The role of tropical cyclones in accelerating dengue transmission: insights from field experiments and dynamical modelling Authors: Yu Yan, Mengjie Geng, Yuwen Zhong, Hui Deng, Shengjun Hu, Yating Chen, Wenting Guo, Yan Li, Haitao Wang, Taiyuan Zhang, Wentao Hu, Qi Zhao, Wei Ma Published: 2026 ABSTRACT AbstractIntroductionTo quantify the impact of tropical cyclones on dengue transmission dynamics and related disease burden by integrating field entomology with dynamic modelling in Guangzhou, China.MethodsGuangzhou was selected as the study site considering that it is China’s most dengue-endemic coastal megacity with frequent tropical cyclones and the country’s largest recorded dengue outbreak in 2014. Mark–release–recapture experiments with laboratory simulations were performed in 2024 to assess the flight distance, reproduction and survival of Aedes albopictus under cyclone and baseline conditions to estimate human–mosquito contact rates. These data informed a dynamic susceptible–incubation–infected–recovered model simulating the 2014 dengue epidemic during Typhoons Rammasun and Kalmaegi. We estimated excess cases and economic losses against counterfactual baselines and projected the impact of enhanced vector control.ResultsDuring cyclones, Aedes albopictus maximum flight distance increased by 63.0% and human–mosquito contact rate by 30.1%. Estimated R0 increased from 1.10 (baseline) to 1.43 (Rammasun) and 1.29 (Kalmaegi). The estimated excess burden reached 217 excess cases (US$76 400) for Rammasun and 5131 cases (US$1.93 million) for Kalmaegi. Enhanced vector control during Kalmaegi was projected to avert 59 360 additional cases (US$20.91 million).ConclusionsTropical cyclones amplify dengue transmission by expanding mosquito dispersal and intensifying human–vector contact. Early vector control, guided by warnings, substantially reduces cases and costs.Keywords: Dengue, Environmental health, Epidemiology WHAT IS ALREADY KNOWN ON THIS TOPIC Dengue is a fast-spreading global vector-borne disease; climate factors shape its transmission, with Guangzhou (southern China) accounting for 82% of China’s 2014 outbreak (471 000 cases). Dengue is a fast-spreading global vector-borne disease; climate factors shape its transmission, with Guangzhou (southern China) accounting for 82% of China’s 2014 outbreak (471 000 cases). Tropical cyclones are tied to elevated dengue risk (eg, 2.13 risk ratio in Guangzhou postcyclone), yet their underlying transmission mechanisms remain poorly understood. Tropical cyclones are tied to elevated dengue risk (eg, 2.13 risk ratio in Guangzhou postcyclone), yet their underlying transmission mechanisms remain poorly understood. Climate change will increase cyclone frequency in coastal China by 18%, but data on cyclone-driven Aedes behaviour, along with the associated models, are lacking. Climate change will increase cyclone frequency in coastal China by 18%, but data on cyclone-driven Aedes behaviour, along with the associated models, are lacking. WHAT THIS STUDY ADDS Via mark–release–recapture experiments and lab simulations, we show cyclones increase Aedes’ flight distance by 63% and human–mosquito contact rates by 30.1%. Via mark–release–recapture experiments and lab simulations, we show cyclones increase Aedes’ flight distance by 63% and human–mosquito contact rates by 30.1%. Integrating these data into a susceptible–incubation–infected–recovered model, we estimate cyclones caused 217 (Rammasun) and 5131 (Kalmaegi) excess 2014 Guangzhou dengue cases (costing US$76 k and US$1.93M). Integrating these data into a susceptible–incubation–infected–recovered model, we estimate cyclones caused 217 (Rammasun) and 5131 (Kalmaegi) excess 2014 Guangzhou dengue cases (costing US$76 k and US$1.93M). Enhanced cyclone-period vector control could avert 59 360 cases (US$20.91M); cyclones amplify transmission via wider mosquito dispersal and more human contact. Enhanced cyclone-period vector control could avert 59 360 cases (US$20.91M); cyclones amplify transmission via wider mosquito dispersal and more human contact. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings support aligning early vector control with cyclone warnings to cut dengue cases and costs in cyclone-prone regions like southern China. These findings support aligning early vector control with cyclone warnings to cut dengue cases and costs in cyclone-prone regions like southern China. Our integrated field entomology-dynamic modelling framework offers a replicable way to study climate-driven vector-borne disease mechanisms. Our integrated field entomology-dynamic modelling framework offers a replicable way to study climate-driven vector-borne disease mechanisms. The results inform targeted dengue responses, strengthen climate-disease surveillance and aid planning for climate-exacerbated extreme weather. The results inform targeted dengue responses, strengthen climate-disease surveillance and aid planning for climate-exacerbated extreme weather. Introduction Dengue is a rapidly expanding vector-borne viral disease predominantly transmitted by Aedes albopictus and Aedes aegypti.1,3 Climatic conditions play a critical role in