KB Analytics
Proprietary intelligence for Australian housing, farming and energy.

Sector outlook · Sustainable energy · Australia · Plate Nº III

The grid runs on weather. Every decision on it starts with a forecast.

Sun and wind are cheap and cannot be ordered. As they replace coal, the value goes to whoever can see the next five minutes, the next evening and the next ten years. With a forecast given as a range, an owner can site, run and finance clean power where it is most likely to pay.

Wind turbines, a solar farm and a transmission line under a published forecast and a forecast rangeArchitectural and mechanical sketch of three wind turbines on a ridge, solar panels in the foreground and a transmission tower. Construction lines remain on the sheet. A hairline traces the published forecast; a yellow curve traces the forecast: solid where output has been measured, a widening band where it is still to come.MEASURED OUTPUTPUBLISHED FORECASTFORECAST RANGEKB ANALYTICSSECTOR OUTLOOK · AUSTRALIAA DAY OF SUN AND WIND — OUTPUT AS A RANGESUSTAINABLE ENERGYPLATE Nº IIINOVEMBER 2026 · NOT TO SCALE
Wind, solar and the line that carries them. The hairline is the published forecast. The yellow curve is a forecast as a range: solid where measured, widening where still to come.

01The sector in numbers

Records at both ends: the most renewable power, and the least demand for the rest.

Most figures here are for the National Electricity Market, the grid that serves the eastern and southern states. The market operator's plan says its least-cost path would reach the national 82% target by 2030. The same plan says only half of the new wind that path needs is in the pipeline [3]. Both are true at once. That is the forecasting problem in one line.

The latest quarter shows the change. Grid-scale batteries supplied 8% of the evening peak, past gas for the first time [2]. The average wholesale price fell to $63 a megawatt-hour, the lowest September quarter since 2021 [2]. And 22 data centres totalling 10.1 GW were working through the connection queue [4].

  • 82%renewable electricity on the grid by 2030 is the national target.Department of Climate Change, Energy, the Environment and Water [1]
  • 44.3%of generation in the eastern grid came from renewables in July to September 2026, a record for that quarter.Australian Energy Market Operator [2]
  • 38 GWof new grid-scale wind and solar is needed by 2030 on the operator's plan. Only 9 of the 18 GW of wind is in the pipeline.Australian Energy Market Operator [3]
  • 4 yearsis the average time a wind or solar project has taken from connection application to full output.Australian Energy Market Operator [3]
  • 9,422 MWwas the lowest demand ever recorded on the eastern grid, at midday on Sunday 20 September 2026.Australian Energy Market Operator [4]
  • 14.4%of five-minute trading intervals had a zero or negative price in the September quarter 2026.Australian Energy Market Operator [4]
  • 6.7%of available solar output was held back by the network in that quarter, up 3.6 points in a year.Australian Energy Market Operator [4]
  • 8%of evening peak generation came from grid-scale batteries that quarter, ahead of gas at 5.1% for the first time.Australian Energy Market Operator [2]
  • $63/MWhwas the average wholesale price in the September quarter 2026, the lowest for that quarter since 2021.Australian Energy Market Operator [2]
  • 546,789home batteries were installed across Australia by 30 September 2026.Clean Energy Regulator [5]
  • 10 GWof grid-scale batteries were in service or commissioning on the eastern grid by the end of that quarter.Australian Energy Market Operator [4]
  • 10.1 GWof data centre load, across 22 centres, was working through the transmission connection queue.Australian Energy Market Operator [4]

02The questions

Six forecasts that decide whether clean power pays.

This sector has the best public data of the three, and real public trials. A federal agency funded 11 projects with $9.41 million to test whether wind and solar farms could forecast their own output five minutes ahead [6]. The lessons are sober: each new site needs its own model, a full year of data is preferred, and cloud inside a five-minute window is hard [7]. Later work combined fourteen real-time models across five solar farms [9], used sky cameras to cut error in high-variability weather [10], and produced nowcasts for all 25 semi- and non-scheduled solar and wind farms in South Australia [11].

We found no public Australian study with results on wear in grid batteries or failures in wind turbines. For each question below: the gap in what is published, the test, the public data, and the pass mark.

