience testing, a different beast). This is a projection of the actual reported number based on the actual ledger, refreshed monthly, expressed as a range with confidence bands.

If you know your year-end number in Q2, you can act on it. If you find out in October, you file it.

What the forecast actually does

The emissions forecasting service runs off the per-facility monthly ledger. Every invoice, docket, and meter read that gets extracted lands on a facility, in a month, with a scope and a category. The forecast takes the partial-year actuals, layers historical seasonality per facility on top, adjusts for known operational changes, and projects the year-end total.

Historical seasonality matters more than most reporters realise. A remote WA project site burns 3x more diesel in Q3 than Q1 because that is when the earthworks phase runs. A Melbourne CBD office spikes electricity in July and August because HVAC runs harder. Refrigerant top-ups follow a service cycle, not a straight line. If you just annualise the year-to-date and multiply by 12/n, you get a number that is confidently wrong.

The projection is not a single number. It comes out as a range with an upper and lower confidence band. The band widens when data is missing (last month's fuel dockets have not landed yet, one supplier is behind on invoicing) and tightens when the ledger is complete and stable. The confidence band is the transparency layer. The CFO does not just see "12,400 tCO2e projected." They see "10,800 to 13,900 tCO2e, mid-point 12,400, confidence 68% based on 8 months of actuals and 4 months of projection."

That is the number to disclose to the board. Not a false-precision point estimate.

Why this matters for Safeguard Mechanism reporters

If you operate a facility covered by the Safeguard Mechanism, you have a declining baseline. The baseline drops 4.9% per year through to 2030. Exceed it and you either surrender ACCUs or SMCs, or you pay the shortfall at the Safeguard penalty rate. The current civil penalty structure sits at $330 per tonne above the limit, plus $33,000 per day while in excess, with criminal exposure for dishonest reporting.

Finding out in October that a Safeguard facility is 8,000 tonnes over baseline is a $2.64 million problem before you buy a single ACCU to close the gap. Finding out in March, when the projection first crosses the trajectory line, is a management problem with time to solve it.

The Safeguard Trajectory Check runs continuously against the facility-specific baseline. It compares the current forecast mid-point plus upper confidence band against the declining baseline for the current reporting year. When the upper band starts overlapping the baseline, the facility gets flagged. When the mid-point crosses, it gets escalated.

The Trajectory Check does not tell you what to do about it. That is the Carbon Planning module's job. But it makes sure the conversation happens while the levers are still available. Reduce diesel through fleet re-routing, bring forward a planned gas-to-electric switchover, buy ACCUs while the market is quiet rather than in an October scramble. All of those decisions need Q2 visibility, not Q4.

Why this matters for AASB S2 disclosure

AASB S2 paragraph 21 requires prior-year comparatives. Which sounds mundane until you realise most Group 2 reporters are staring at a first-year AASB S2 disclosure where the current-year figure is still moving in September while the auditor is scheduled for November.

Forecasting closes that gap. The projection sits alongside the prior-year snapshot in the Auditor Workspace, so the CFO can see what the disclosure conversation is going to look like months before the auditor arrives. If the current-year forecast is up 15% versus prior year, the CFO has two options: dig into the drivers now and fix them, or prepare the narrative for why it moved.

Either option beats being surprised.

The disclosure also has to sit inside a coherent transition plan under AASB S2 paragraph 14. If your climate transition plan says emissions are on a trajectory toward net zero by 2050, and your Q2 forecast shows a 12% year-on-year increase, that is a governance problem the board wants to know about before the sustainability report hits the ASX. Not after.

What the forecast is built from

Three inputs. First, the per-facility monthly ledger. Every extracted document, every activity record, every meter reading, tagged to a facility and a month. That is the actuals baseline.

Second, per-supplier and per-facility seasonality. Twelve months of historical patterns per site, calibrated on prior reporting years. A construction facility with three completed years of Carbonly data has a much tighter forecast than one loaded from scratch six months ago. The service is honest about this: confidence is lower when history is thin.

Third, factor version pinning. This is a detail that trips a lot of reporters. The NGA Factors workbook is updated annually, and the 2025 edition changed several state grid factors materially. If the forecasting engine silently applies the new factor to prior-period activity, the year-on-year comparison stops meaning anything. Factor version gets pinned to the reporting year. Change in factor edition is treated as a separate line item, not baked invisibly into the trajectory.

Same principle for AR5 versus AR6 GWPs. NGER submissions still use AR5 for the current cycle. AASB S2 requires AR6. Both reports run off the same ledger, but the GWP is toggled at render time. The forecast does the same. You get an NGER-basis forecast for the Clean Energy Regulator submission and an AASB S2-basis forecast for the disclosure. Same underlying activity data. Different presentation.

The Variance Explanation Agent

Forecasts move. That is fine. What kills the board conversation is when the forecast moves and nobody can explain why.

The Variance Explanation Agent watches every material shift in the forecast between refreshes and writes a plain-English explanation of the driver. Not a bar chart. A written explanation.