shaping the spatiotemporal distribution of dengue, with certain elements such as warmer season, temperature and rainfall widely recognised as key determinants.3,6 In line with the changing climate in the past decades, 4 billion people are at risk globally and 50–100 million infections occur annually, making it a major global public health threat.7 In China, dengue has been the only vector-borne disease to show a consistent rise in incidence since 2005, with southern coastal cities such as Guangzhou emerging as persistent transmission hotspots.8 9 Extreme weather events, particularly tropical cyclones (TCs) have emerged as potential short-term amplifiers of dengue outbreaks.10,14 TCs impose abrupt hydroclimatic changes, including intense rainfall, elevated humidity and transient cooling, all of which can theoretically create favourable conditions for mosquito survival, dispersal and contact with humans.15,18 Previously, several studies have reported temporal associations between TCs and dengue incidence. For example, research in the Pearl River Delta region of China indicated a relative risk of 1.62. Similarly, in Guangzhou, TCs were linked to a cumulative risk ratio of 2.13 (95% CI 1.28 to 3.56) over lag 0–4 weeks. However, these findings are largely ecological and offer limited insight into the mechanisms through which TCs alter transmission dynamics. Under projected climate scenarios, the frequency and intensity of TCs along China’s coastlines are anticipated to escalate, with Guangdong Province experiencing the most substantial increase (0.27 yr−1). Furthermore, national TC landfalls are projected to increase by 18%, rising from a historical baseline of 3.98 yr−1 (1986–2005) to 4.69 yr−1 by the end of the century. This escalating threat highlights an imperative need to unravel how TCs influence the transmission dynamics of dengue.19 Despite the dramatic development of cyclone epidemiology for dengue, several knowledge gaps persist. First, the effect of TCs on Aedes behaviours has not been characterised through field-based approaches, specifically for flight distance, survival and host-seeking activity. Second, reliable estimates of human–mosquito contact rate during cyclone periods are lacking. Although the human landing catch method is the standard for measuring this key parameter driving vector-borne disease transmission,20 its ethical limitations restrict the wide application. Alternatives such as retrospective surveys are prone to bias and lack temporal precision.21 Furthermore, previous case–control investigations have established that elevated mosquito biting frequency constitutes an independent risk factor for dengue infection during TCs.22 This underscores human–mosquito contact intensity as the primary pathway mediating TCs associated transmission, highlighting the necessity to quantify this parameter. Third, while robust dynamic transmission models have been widely deployed to evaluate dengue responses to standard meteorological variables (eg, temperature and precipitation) using laboratory-validated thermal performance curves, few have integrated empirical entomological data to assess cyclone-induced alterations in transmissibility.23 Additionally, existing models predominantly use temperature-dependent vector parameters sourced from non-local populations, thereby underusing critical estimates specific to indigenous Aedes albopictus populations in China.24 These limitations hinder accurate prediction and timely response to cyclone-related dengue outbreaks. To address these gaps, a study integrating field experiments and mathematical modelling is necessary in cyclone-prone locations. Guangzhou is a dengue-endemic subtropical megacity with frequent cyclone exposure (averaging 3.4 occurrences per year from 1980 to 2020).25 The city experiences a disproportionately high dengue burden, with 43 802 notified cases and nine associated deaths documented from 2011 to 2019, yielding an accumulated incidence of 29.88 per 100 000.26 27 Prior to 2014, the region was essentially non-endemic, characterised by minimal local activity and the circulation of single serotypes.28 29 This historical baseline was shattered by the 2014 epidemic, marking the most severe dengue outbreak in China in the past half-century. Guangzhou was the epicentre of this crisis, contributing 82.6% of the nationally notified cases and recording 14 048 local hospitalisations.30 31 This dengue outbreak is also marked by the passage of two major TCs, Rammasun and Kalmaegi. These factors make Guangzhou an ideal location to investigate the impacts of TCs on dengue transmission dynamics. In this study, we developed a two-stage time-varying susceptible–incubation–infected–recovered (SEIR) model parameterised by mark–release–recapture (MRR) experiments and laboratory simulations. The estimated excess dengue burden associated with the coincided TCs during Guangzhou’s 2014 epidemic was estimated. By linking empirical entomology with dynamic modelling, this work provides new insights into the climate sensitivity of dengue transmission and informs anticipatory public health strategies in cyclone-prone regions. Methods Study design and ethical