  1. How much will this farm produce in the next five minutes, and the next two days?

    Farms on the eastern grid may give the market operator their own five-minute forecast, and a poor one costs them money. Cloud edges and wind ramps are where forecasts break.

    The gap
    Federal trials found that each new farm needs its own model, and that cloud inside a five-minute window is hard [7]. Whether a model beats the no-change forecast at a given farm, in its cloudiest intervals, cannot be known in advance.
    The test
    For a named solar farm, forecast five-minute output from satellite-derived sunlight and the last hour of output. Score it over a 12-month hold-out against a no-change forecast, including the cloudiest tenth of intervals.
    Public data
    AEMO market data (NEMWEB), OpenElectricity, Bureau of Meteorology solar data.
    Pass mark
    Lower average error than the no-change forecast. If the gain vanishes in the cloudiest intervals, the model has failed where it matters.

  2. How low will midday demand fall, and how high will the evening climb?

    Rooftop solar pushed the eastern grid to a record low in September 2026, and South Australia briefly drew less than nothing from the grid [4]. Home batteries are arriving by the hundred thousand.

    The gap
    We found no published result on whether monthly postcode battery counts improve forecasts of the lowest-demand half-hours.
    The test
    Add monthly postcode battery counts to a regional demand model. Compare next-season forecasts of the lowest 1% of half-hours with a model that uses rooftop solar capacity and weather alone.
    Public data
    AEMO market data (NEMWEB), APVI solar map, Clean Energy Regulator postcode data, AEMO Integrated System Plan inputs.
    Pass mark
    Lower error on the lowest-demand half-hours of a held-back season. If counts add nothing, battery behaviour is not yet predictable from counts, and we say so.

  3. How often will the price be negative next year?

    Negative-price intervals fell 4.5 points in a year as batteries began charging through the middle of the day, and batteries now set the price in almost half of intervals [2]. The market of two years ago is a weak guide to the next one.

    The gap
    How far a model trained on the old market over-predicts negative prices, and whether installed battery capacity corrects it.
    The test
    Train only on data to June 2025. Measure the over-prediction of negative-price intervals for the September quarter 2026 in each mainland region. Then add installed battery capacity and measure again.
    Public data
    AEMO market data (NEMWEB), OpenElectricity, Australian Energy Regulator wholesale charts, AEMO Quarterly Energy Dynamics.
    Pass mark
    Forecast share of negative intervals against the actual share, region by region.

  4. When should a battery charge and discharge, and how fast will it wear?

    The daily pattern, charge through the middle of the day and discharge into the evening, is so regular that a simple rule already does well. The margin over that rule is the honest test.

    The gap
    The margin, per full cycle, of dispatch driven by a price forecast given as a range. Degradation data for Australian grid batteries is held by owners and is not public.
    The test
    Over a 12-month walk-forward for a two-hour battery in a named region, drive dispatch with a price forecast given as a range. Score energy-arbitrage revenue per full cycle against the fixed charge-at-midday, discharge-at-evening rule.
    Public data
    AEMO market data (NEMWEB), OpenElectricity, AEMO Quarterly Energy Dynamics, Bureau of Meteorology solar data.
    Pass mark
    More revenue per cycle than the fixed rule on held-back months. Where the margin is within noise, the forecast adds nothing, and we say so.

  5. Will a new project lose income to a crowded network?

    Solar held back by the network is concentrated in particular areas, and each generator's loss factor is reset every year as neighbours connect. Both depend on what other people build.

    The gap
    We found no published result on whether new solar committed nearby predicts the direction of next year's loss-factor change.
    The test
    For existing solar farms, test whether new solar committed in the same network area in one year predicts the direction of the farm's loss-factor change the next, across at least five past years.
    Public data
    AEMO market data (NEMWEB), AEMO Quarterly Energy Dynamics, AEMO Integrated System Plan inputs. A statistical model can rank sites by risk. It does not replace a network power-flow study.
    Pass mark
    More loss-factor moves called in the right direction than chance, on held-back years.

