Abstract green illustration of a data network and forecast curve over a city skyline

AI and asthma prediction in 2026: what forecasts can and can't tell you about your home

AI & healthBy Dr Cameron Jones, PhD, environmental microbiologist, Biological Health Services, Melbourne

Artificial intelligence is now part of how Victoria watches the air in spring. Machine-learning pollen forecasts, robot pollen counters and software that spots asthma spikes in emergency departments all feed into one of the world's most advanced thunderstorm asthma warning systems. Researchers are also training models to predict asthma attacks for individual patients. Here's what is real in 2026, what's still promise, and what it all means for the air inside your home.

Key takeaways

  • Since 2017, Victoria has run the world's first routine daily thunderstorm asthma risk forecast covering a wide area. It now uses machine-learning pollen forecasts and automated pollen counters.
  • AI pollen counters in Melbourne closely matched traditional manual counts for grass pollen.
  • AI models that predict individual asthma attacks show promise, but reviews say most need more validation before everyday clinical use.
  • Forecasts describe outdoor, regional conditions. They can't tell you what's growing in your bedroom.

Victoria's thunderstorm asthma forecast

After the November 2016 epidemic, the Victorian health department asked the Bureau of Meteorology and research partners to build a forecasting system. It began on 1 October 2017. It combines forecasts of gusty winds in severe thunderstorms with statistical forecasts of high grass pollen, and was the first routine daily thunderstorm asthma risk forecast in the world to cover a wide area [1].

Today the forecast runs from 1 October to 31 December. It covers three days at a time and uses a colour scale of green (low), orange (moderate) and red (high), published on VicEmergency [2]. The department is upfront that it is not a guarantee. A higher forecast doesn't mean an event will happen, and a lower one doesn't mean it won't [2]. In its first three years, the forecasts showed skill at telling apart days with and without health impacts, but some busy hospital days happened in districts that had been rated low risk [1].

Where AI comes in

Smarter pollen forecasts

A 2024 review of the system by the partners behind it lists the improvements made since launch. They include more accurate grass pollen forecasts based on machine learning, automated pollen monitors replacing manual counting, and near-real-time detection of asthma spikes in emergency departments using natural language processing [3]. The pollen forecasts draw on wind, temperature, rainfall, humidity and satellite estimates of grass cover and condition [2].

Robot pollen counters

For decades, pollen was counted by people looking down microscopes [4]. Victoria now has automated counters at Parkville, Geelong, Hamilton, Ballarat, Bendigo, Wodonga and Churchill [2]. The counter tested in Melbourne captures holographic images of each particle as it flows through the device, and an AI model identifies the pollen [4]. A 2026 study compared one of these counters with a traditional trap at the University of Melbourne over the 92-day spring season of 2022. For grass pollen, the two agreed closely (R² = 0.92), including when sorting days into low, moderate and high thresholds [4].

Next: burst pollen and fungal spores

Current automated counters can't tell intact pollen from ruptured pollen, which is the form linked to thunderstorm asthma. A 2025 review argues that AI could learn to make that distinction and improve warnings [5]. Researchers in Europe have trained the same holographic technology to recognise airborne spores of the mould Alternaria in real time, with results that correlated well with a manual trap [6]. That's an exciting step for outdoor monitoring, though it's still early days.

Forecasts watch the sky. Who's watching your house?

Outdoor forecasts can't see inside your walls. A lab-analysed settle plate test shows you which moulds are settling in the rooms where you sleep, work and play.

Can AI predict your next asthma attack?

This is the other big area of research: using health records, environmental data and sometimes sensors to estimate a person's risk of an asthma flare-up.

What works best so far

A 2024 review of 20 studies found that ensemble methods based on decision trees generally performed best for classifying future attacks, and that shorter prediction windows improved results [7].

How accurate it is

A 2024 meta-analysis of models for children found a pooled score (AUC) of 0.67 for predicting emergency visits and 0.79 for hospital admissions. The authors say external validation is needed before real-life use [8].

The catch

A 2025 review of adult models published from 2017 to 2023 rated 80% as having a high risk of bias, and concluded they had limited practical use so far [9].

In plain terms, an AUC of 0.5 is a coin toss and 1.0 is perfect. Scores in the 0.7s are useful but far from certain. These tools may help doctors one day, but they aren't something to rely on at home in 2026.