approval This study developed a two-stage time-varying SEIR model to quantify the impact of TCs on dengue transmission (online supplemental figure S1). In the first stage, MRR and laboratory simulations were performed to assess the flight, reproduction and survival of Aedes albopictus across varying conditions under cyclone and baseline conditions, enabling the calculation of human–mosquito contact rates. In the second stage, these entomological parameters were integrated into the SEIR model to simulate the dengue transmission dynamics and related economic loss during the 2014 dengue epidemic that coincided with two major TCs. All analyses were conducted using MATLAB (V.R2024b). Definition of tropical cyclone exposure TC events were identified using the Yearbook of Tropical Cyclone and the China Weather Typhoon Network (http://typhoon.weather.com.cn/), categorised into six intensity levels based on the maximum wind speed near centre (online supplemental table S1).32 A city was considered cyclone-affected if (1) a typhoon warning was issued; and (2) daily rainfall exceeded 25 mm or maximum wind speed reached ≥ level 7. Following previous studies, the ‘cyclone-effect period’ was defined as 7–28 days after the initial impact of TC.11,13 Data collection Daily dengue case in Guangzhou during 2014 were obtained from the China National Notifiable Disease Surveillance System. All cases were diagnosed according to the standard guidelines issued by the National Health Commission.33 Monthly Aedes albopictus Breteau Index (BI) data in 2014 were provided by the Guangdong Centre for Disease Control and Prevention. Daily mean temperature of Guangzhou in 2014 were obtained from the China Meteorological Administration (0.25°×0.25°) by weighting city boundary.34 35 The number of permanent residents were derived from the statistical yearbook.36 Field experiment Marker release recapture MRR experiments were conducted on Shazai Island, Nansha District, in September 2024, during Super Typhoon Yagi (September 15–21) and a postcyclone baseline period (September 26–October 2). Marked adult Aedes albopictus were released, and 18 BG traps (without CO2) within a 200 m radius were used for recapture over 7 days (online supplemental figure S2). Detailed experimental procedures can be found in online supplemental file 1. Calculation of human–mosquito contact rate The human–mosquito contact rate (c) is defined as the total number of contact events between humans and mosquitoes per unit time within a specified area. Following previously reported evidence,37 38 c is calculated using the following formula, with its value constrained to 0.4–0.9539: c0: The baseline human–mosquito contact rate, set at 0.7 based on previous studies.39 D: The observed maximum flight distance of mosquitoes. D0: The reference flight distance, set at 80 m based on previous studies.37 φ: The observed daily survival rate of mosquitoes, calculated from the MRR experiment. φ0: The baseline daily survival rate, determined from laboratory experiments. ρ: The recapture rates of female mosquitoes during the experimental period, defined as the number of recaptured female divided by the number of released female. ρ0: The baseline recapture rates of female mosquitoes, based on baseline period data and prior studies, set at 2.5%.37 Statistical analysis The mathematical model The SEIR model incorporated mosquito life stages and human infection states (figure 1). To model mosquito population dynamics, we adopt the framework proposed by Yang et al,40 which classifies mosquitoes into aquatic immature and mature adult stages. We further extended the adult female mosquitoes into epidemiologically relevant stages for dengue transmission: susceptible (Sm), incubating (Em) and infectious (Im). The human population was categorised into susceptible (Sh), incubating (Eh), symptomatic (Ih) and asymptomatic infections (Yh), imported cases (Wh) and recovered (Rh), with the assumption that the population is fully susceptible to the virus. The transmission of dengue virus between mosquitoes and humans. A, aquatic immature stages; Eh, latent period population; Em, incubating vector; Ih, symptomatic infected individuals; Im, infected vector; Rh, recoverer; Sh, susceptible population; Sm, susceptible vector; Wh, imported infections; Yh, asymptomatic carrier. Equations (2)–(5) describe the dynamics of dengue transmission within mosquito populations: Where θ is the intrinsic oviposition rate of an adult mosquito, μa and μm are the respective mortality rates of immature forms and adult mosquitoes, ε is the transition rate from immature stages into adult mosquitoes, k is the fraction of eggs hatching to larvae, η is the inhibition rate of precipitation on larva and f is the proportion of female mosquitoes hatched from all eggs. A is the immature population size. A0 was not directly fitted. Instead, its value was determined via sensitivity analysis, with details provided in Supplementary (see Statistical analysis and online supplemental figure S3). The effective elimination rates for Aedes larvae (αa) and adult mosquitoes (αm) were assumed to be equivalent. c is the human–mosquito contact