Example of what lands in the sustainability lead's inbox after a monthly refresh: "Diesel forecast up 12% versus last month. Driver: Perth site added a night shift in May, extraction pattern suggests fuel consumption will run 40% above baseline for the remainder of the reporting year. Refrigerant forecast down 8%. Driver: R410a service dockets from the Melbourne DC are 40% behind pace against the historical service cycle; two options, either service is genuinely delayed (adjust forecast down) or invoices are late (widen confidence band). Recommend confirming with facilities."

That is board-briefing quality. The CFO can walk into the risk committee with the number, the direction of travel, and the specific business decision that drove it. Not a spreadsheet.

The Proactive Report Agent

Quarterly, the Proactive Report Agent auto-generates draft NGER and AASB S2 reports based on the current forecast and emails them to the sustainability lead and CFO. Not to file. To review.

The draft contains the current forecast, the confidence range, the top variance drivers from the Variance Explanation Agent, the anomaly list from the Data Health Agent, and a methodology snapshot pinning the factor versions and calculation basis. The reader can see, in mid-cycle, what the year-end submission is trending toward.

Most reporters do not need to file a quarterly report. They still need to think like one. The Proactive Report Agent forces that discipline without asking anyone to do the work.

Where the MACC curve fits

The Carbon Planning module maintains a marginal abatement cost curve. That is a ranked list of every emission-reduction opportunity in the plan, priced by dollars per tonne abated. When the forecast projects a Safeguard breach or an AASB S2 trajectory concern, the MACC becomes the answer to "what do we do about it?"

Rooftop solar at the Melbourne DC might abate 800 tonnes at $12 per tonne. A fleet electrification pilot might abate 400 tonnes at $180 per tonne. An HVAC upgrade might abate 200 tonnes at negative $30 per tonne (net saving from energy). The MACC ranks them, and the forecast tells you how many tonnes need closing.

That combination (forecast tells you the gap, MACC tells you the cost to close it) lands in the board pack with a real emission and cost delta. Not a slide with a green arrow.

Reading the confidence band

We keep coming back to the confidence band because most tools do not have one, and reporters get burned by the false precision. Here is how to read it.

Narrow band, high confidence: your ledger is complete, historical seasonality is well-calibrated, no operational changes flagged. Trust the mid-point.

Wide band, medium confidence: activity data is missing for one or two months, or a Scope 3 category is running behind on supplier engagement. Trust the range, not the mid-point. Do not commit to a point estimate in the board pack.

Very wide band, low confidence: you are asking for a forecast on a facility with three months of history. The number will move materially as more data lands. Use it as a directional signal, not a disclosure.

The transparency is deliberate. The CFO needs to know what the auditor will ask. If the auditor sees a tight forecast band six months out, they will interrogate the assumptions. If they see an honest wide band with a clear explanation of what would tighten it, they engage very differently.

What forecasting is NOT

One thing worth stating plainly. Emissions forecasting is not scenario analysis under AASB S2 paragraph 22. Scenario analysis is a climate-resilience exercise, running the business through 1.5-degree and 3-degree scenarios to test transition and physical risk exposure. That is a strategic exercise, usually run annually, with qualitative and quantitative outputs.

Forecasting is a projection of the actual reported number based on the actual ledger. It answers the operational question ("what is our Q4 going to look like?") not the strategic one ("how would we perform under a disorderly transition scenario?"). Both belong in an AASB S2 report. They are not the same thing.

Reporters conflate them because both involve looking forward. The distinction matters for how you use the output and how you present it to the board.

Where the MCP server comes in

The forecasting service is exposed through the Carbonly MCP server. Once ChatGPT or Claude is connected to your Carbonly workspace, the CFO or sustainability lead can ask ask_copilot directly: "what does the annual forecast look like for the Perth site?" The answer lands with the range, the confidence, and the top three drivers. No dashboard navigation. No filter juggling.

For the CFO who spends the reporting cycle answering board questions on a laptop between meetings, this is the difference between "let me pull the number and get back to you" and "the mid-point is 8,400 tonnes, upper band 9,200, and the driver is the night shift that started at Perth in May."

Practical starting point

If you have never built a forecast against your emissions ledger, start with the highest-emission facility. Not the whole portfolio. One site.

Load six months of ledger data (invoices, dockets, meter reads), whatever you have in the 8 formats the document engine accepts. Confirm the extractions with the 5-tier material matching sitting on top. Let the seasonality calibration run. Then ask ask_copilot: "what does the annual forecast look like for [site name]?"

The answer will land with a confidence band. If the band is narrow, you have found the format. Roll it out across the portfolio. If the band is wide, the tool is telling you exactly which months and which categories need more data before the forecast becomes decision-quality. That is not a limitation. That is the tool doing its job.

We are honest that this is the frontier of the product. Forecasts on well-instrumented facilities with 24+ months of history are tight and boring. Forecasts on new sites or on Scope 3 categories with poor supplier data are wide and interesting. Either way, the CFO knows what they are looking at eight weeks before the auditor lands. That was the goal.

Carbonly is priced per project (Small, Medium, Large, Enterprise) with a $100 per month workspace minimum. Consultants running an ASRS or NGER engagement can spin up a client workspace, load six months of ledger, and have a defensible forecast in front of the CFO within a fortnight. Reach out at hello@carbonly.ai if you want to run this against your Q4.

Emissions Forecasting for Australian NGER and AASB S2 Reporters: Knowing Your Q4 Number in Q2