What this means for your home

  • Use the forecasts. Check VicEmergency during grass pollen season, set a watch zone, and use the Melbourne Pollen app or website for local pollen counts [2].
  • Remember what they measure. Forecasts describe outdoor pollen and storms across a district. They don't measure your home's indoor air.
  • Don't forget damp. Building dampness and mould are associated with roughly 30–50% increases in a range of respiratory and asthma-related outcomes [10]. That's a year-round, indoor issue no pollen forecast will flag.
  • Measure, don't guess. Good AI depends on good measurements. The same goes for your home.

Where a mould test fits

Our DIY kits use potato dextrose agar (PDA) settle plates exposed for 45 minutes in the rooms you choose. The plates are analysed by Biological Health Services, which holds ISO 9001:2015 certification (AUS-Q-900149) and an ISO/IEC 17025:2017 Certificate of Compliance (AUS-TL-17025-149). Your report shows the mould count, the moulds identified to genus where possible, and an estimated CFU/m³ calculated with the published Omeliansky formula.

A settle plate doesn't detect pollen and it can't forecast asthma. What it gives you is real, lab-checked data about mould in your own indoor air, which is something no forecast can provide. Read how the method works on our Our Method page.

Please note: This article is general information, not medical advice. If you have asthma, follow your asthma action plan and speak with your doctor. In an emergency, call 000.

Add real data from your own home

Pair the outdoor forecasts with a lab-analysed look at your indoor air. Simple to use, posted back to our Melbourne lab.

Renting and worried about mould? See our Mould & Your Rental guides.

References

  1. Bannister T, Ebert EE, Williams T, et al. (2021). A pilot forecasting system for epidemic thunderstorm asthma in southeastern Australia. Bulletin of the American Meteorological Society 102(2):E399–E420. https://doi.org/10.1175/BAMS-D-19-0140.1
  2. Department of Health, Victoria. Epidemic thunderstorm asthma risk forecast. health.vic.gov.au
  3. Ebert E, Csutoros D, Black J, Lampugnani ER, Nattala U, Newbigin E (2024). Using a value chain to develop a new warning service from scratch: epidemic thunderstorm asthma risk forecasts in Victoria, Australia. EMS Annual Meeting 2024, EMS2024-631. https://doi.org/10.5194/ems2024-631
  4. Lampugnani ER, Nattala U, Csutoros D, et al. (2026). Evaluating automated pollen monitoring for thunderstorm asthma and public health forecasting: a comparative study in Melbourne, Australia. Environmental Monitoring and Assessment 198(11):1187. https://doi.org/10.1007/s10661-026-16017-2
  5. Hanoun K, Hughes KM, Suphioglu C (2025). Review article: how emerging technologies could reshape pollen monitoring for epidemic thunderstorm asthma. Aerobiologia 41(3):559–567. https://doi.org/10.1007/s10453-025-09874-w
  6. Erb S, Berne A, Burgdorfer N, et al. (2024). Automatic real-time monitoring of fungal spores: the case of Alternaria spp. Aerobiologia 40(1):123–127. https://doi.org/10.1007/s10453-023-09780-z
  7. Darsha Jayamini WK, Mirza F, Asif Naeem M, Chan AHY (2024). Investigating machine learning techniques for predicting risk of asthma exacerbations: a systematic review. Journal of Medical Systems 48:49. https://doi.org/10.1007/s10916-024-02061-3
  8. Votto M, De Silvestri A, Postiglione L, et al. (2024). Predicting paediatric asthma exacerbations with machine learning: a systematic review with meta-analysis. European Respiratory Review 33(174):240118. https://doi.org/10.1183/16000617.0118-2024
  9. Liu A, Zhang Y, Yadav CP, Chen W (2025). An updated systematic review on asthma exacerbation risk prediction models between 2017 and 2023: risk of bias and applicability. Journal of Asthma and Allergy 18:579–589. https://doi.org/10.2147/JAA.S509260
  10. Fisk WJ, Lei-Gomez Q, Mendell MJ (2007). Meta-analyses of the associations of respiratory health effects with dampness and mold in homes. Indoor Air 17(4):284–296. https://doi.org/10.1111/j.1600-0668.2007.00475.x
Back to blog