rate, and b is the biting rate of mosquitoes. Equations (6)–(11) describe the dynamics of dengue transmission within human populations: For humans, on exposure to infectious mosquito bites, susceptible individuals Sh enter the latent stage Eh at a rate determined by the human–mosquito contact rate (c), the biting rate of mosquitoes (b) and the mosquito-to-human transmission probability per bite (βmh). Then, they progress to either symptomatic and asymptomatic infection at rates p. The incubation period is σh days, the durations of symptomatic, asymptomatic and imported infections are γh, γy and γw days, respectively. Transmission is further influenced by the human-to-mosquito transmission probability per bite (βhm). Human mortality was not considered in the model. Model assumptions Three core assumptions were adopted in this SEIR transmission model. First, the total human population was assumed to remain constant throughout the study period. As each period lasted merely 21 days, demographic changes including natural birth, death and population migration were negligible and thus excluded from the model. Second, we assumed complete susceptibility of the local human cohort to dengue virus infection. Epidemiological surveillance prior to the 2014 epidemic indicated non-endemicity, characterised by rare local transmission and limited serotype circulation. Consequently, the influence of pre-existing herd immunity and multiserotype cocirculation was deemed negligible, allowing us to exclude immune-mediated phenomena such as antibody-dependent enhancement from the model. Third, the effective elimination rates for Aedes larvae and adult mosquitoes were assumed to be equal. This structural simplification is a standard convention in vector-borne disease modelling, used to reduce the computational complexity of the mosquito life cycle.15 To account for empirical vector control efforts, the study period was stratified into two distinct phases, separated by 24 September 2014.41 The preintervention phase (encompassing the baseline and Rammasun periods) used baseline elimination rates, whereas the postintervention phase (the Kalmaegi period) incorporated enhanced elimination rates to reflect intensified adulticidal fogging and larval source management. The assumption of equal elimination rates for larvae and adults was strictly maintained within each respective phase. Parameterisation was executed independently for each phase using grid-search optimisation to minimise the root mean square error (RMSE) (online supplemental figures S8 and S9). To validate the model’s consistency, the c values were estimated using a grid search method that minimised the RMSE between model output and observed data (online supplemental figure S10). Table 1 summarises parameters incorporated in the model, which are related to transmission dynamics, human and mosquito populations, and environmental factors. Key parameters were determined from previous studies and laboratory experiments. Specifically, based on the temperature-dependent non-linear trait relationships verified in the original literature, we directly matched and imported published measured parameters corresponding to the observed average temperature intervals during each study period.24 The relationship between temperature and biting rate, as well as other parameter details, can be found in the online supplemental tables S2–S5. Table 1Model parameters descriptionTypeDescriptionNotationAccessRammasunKalmaegiBaselineTransmission dynamicsTransmission probability of human-to-vectorβhmFixed parameter0.5500.5500.550Transmission probability of vector-to-humanβmhFixed parameter0.4000.4000.400Human–mosquito contact ratecMRR result0.9500.9500.730Mosquito effective biting ratebExperimental result0.2190.2170.219Human population characteristicsHuman latency period (day)σhFixed parameter1/71/71/7Duration of symptomatic infection (day)γhFixed parameter1/71/71/7Duration of asymptomatic infection (day)γyFixed parameter1/71/71/7Fraction of symptomatic infectionpFixed parameter0.2500.2500.250Mosquito characteristicsFraction of female mosquitoes hatched from all eggsfFixed parameter0.5000.5000.500Fraction of eggs hatching to larvaekFixed parameter0.5000.5000.500Intrinsic oviposition rate of adult mosquitoθExperimental result510.794504.140510.794Mortality rate of aquatic mosquitoμaExperimental result0.1750.1200.175Mortality rate of adult mosquitoμmExperimental result0.0390.0310.039Transition rate from aquatic into adult mosquitoεExperimental result0.8250.8800.825Extrinsic incubation period of adult mosquitoσmExperimental result3.4258.1403.425The inhibition rate of precipitation on larva η Fixed parameter3.84×10−73.84×10−73.84×10−7Input factorImported dengue casesBDisease surveillance system0.0502.0000.050Duration of imported infection (day)γwFixed parameter1/51/51/5Mosquito eliminationEffective mosquito-borne clearanceαa/αmGrid retrieval0.0060.8020.006MRR, mark–release–recapture. Model parameters description MRR, mark–release–recapture. Calculation of basic reproduction number The basic reproduction number R0 was calculated using the next-generation matrix method.42 Rmh denotes the expected