Still Air, Falling Temperatures: Weather Patterns Drive Influenza Surges in a Chinese City

In the subtropical monsoon city of Nanchang, in southeastern China’s Jiangxi Province, influenza does not strike at random. A new surveillance-based analysis of more than 120,000 laboratory-confirmed cases suggests that the rhythm of flu seasons in this city is tightly entwined with the weather, and that a handful of atmospheric variables—above all, wind speed—may offer public health officials an early warning signal before outbreaks crest. The study, published in BMC Infectious Diseases by a team led by Maohong Hu of the Nanchang Center for Disease Control and Prevention, draws on weekly case counts and ten meteorological indicators spanning January 2023 through March 2026, offering one of the most granular portraits to date of how climate and contagion interact in a rapidly urbanizing subtropical setting.

The scale of the dataset is itself notable. Between the start of 2023 and the end of March 2026, Nanchang recorded 120,308 confirmed influenza cases, aggregated and analyzed on a weekly basis alongside local measurements of temperature, humidity, wind speed, cloud cover, precipitation, and related atmospheric conditions. The researchers found that influenza activity in the city peaks in winter and spring, a pattern familiar from temperate latitudes but shaped here by the distinctive dynamics of a monsoon climate, in which hot, humid summers give way to cool, damp winters. The highest annual incidence rate of the study period was recorded in 2025, at 837.36 cases per 100,000 population, with the outbreak peak arriving in Week 49 of that year—a reminder that flu seasons can intensify sharply and early.

Who bears the burden of these outbreaks is equally clear from the data. Individuals aged 0 to 19 accounted for fully 76 percent of all confirmed cases, with students constituting the largest occupational group. That finding underscores a well-established but still underappreciated feature of influenza epidemiology: school-age children act as the engine of seasonal transmission, seeding households and communities as the virus moves through classrooms. Geographically, the heaviest disease burden fell on Xihu District and Honggutan District, dense urban areas where population mixing is likely to accelerate viral spread. For health planners, the message is that prevention and control efforts in Nanchang should prioritize children and adolescents, whether through vaccination campaigns timed ahead of the winter-spring peak or through school-based measures during periods of high meteorological risk.

The study also examined how quickly patients were diagnosed after symptoms began, a metric with direct consequences for treatment and transmission. The picture here was largely encouraging: 89.64 percent of cases were diagnosed within two days of symptom onset, suggesting that Nanchang’s surveillance and clinical systems are generally responsive. But the exceptions were telling. Diagnostic delays were significantly more common among people aged 60 and older and among farmers, groups that may face barriers of access, awareness, or health-seeking behavior. In an era when antiviral treatment is most effective when started early, and when rapid diagnosis can help interrupt household and workplace transmission, these gaps point to a concrete target for intervention: strengthening early diagnostic capacity for elderly and rural populations.

To disentangle the weather’s role, the researchers deployed a battery of statistical methods, each chosen to illuminate a different aspect of the relationship. Seasonal-Trend decomposition using Loess, or STL, allowed them to strip away long-term trends and recurring seasonal cycles from the weekly time series, isolating the short-term fluctuations that might be attributable to changing weather. Spearman’s correlation provided a first, non-parametric read on which meteorological variables moved in step with case counts. Then came the heavier machinery: a Random Forest regression model, an ensemble machine-learning technique that builds hundreds of decision trees on random subsets of the data to rank the predictive importance of each variable while capturing nonlinear relationships that simpler models would miss.

The Random Forest results delivered a clear hierarchy of influence. Average temperature emerged as the most important variable, and intriguingly, its relationship with influenza incidence was U-shaped—meaning that both cold and, to a lesser extent, hot extremes were associated with elevated risk, while moderate temperatures were linked to the lowest activity. Wind speed ranked second in importance, followed by cloud cover and precipitation. This kind of variable-importance ranking is valuable because it tells modelers and health officials which measurements to watch most closely when trying to anticipate the next surge, and it hints at the underlying physics and biology: temperature shapes viral stability and host defenses, while wind and precipitation influence how aerosols disperse and how much time people spend indoors.

To move from correlation toward a more nuanced, time-aware picture of risk, the team turned to the Distributed Lag Nonlinear Model, or DLNM, a framework widely used in environmental epidemiology. The DLNM’s power lies in its ability to model two dimensions simultaneously: how strongly an exposure—say, a given temperature—affects risk, and how that effect is distributed across the days and weeks that follow. This matters because weather’s influence on influenza is not instantaneous. A cold snap this week may suppress immunity, alter behavior, or increase indoor crowding, with consequences for case numbers that unfold over a lagged window. By modeling exposure and lag together in a nonlinear way, the DLNM can estimate cumulative risk across that entire window rather than at a single point in time.

The DLNM findings were striking. Low wind speed was the single strongest factor associated with influenza activity: at a wind speed of just 1.43 meters per second, the cumulative relative risk reached 6.26, meaning that flu incidence under such still conditions was more than six times higher than under the reference conditions. Temperature showed a pronounced asymmetry. Low temperatures carried a cumulative relative risk of 2.88, while high temperatures were associated with a cumulative relative risk of just 0.03—effectively a near-absence of excess risk. In other words, cold, calm weather is the signature atmospheric condition of influenza’s rise in Nanchang, while hot weather offers something close to protection.

Why would still air be so dangerous? The authors’ findings align with a growing body of research on airborne viral transmission. Calm conditions allow virus-laden respiratory aerosols to linger in shared air rather than being dispersed and diluted, effectively raising the dose of infectious particles that susceptible individuals inhale in indoor and semi-enclosed settings. Still, cold days also drive people indoors, where crowding and poor ventilation compound the effect, and cold air itself can impair mucociliary clearance in the airways, weakening a first line of defense. The U-shaped temperature relationship adds a further wrinkle: while cold dominates the risk profile, extreme heat may also create conditions—perhaps through behavioral shifts or physiological stress—in which transmission edges upward, though the effect is far weaker than that of cold.

The practical implications extend beyond Nanchang. The authors argue that meteorological indicators could provide supportive information for influenza surveillance and early-warning systems, complementing virological monitoring with cheap, continuously available weather data. A city that sees wind speeds dropping toward the critical low range while temperatures fall could ratchet up messaging, vaccination outreach, and school-based precautions before case counts climb. For a virus that imposes a heavy global burden year after year, and whose seasonality may shift as climate change alters regional weather patterns, the ability to read the atmosphere as a forecasting tool is an increasingly attractive proposition. The Nanchang study, grounded in three years of dense surveillance and a rigorous modeling framework, offers a template for how other cities in subtropical monsoon zones might build that capability—while making clear that the ultimate beneficiaries should be the children who drive transmission and the elderly and rural residents whose diagnoses too often come late.

Subject of Research: Association between meteorological factors and influenza incidence in Nanchang, China

Article Title: Association between meteorological factors and influenza incidence in Nanchang, China, 2023–2026: a surveillance-based time-series analysis

Article References: Association between meteorological factors and influenza incidence in Nanchang, China, 2023–2026: a surveillance-based time-series analysis. (n.d.). https://doi.org/10.1186/s12879-026-14524-8

Image Credits: AI Generated

DOI: 10.1186/s12879-026-14524-8

Keywords: influenza, meteorological factors, wind speed, temperature, time-series analysis, random forest, distributed lag nonlinear model, epidemiological surveillance, diagnostic delay, Nanchang, China, public health