The review, led by Zecheng Li and Chng Saun Fong of the Institute for Advanced Studies at Universiti Malaya, together with colleagues spanning software engineering, public health, and chemical engineering, synthesised sixty-nine studies on how compound environmental events are defined, forecast, and communicated within early warning system frameworks. The team followed established systematic review methodology, including PRISMA reporting guidelines, and mapped every study against a purpose-built taxonomy that organises the field into three connected stages: how compound events are operationally defined, how they are jointly forecast, and how the resulting risk reaches the public as an actionable warning.

The first stage, event definition, turns out to be far more consequential than it might appear. Most existing systems identify a compound event simply by overlapping fixed or percentile-based thresholds, for example declaring a compound hot-and-polluted episode when both temperature and ozone or particulate matter exceed chosen cut-offs simultaneously. This approach is transparent and easy to automate, but the review shows it is also arbitrary: studies of mortality in Wuhan, Montreal, Beijing, and dozens of other cities demonstrate that the health impact of a given pollutant concentration shifts substantially with temperature, meaning a threshold defined without reference to measured health outcomes can systematically misclassify genuinely dangerous days. Impact-based interaction models, which use epidemiological relationships to weigh the joint effect of heat and pollution, offer a more defensible alternative, and the review documents a rapidly growing body of work fitting distributed lag non-linear models and machine learning methods to hospital admissions and mortality records to derive compound definitions grounded in actual harm.

On the forecasting side, the technological picture has changed dramatically in just a few years. Coupled meteorology-chemistry models such as WRF-Chem and operational systems built on the NOAA Global Forecast System can now simulate how heat, boundary-layer dynamics, and atmospheric chemistry interact, capturing for instance how a heat dome suppresses ventilation and allows pollutants to accumulate. More striking still is the arrival of artificial intelligence in numerical weather prediction. The review highlights machine learning systems that have achieved skillful medium-range global forecasts and, more recently, probabilistic ensemble forecasts generated entirely by neural networks, alongside hybrid approaches that combine physical chemical transport models with machine learning bias correction to deliver high-resolution particulate matter predictions. Interpretable models such as random forests equipped with SHAP explainability are also being used to unpick which environmental variables drive ozone formation and health outcomes, giving forecasters both accuracy and a measure of transparency.

These forecasting advances feed into a third strand of technology: unified risk indices. Rather than issuing a heat alert and an air quality alert independently, several research groups have constructed combined indices that merge temperature and multiple pollutants into a single health-relevant number, validated against mortality in places as varied as Monterrey, Taiwan, and Beijing. Studies building graded heat-health risk forecasts with full-season coverage across China demonstrate that such integrated products can be produced at national scale. The review treats these indices as a crucial bridge between the raw machinery of forecasting and the blunt reality of public communication, because a single number with a clear protective message is far easier to act on than two parallel warnings that may never be reconciled.

The final stage, risk communication and warning triggers, is where the review finds both encouraging innovation and stubborn weaknesses. Impact-based forecasting, now promoted by the World Meteorological Organization, shifts the emphasis from describing what the weather will be to describing what it will do, and evaluation studies from South Korea and New Zealand show that warnings framed around expected impacts measurably improve risk perception and protective behaviour. Probabilistic trigger rules drawn from forecast-based financing allow humanitarian agencies to release funds before a disaster strikes, and research on visual and verbal communication of uncertainty shows that well-designed probabilistic messages help decision makers rather than confuse them. Newer experiments, including digital heat warning platforms for older adults and generative AI chatbots tailored to multi-lingual communities, hint at how compound warnings might one day reach vulnerable individuals directly.

Yet despite this rapid progress at every stage, the review’s central finding is a persistent integration gap. Definitions are developed by epidemiologists, forecasts by atmospheric modellers, and warning triggers by disaster management agencies, usually without shared targets or common validation. A compound event may be defined one way in a research paper, forecast another way in an operational model, and trigger a warning under a third set of rules entirely. Uncertainty estimates produced by ensemble forecasts rarely survive the journey into the public warning product. The authors argue that this fragmentation means the impressive technical capabilities documented across the literature are not being converted into coherent, end-to-end systems, and that the missing ingredient is a shared compound-risk target that all three stages are explicitly designed to serve.

To remedy this, the review proposes both a taxonomy and a research agenda. The taxonomy classifies compound-event generation methods by how they combine hazards, whether through threshold overlap, statistical interaction models, coupled physics-based simulation, or AI-driven fusion, and by how they connect to impacts and triggers. The research agenda calls for uncertainty-aware model fusion so that probabilistic information flows unbroken from forecast to warning; for impact-based warning thresholds calibrated against health and service outcomes rather than convenient percentiles; and for end-to-end validation that tests not whether a model predicted a hazard accurately, but whether the resulting warning changed behaviour and reduced harm. The authors also stress equity considerations documented in the underlying literature, including evidence that the health effects of compound heat and pollution fall disproportionately on historically marginalised neighbourhoods, and that people-centred design remains essential even as the technology becomes more sophisticated.

The timing of this synthesis is significant. Under the United Nations Early Warnings for All initiative, governments worldwide are being pressed to expand multi-hazard early warning coverage, and climate change is steadily increasing the frequency with which extreme heat and air pollution coincide over populated regions, from the Pearl River Delta to California, Delhi, and London. The review’s message to policymakers is that simply adding more hazards to existing single-hazard systems will not suffice; the architecture itself must become compound-aware. If the integration gap can be closed, the authors conclude, the technological ingredients for warnings that genuinely reflect the compound risks people face already exist. What remains is the harder work of connecting them.

Subject of Research: Technological advances in generating compound heat and air pollution events within early warning system frameworks

Article Title: A systematic review of technological advances in compound-event generation within early warning system frameworks

Article References: Li, Z., Fong, C. S., Ab Hamid, S. H., Aghamohammadi, N., Jamali, S. N., & Sulaiman, N. M. (2026). A systematic review of technological advances in compound-event generation within early warning system frameworks. Air Quality, Atmosphere & Health, 19(9), Article 204. https://doi.org/10.1007/s11869-026-02091-5

Image Credits: AI Generated

DOI: 10.1007/s11869-026-02091-5

Keywords: compound events, early warning systems, extreme heat, air pollution, impact-based forecasting, machine learning, risk communication, public health, climate risk, heat-health warning, coupled meteorology-chemistry models, systematic review