number of humans that a single infected mosquito can infect over its lifetime, whereas Rhm represents the expected number of mosquitoes that can be infected by an infected human during their infectious period. R0, defined as the geometric mean of Rmh and Rhm, reflects the overall intensity of transmission. Assessment of excess cases and economic loss. To quantify the excess dengue cases attributable to TCs, we compared the primary output of the SEIR model with a simulated counterfactual scenario. The difference between these two projections is hereafter defined as the estimated excess cases. Given that our analysis does not encompass all cyclone-associated transmission drivers (eg, vector breeding changes, human mobility, healthcare access), these estimates should be interpreted as scenario-based projections rather than absolute causal attribution. Economic burden was calculated from excess symptomatic cases during cyclone period, using national survey data on hospitalisation rates (83% of all symptomatic individuals), direct costs (US$270 per inpatient and US$70 per outpatient), and indirect productivity losses (US$70 per inpatient and US$20 per outpatient).43 Costs were expressed in 2014 USD. Uncertainty quantification To quantify the uncertainty of these scenario-based point estimates, we employed a parametric bootstrapping method. Specifically, we treated the deterministic model predictions as the expected values and resampled the daily incident cases using a Poisson distribution across 1000 iterations to simulate observation stochasticity. The 95% CIs for the excess cases and economic costs were then derived from the 2.5th and 97.5th percentiles of these bootstrapped outcomes. Patient and public involvement Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research. Result Epidemiological overview In 2014, Guangzhou experienced a large dengue outbreak from July to October (online supplemental figures S4 and S5). This outbreak coincided with Super Typhoon Rammasun (July 22–August 11) and Typhoon Kalmaegi (September 24–October 14), which was separated by a baseline period (August 19–September 19). The daily infections peaked at 713 cases during Kalmaegi’s affected period. The BI was highest in June and remained ≥6 through October (online supplemental figure S6). In 2024, relative humidity was 60–75% during cyclones, which decreased steadily to 50% in baseline (online supplemental figure S7). The average temperature in both periods was similar, ranging between 30 and 34℃. MRR and human–mosquito contact rate A total of 2188 (75.0% female) and 1600 (77.0% female) Aedes albopictus were released during cyclone conditions and during baseline conditions, respectively. Recapture rates were higher during cyclones (5.35%) than the baseline (2.56%) (online supplemental tables S6 and S7). Specifically, the recapture rates of female mosquitoes were 3.17% and 2.35%, respectively. Between the baseline and cyclone periods, the maximum flight distance of mosquitoes increased by 63.0% from 119 m to 194 m, and the human–mosquito contact rate increased by 30.1% from 0.73 to 0.95. Model parameter changes During Typhoon Kalmaegi, the average temperature decreased from 29.5℃ (during the period of super Typhoon Rammasun) to 26.2℃. The egg-laying and biting rates of adult mosquitoes remained relatively stable. Larval and adult mortality decreased, while the aquatic-to-adult transition increased, and extrinsic incubation period lengthened (table 1). Mosquito elimination rates (αm and αa) increased significantly from 0.006 to 0.087 before and during the effect period of Typhoon Kalmaegi (online supplemental figures S8 and S9). Transmission dynamics As shown in figure 2, the fitted SEIR model reproduced observed trends of the 2014 epidemic, with the long-term transmission trajectories further detailed in the 100-day extended simulations (online supplemental figure S11). Cyclone conditions accelerated epidemic growth, with faster increases in infectious mosquito and human populations compared with baseline periods. For example, the estimated R₀ was 1.10 at the baseline, which increased by 30.0% to 1.43 during Rammasun and increased by 17.3% to 1.29 during Kalmaegi, respectively. Mosquito infection dynamics differed between cyclones (figure 3, online supplemental figure S12): During Typhoon Kalmaegi, mosquito populations in all stages declined faster than in the baseline and Rammasun periods, whereas Rammasun saw a sharper rise in incubating mosquitoes and a slower decline in infectious ones. The simulation result of dengue in difference period in 2014. Mosquito population dynamics during different periods. (A) susceptible mosquitoes; (B) incubating mosquitoes; (C) infectious mosquitoes. Additional burden and cost As shown in table 2, Rammasun was associated with an estimated 217 excess symptomatic cases (95% CI 203 to 230), incurring US$76 400 in costs (95% CI 71 507 to 81 018). Conversely, Kalmaegi contributed 5131 excess cases (95% CI 5004 to 5265), resulting in an economic burden of US$1.93 million (95% CI 1.76 to 1.85 million). Notably, enhanced vector control during the Kalmaegi period was projected to have prevented 59 