Climate Informed Dengue Prediction and Future Risk Under a Changing Monsoon Climate in Kerala, India Authors: Yacob Sophia, Mathew Koll Roxy, Surendranath Bindu Vaisakh, Pallipurathu Raghunathan Nair Sreelakshmi, Thekkumkara Surendran Anish, Pillaveetil Sathyadas Indu Published: 2026 ABSTRACT AbstractDengue fever has emerged as a major public health threat in India, with its burden projected to rise under climate change. This study investigates the influence of climate variability on dengue incidence in Kerala, a dengue hotspot in southern India, using 14 years of climate and dengue data (2006–2019). Temperature, monsoon rainfall and its active‐break phases, relative humidity (RH), and the El Niño–Southern Oscillation are identified as key climatic predictors of dengue transmission in the region. Dengue seasonality in Kerala closely follows its rainfall cycle, with cases rising in May and peaking during the southwest monsoon (June–September), reaching a maximum in June–July. To capture the nonlinear and combined relationships between climatic predictors and dengue incidence, several machine learning (ML) models were trained and evaluated, with the XGBoost regression model showing the best predictive performance (R 2 = 0.72, Normalized Root Mean Square Error = 0.53). Locally, warmer temperatures (30.5–32.5°C) and moderate humidity (66%–79%) during the pre‐monsoon season, together with moderately high rainfall (180–450 mm/month) during the monsoon season, favor high dengue incidence. Remotely, El Niño–like conditions increase dengue incidence by shifting large‐scale circulation toward warmer and drier pre‐monsoon conditions over Kerala, while La Niña–like conditions have the opposite effect, providing about 3 months of potential predictability. Future simulations from Coupled Model Intercomparison Project Phase 6 models project an increase in dengue incidence by 16%–20% in the near future (2021–2040) and 46%–113% in the far future (2081–2100) under low‐to‐high emission pathways. These findings highlight the need for climate‐informed dengue early warning and future risk planning in Kerala.Keywords: dengue, climate change, El Niño, future projections, Kerala, machine learning Introduction Dengue fever is a mosquito‐borne viral disease that has become a major public health concern in tropical and subtropical regions worldwide. The World Health Organisation (WHO) reports that the global dengue cases have increased tenfold over the past two decades, from half a million in 2000 to more than five million in 2019 (WHO, 2024). India reports more than 250,000 dengue cases annually, making it a major contributor to this burden (NCVBDC, 2026). Among Indian states, the southwestern state of Kerala—with its warm and humid climate—has experienced consistently high dengue cases, with annual reported cases exceeding 20,000 in recent years. While the long‐term annual average of reported dengue cases in Kerala is approximately 5,000, exceptional surges were recorded in 2017 (21,993 cases) and more recently during the post‐COVID period in 2023 (16,766 cases) and 2024 (20,088 cases). Dengue cases, including deaths, have been reported in Kerala since 1997, with the disease burden increasing over time. The state is now considered hyper‐endemic, with all four dengue virus serotypes (DENV‐1 to DENV‐4) co‐circulating and occasional co‐infections reported (Karunakaran et al., 2014; Manakkadan et al., 2010). Dengue transmission in Kerala is primarily driven by Aedes aegypti, with contributions from Aedes albopictus also reported in the region (Samuel et al., 2014). Dengue is a climate‐sensitive disease, with its transmission dynamics strongly influenced by environmental factors such as temperature, rainfall, and humidity (Focks et al., 1993; Gagnon et al., 2001; Hales et al., 1996; Koopman et al., 1991). These variables regulate the life cycle, reproduction, and biting behavior of mosquito vectors, as well as the replication of the dengue virus (DENV) within the mosquito (Brady et al., 2013; Chan & Johansson, 2012; Rueda et al., 1990; Russell, 1998). Numerous studies have shown that favorable meteorological conditions, especially during the monsoon season, can lead to increased dengue risk and trigger spikes in dengue cases across India (Shil et al., 2020; Sophia et al., 2025). However, the relationship between climate and dengue is non‐linear, with a lag between favorable climatic conditions and dengue occurrence—driven by delays in mosquito breeding, viral incubation, and human transmission. Temperature influences multiple components of the dengue transmission cycle, including mosquito development, survival, biting rate, and viral replication, all of which exhibit strong nonlinear responses to thermal conditions (Chan & Johansson, 2012; Christophers, 1960; Reinhold et al., 2018). Dengue transmission generally increases within a favorable temperature range but declines beyond critical thresholds due to adverse effects on mosquito survival and viral development. However, the optimal temperature range varies across regions, highlighting the complex and location‐specific nonlinear relationship between temperature and dengue transmission (Fan et al., 2014; Sophia et al., 2025; Xiang et al., 2017). Rainfall also exhibits a complex relationship with dengue. Moderate rainfall can enhance breeding by increasing the availability of water‐filled containers, whereas excessive rainfall may flush out larvae and reduce vector populations (Benedum et al., 2018; Koenraadt & Harrington, 2008). Conversely, low rainfall in urban settings may promote water storage practices, inadvertently creating breeding habitats. These interacting effects, further shaped by local infrastructure and human behavior, result in region‐specific and non‐linear rainfall–dengue relationships. Climate change has emerged as a significant driver of vector‐borne diseases, including dengue, by altering ecological and climatic conditions that impact mosquito proliferation and virus transmission (Epstein, 2000; Pörtner et al., 2022). Rising global temperatures, shifting precipitation patterns, and increasing humidity are creating more favorable environments for Aedes mosquitoes, particularly in regions that were previously less conducive to dengue transmission (Bhattacharya et al., 2006; Dhiman et al., 2010). A recent study conducted in urban Pune using climate change projections