  6. Which assets are quietly under-performing?

    University researchers reported a council solar system that under-performed for five months before anyone noticed [8]. Component data such as inverter logs is held by owners and is not public.

    The gap
    Whether public five-minute output data is fine enough to catch a 5% shortfall at farm level. It may be too coarse; that is part of the test.
    The test
    From public five-minute farm output and satellite sunlight alone, flag periods of at least 5% unexplained shortfall at a named solar farm. Check the flags against recorded outages.
    Public data
    AEMO market data (NEMWEB), OpenElectricity, Bureau of Meteorology solar data.
    Pass mark
    Flags match outage records more often than chance. If public data cannot separate a fault from a network limit, that is worth knowing before anything is promised.

03What we built

We start with the published number. We return a range.

Our own method takes the number already in use and returns a range for this project, this paddock, this farm. It is scored on records the model has never seen. If it does not beat the number in use, we say so.

For energy, the method covers output over the next five minutes and the next two days, the midday trough and evening climb, negative prices, battery dispatch and wear, crowding on the network and quiet under-performance. Each is a range for a site and an interval, set against the published forecast and the simple daily rule.

  1. Lock the ruleThe number in use today: a statewide average, a map cell, last year's rate, a no-change forecast.
  2. Build the modelOn Australian records, for the decision that has to be made now.
  3. Give a rangeA range you can check. Not a single figure.
  4. Score itOn records it has never seen, against the rule. Misses are named.
  5. Keep scoringAs each quarter, season or five-minute interval arrives.

What stays private

The models, the features, the scoring, and the way a miss shapes the next range. None of it is published. Client data stays with the client. This site shows the questions and the tests, not the working.

04Sources

Every figure on this page comes from one of these publications. Official statistics are revised, so check the source before relying on a number.

  1. Department of Climate Change, Energy, the Environment and Water. Capacity Investment Scheme. 24 June 2026. https://www.dcceew.gov.au/energy/renewable/capacity-investment-scheme
  2. Australian Energy Market Operator. Quarterly Energy Dynamics Q3 2026 (media release). 23 October 2026. https://www.aemo.com.au/newsroom/media-release/qed-q3-2026
  3. Australian Energy Market Operator. 2026 Integrated System Plan. 25 June 2026. https://www.aemo.com.au/energy-systems/major-publications/integrated-system-plan-isp/2026-integrated-system-plan-isp
  4. Australian Energy Market Operator. Quarterly Energy Dynamics Q3 2026 (report). 23 October 2026. https://www.aemo.com.au/-/media/files/major-publications/qed/2026/qed-q3-2026.pdf
  5. Clean Energy Regulator. Small-scale installation postcode data. 20 October 2026. https://cer.gov.au/markets/reports-and-data/small-scale-installation-postcode-data
  6. Australian Renewable Energy Agency. $9 million funding to enhance short-term forecasting of wind and solar farms. 18 March 2019. https://arena.gov.au/news/9-million-funding-to-enhance-short-term-forecasting-of-wind-and-solar-farms
  7. Australian Renewable Energy Agency. Short-Term Forecasting Trial: summary of key insights, April to October 2020. read 10 November 2026. https://arena.gov.au/knowledge-bank/short-term-forecasting-trial-summary-of-key-insights-apr-to-oct-2020/
  8. UNSW Sydney. Sensing a solar revolution for underperforming panels. 2 May 2024. https://www.unsw.edu.au/news/2024/05/sensing-a-solar-revolution-for-underperforming-panels
  9. Australian Renewable Energy Agency. Solar Power Ensemble Forecaster: final report. June 2021. https://arena.gov.au/knowledge-bank/solar-power-ensemble-forecaster-final-report
  10. Australian Renewable Energy Agency. Innovative Optimally Combined Solar Forecasts (Proa Analytics): final report and cost-benefit analysis. read 10 November 2026. https://arena.gov.au/knowledge-bank/pro-analytics-final-report-and-cost-benefit-analysis
  11. Australian Renewable Energy Agency. Gridded Renewables Nowcasting Demonstration over South Australia. read 10 November 2026. https://arena.gov.au/knowledge-bank/gridded-renewables-nowcasting-demonstration-over-sa-public-report