360 additional cases (95% CI 58 836 to 59 904), thereby averting US$20.91 million in economic costs (95% CI 20.73 to 21.10 million). Table 2The increased economic burden during different periods of tropical cycloneScenariosExcess reported casesHospitalisation costsOutpatient costsTotal economic burdenDirectIndirectDirectIndirectRammasun217(203 to 230)57 343.32(53 677.405 to 60 816.764)14 768.03(13 823.920 to 15 662.570)3427.15(3196.567 to 3621.726)867.33(808.973 to 916.570)76 405.83(71 506.865 to 81017.630)Rammasun(No elimination)*−11(-41 to 0)−2867.17(−10 841.249 to 0.000)−738.40(−2723.925 to 0.000)−185.25(−277.882 to 0.000)−46.88(−70.325 to 0.000)−3837.70(−13913.381 to 0.000)Kalmaegi5131(5004 to 5265)1 356 806.61(1 323 161.250 to 1 392 175.056)349 427.97(340 763.040 to 358 536.652)80 769.66(78 796.158 to 82 906.030)20 440.83(19 941.383 to 20 981.491)1 807 445.06(1 762 661.830 to 1 854 599.228)Kalmaegi(No elimination)*−59 360(−59 904 to −58 836)−15 695 821.75(−15 839 858.412 to −15 557 457.090)−4 042 255.58(−4 079 350.352 to −4 006 621.550)−934 686.47(−943 286.383 to −926 468.977)−236 546.31(−238 722.740 to −234 466.666)−20 909 310.11(−21 101 217.887 to −20 725 014.283)*Indicates the reduction in reported cases attributed to mosquito elimination measures during difference tropical cyclone period. The increased economic burden during different periods of tropical cyclone Indicates the reduction in reported cases attributed to mosquito elimination measures during difference tropical cyclone period. Discussion This study found that TCs measurably amplified dengue transmissibility. Using a SEIR framework parameterised with entomology, the R0 was estimated to increase by 17.3–30.0% during the two cyclone periods than the baseline in Guangzhou’s 2014 epidemic, tracking the observed acceleration in incident cases during cyclone-effect windows. These surges coincided with increased human–mosquito contact (by 30.1%) and longer Aedes albopictus dispersal (by 63.0%), indicating enhanced opportunity for vector–host encounter at a short range. While projected cyclone-associated burden was non-trivial, intensified vector control undertaken for Kalmaegi plausibly averted substantial dengue infections and related economic loss, underscoring the value of timely surge actions when cyclone alerts are issued. Added value of this study Most previous studies linking cyclones to dengue have been ecological, with limited ability to separate epidemiological mechanism from statistical correlation.10,14 We integrate three components rarely combined in a single analysis: (1) field-based estimates of Aedes albopictus movement and survival derived from a MRR experiment; (2) laboratory-informed, temperature-dependent life-history parameters; and (3) a time-varying SEIR model that explicitly represents latent and infectious stages in both vectors and humans and links transmission to an empirically derived, cyclone-responsive contact process. The resulting framework reproduced epidemic trends and isolated cyclone-period dynamics from baseline behaviour, providing a transparent bridge between atmospheric shocks and transient changes in R0. Notably, our MRR was conducted in Guangzhou in 2024 (Typhoon Yagi and a postcyclone baseline) to obtain contemporary behaviour and survival inputs for contact-rate estimation. These empirics were then coupled to 2014 surveillance and weather to interrogate the historical epidemic, an approach that is pragmatic in settings where high-resolution 2014 entomology is unavailable. Our findings complement and extend a growing body of literature showing that extreme weather perturbs infectious disease risk and health service demand—for example, elevated postcyclone infectious-disease hospitalisations—and that dengue risk is rising globally as climate suitability expands.10 44 45 Comparing estimated parameters to existed evidence The R0 values inferred by our study were lower than some Guangzhou estimates derived from mosquito-abundance indices (MOI, reaching 4.0) and from simpler SIR (Susceptible-Infected-Recovered) formulations (2.2).17 46 Differences likely reflect methodological choices. For example, MOI-based approaches may overweight vector abundance by focusing solely on mosquito-to-human transmission, whereas SIR models omit latent periods and vector life-history constraints. By contrast, our SEIR structure explicitly couples vector and human infection stages and ties transmission to empirically anchored contact dynamics. It finally helps produce a more balanced R0 estimate. Nevertheless, the estimated higher R0 values during the TC effect period were attributed to the increased human–mosquito contact rates. During the cyclone window, our MRR data indicated longer active dispersal by Aedes albopictus, with maximum recapture distance being 75 m beyond the baseline. Previous studies have reported that the maximum flight distance could range between 50 and 150 m, which was in line with the monitoring results in our baseline period.47,49 However, one study reported a maximum flight distance exceeding 900 m. These discrepancies largely reflect methodological and environmental drivers (eg, trap density and spacing, attractants, landscape structure and wind),50 underscoring that dispersal estimates are context-sensitive and design-sensitive. At the city scale, our results are consistent with analyses of Guangzhou’s record 2014 epidemic, which emphasised the interplay of climate, imported infections and timing of interventions; our work adds a mechanism-centred quantification of cyclone perturbations superimposed on those drivers.16 Potential mechanisms By altering key environmental conditions, such as increasing humidity through oceanic moisture and rainfall, TCs affect multiple components of the vectorial capacity equation at once.51 52 In our data, cyclone-period humidity was higher than baseline while temperatures were similar on average; MRR-derived movement and survival increased; and the resulting contact rate rose within the plausible range defined by prior literature and our constraint set. Mechanistically, higher humidity plays a critical role in mosquito ecology, such as by reducing desiccation stress, prolonging adult lifespan and sharpening responsiveness to host cues. These effects finally increase the likelihood that mosquitoes locate warm-blooded hosts,53,57 particularly where the short-range dispersal extends from tens to low hundreds of metres as observed in our study. In our model, these behaviour-linked and survival-linked changes increased the force of infection during cyclone windows and reproduced the faster growth of infectious mosquito and human compartments compared with the baseline. Between the two cyclones, dynamics differed in ways consistent with temperature-sensitive life-history traits and control intensity. For example, the mean temperature was higher during Typhoon Rammasun relative to Kalmaegi (29.5℃ vs 26.2℃). Warmer temperature could provide favourable conditions for mosquito development and virus replication cycle.58 These environmental changes significantly enhanced mosquito activity, resulting in a higher R0 during Rammasun. In contrast, during Typhoon Kalmaegi, the average temperature dropped, slowing down mosquito maturation and reproduction while slightly improving its survival rates. However, the lower temperature also extended the extrinsic incubation period of dengue virus, and the resulted higher human–mosquito contact rate might have prolonged the epidemic if without timely intervention.59 60 Period independence and counterfactual effect attribution Under the no-cyclone counterfactual framework, the predicted local epidemic potential for dengue was negligible during the Rammasun period, suggesting that infections observed within this timeframe were closely associated with cyclone-induced perturbations. This counterfactual model was specifically designed to isolate the impact of variations in human–mosquito contact rates—identified as the primary transmission driver via our MRR field experiments—while keeping all other parameters constant. To account for and eliminate potential lagged carryover effects from the preceding TC, a complete extrinsic incubation period for dengue was used as an independent buffer interval. This buffer separated the termination of the Rammasun effective period from the initiation of the subsequent baseline phase. As all epidemiological and entomological model parameters were independently calibrated and assigned to each discrete study period, the sporadic local cases detected during the baseline phase were indicative of inherent background transmission rather than residual impacts from Typhoon Rammasun. Furthermore, the selection of a 21-day duration for each cyclone effect period was explicitly informed by existing ecological evidence.11,13 Given the exclusion of other unmeasured cyclone-modulated mechanisms, the estimated excess dengue cases are interpreted as scenario-based projections rather than measures of absolute causal attribution. This constraint underscores our methodological focus on quantifying the contribution of the most empirically supported transmission pathway (human–mosquito contact intensity), while recognising the multifaceted nature of cyclone-driven transmission dynamics. Public health implications During Typhoon Kalmaegi, Guangzhou intensified control across the Aedes life cycle, including source reduction for larval habitats and insecticide spraying against adult Aedes. These measures coincided with a faster decline in mosquito stages despite elevated contact pressure. Taken together, it demonstrates that rapid, integrated vector management (IVM) during cyclone and even other extreme weather events can effectively reduce mosquito density and the risk of dengue transmission. In addition to immediate IVM, sustained programmes could also be important. Since the 2014 epidemic, Guangzhou has maintained the BI below five, consistent with substantial reduction in vector density and a corresponding decline in dengue cases.61 62 It suggests that the effectiveness of long-term vector control efforts in preventing epidemic, even under extreme weather conditions. These findings align with the global vector control response (2017–2030) of WHO, which emphasises locally adapted IVM, strong surveillance and community engagement. In cyclone-prone cities, embedding cyclone warnings as explicit triggers of IVM should be recommended. Moving forward, findings from previous studies imply that strong public participation in source-reduction campaigns combined with awareness initiatives may limit container breeding that surges with cyclone-related rainfall.63 Real-time street-level mosquito monitoring maps across Guangzhou further enable microtargeting of IVM based on current conditions.64 Integrating community action with responsive, high-resolution surveillance is therefore essential for breaking transmission chains during cyclone windows. Finally, the economic rationale is compelling. Under Kalmaegi, intensified control was projected to avert 59 360 cases and US$20.91 million costs, illustrating that anticipatory IVM can be cost-saving in cyclone seasons. A precise quantification of intervention costs is limited by the lack of disaggregated expenditure records for the 2014 emergency programme, as these expenses were absorbed into overarching public health emergency funds. Nevertheless, published regional data for comparable Aedes albopictus suppression activities estimate annual intervention costs at US$108–163 per hectare.65 Contextualised within our estimates of averted economic burden, these reference figures validate the highly favourable cost-benefit ratio of implementing cyclone-responsive vector control actions. Strengths and limitations This study has several notable strengths. First, by integrating MRR experiments with empirical evidence, we effectively estimated the human–mosquito exposure rate. This methodology eliminates the infection risks associated with traditional human-baiting techniques, mitigates recall bias inherent in questionnaire-based surveys and provides an accurate representation of real-world human–mosquito interactions. Second, by developing a dynamic model, this study conducted an in-depth analysis of the specific impact of the TC effect period on dengue transmission. These findings offer a critical scientific foundation for understanding how extreme weather events influence the dynamics of dengue transmission. Several limitations should be acknowledged. First, a temporal discrepancy exists between the 2024 MRR entomological parameters and the 2014 dengue incidence data utilised for modelling. The 2014 dataset was specifically selected to optimise model stability given the high incidence rate during that outbreak. While long-term shifts in urban microenvironments may influence human–mosquito contact dynamics, existing evidence confirms that Aedes’ flight and host-seeking behaviours remain evolutionarily conserved over such periods.66 Consequently, the application of 2024 MRR results to parameterise the 2014 transmission dynamics is methodologically sound, serving as the best available proxy, though concurrent data would ideally strengthen future frameworks. Second, the MRR field observations were confined to a single TC, preventing the analysis of storms with varying intensities. Third, the estimated contact rates reflect isolated changes in mosquito behaviour, unadjusted for human-centric variables such as infrastructure degradation and population displacement following a cyclone. As a composite index, the reliance on this contact rate as the singular driver in counterfactual scenarios constitutes a model simplification, restricting spatial extrapolation across different landscape typologies. Finally, the current model structure does not explicitly account for daily precipitation fluctuations and uses laboratory-derived reproduction parameters, potentially introducing bias by underrepresenting the complexity of natural ecological environments. TCs significantly amplify dengue transmissibility by enhancing Aedes albopictus dispersal and accelerating viral transmission cycles under climate-influenced conditions. Strengthening vector control measures and adaptive strategies should be recommended for cyclone-prone regions. Supplementary material Acknowledgements We are grateful to the staff of the Guangdong Provincial Center for Disease Control and Prevention and the Huangge Town Shazai Village Committee (Nansha District, Guangzhou) for their valuable assistance in data collection. Footnotes Funding: This study was supported by the National Natural Science Foundation of China (grant numbers 82073615, 82304200 and 82574161) and the Natural Science Foundation of Shandong Province in China (grant number ZR2023QH319). The funder had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Provenance and peer review: Not commissioned; externally peer reviewed. Handling editor: Luca Nelli Patient consent for publication: Not applicable. Ethics approval: The study was approved by the Ethical Review Committee of the School of Public Health, Shandong University (LL20240709). Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research. Data availability statement Data will be made available from the corresponding author on reasonable request. References Associated Data Supplementary Materials Data Availability Statement Data will be made available from the corresponding author on reasonable request.