from the Coupled Model Intercomparison Project Phase 6 (CMIP6) models found that dengue risk could increase by 12%–112% under low‐to‐high future emission scenarios, primarily due to the projected rise in mean temperature (Sophia et al., 2025). Extending this perspective, a global analysis across 122 countries documented a significant rise in dengue incidence since 1990, with stronger temperature–dengue associations observed above 21°C, and projections under CMIP6 scenarios indicating a substantial expansion of year‐round endemic regions by 2100 under high‐emission pathways (Feng et al., 2024). Consistent with these findings, another study in Guangdong, China, revealed that climate change is not only intensifying dengue outbreaks but also extending transmission periods, shifting them from seasonal to nearly year‐round in some regions (Xu et al., 2024). In addition to global warming, large‐scale climate modes such as the El Niño–Southern Oscillation (ENSO) are also projected to intensify in the future (Fredriksen et al., 2020; Heede & Fedorov, 2023). Evaluating how future changes in these climate modes will alter the local environmental conditions is crucial for anticipating shifts in dengue risk and informing public health preparedness. Kerala's unique climatic conditions—characterized by the dual influence of the southwest and northeast monsoons, persistently high humidity, orographic effects from the Western Ghats, and proximity to the Arabian Sea—create ideal conditions for year‐round mosquito breeding and elevated dengue risk. Given the growing threat of climate change on human health and its potential to exacerbate vector‐borne diseases, it is critical to assess how future changes in temperature, rainfall, humidity, and ENSO variability may alter the burden of dengue in Kerala. While ENSO has been identified as a potential driver of dengue variability in many global studies (Andhikaputra et al., 2023; Fuller et al., 2009; Poveda et al., 2000), its role in shaping dengue risk at the regional scale, particularly in Kerala, remains insufficiently explored. In this study, we aim to (a) identify key climatic predictors of dengue incidence over Kerala along with their relative importance and optimal lag periods, (b) develop a predictive dengue model using machine learning (ML) techniques, (c) assess future dengue incidence using CMIP6 climate projections under three Shared Socioeconomic Pathways (SSP): SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5, and (d) investigate the role of long‐term climate trends and ENSO variability in shaping the future dengue incidence in the region. The results from the current study provide important insights that can enhance prevention strategies, inform policymaking, and support climate‐adaptive public health planning, aiding in the mitigation of dengue outbreaks in a warming climate. Data Kerala is a state located in the southwestern part of India, bordered by the Arabian Sea to the west and the Western Ghats to the east. Covering an area of approximately 38,863 km2, the state has a population of about 35.5 million as of 2023 (Annual vital statistics report, 2023). The state features diverse landscapes, including coastal plains, hilly regions, and highlands (H. C. Nair et al., 2018). It receives the first spell of rainfall during the southwest monsoon season over India in early June (Ananthakrishnan & Soman, 1988) and acts as the “Gateway of Summer Monsoon.” The southwest monsoon typically sets in over Kerala in June and lasts until September, followed by the northeast monsoon from October to November. Observed Climate and Health Data The current study uses monthly dengue case data for 2006–2019 obtained from the annual reports of the State Surveillance Unit, Directorate of Health Services, Government of Kerala. Data from 2020 to 2022 were excluded because the surveillance and reporting during these years were affected by the COVID‐19 pandemic due to reduced case detection, underreporting, and altered mobility patterns (Mayilsamy et al., 2023; Phadke et al., 2021). To ensure consistency, the current study utilized only the continuous pre‐pandemic monthly state‐wise data set spanning 2006–2019. On average, Kerala reports around 5,000 dengue cases annually; however, 2017 recorded an anomalously high count of 21,993 cases. More recently, the post‐COVID years 2023 and 2024 also reported elevated case numbers of 16,766 and 20,088, respectively. For calculating dengue per lakh population which is the target variable in the dengue model developed for Kerala, the annual population data for Kerala was obtained from the Vital Statistics Department of Kerala. To study the climate‐dengue associations in Kerala, the current study uses high‐resolution daily gridded rainfall (0.25° × 0.25° grid) (Pai et al., 2014) and temperature (1° × 1° grid) data from the India Meteorological Department (IMD) (Srivastava et al., 2009). Relative humidity (RH) data (0.25° × 0.25° grid) covering the same period as the health data (2006–2019), were obtained from the ERA5 reanalysis data set due to the unavailability of gridded observational RH data from IMD. While ERA5 provides spatially and temporally consistent estimates, it may contain regional biases, and the lack of independent ground‐based RH data has limited its bias‐correction. For each variable, values were spatially averaged over the state of Kerala (8°–12.5°N, 74.5°–77.5°E). In addition to meteorological variables, the Niño 3.4 sea surface temperature (SST) and the extrinsic incubation period (EIP) were also included as input predictors while developing the dengue model for Kerala. The Niño 3.4 SST is calculated as the average SST over the central equatorial Pacific region (5°N–5°S, 170°W–120°W) and is widely used to monitor the ENSO and its influence on climate variability. Niño 3.4 SST anomalies were obtained from the HadISST data set (1870–2023) to assess the long‐term influence of ENSO on dengue‐related climatic predictors in Kerala, while for dengue modeling, values corresponding to the same period as the dengue data (2006–2019) were used. The EIP refers to the time required for the dengue virus to complete its development inside the mosquito vector before it can be transmitted to humans. It is a temperature‐sensitive epidemiological parameter, with shorter incubation periods observed at higher temperatures. In this study, EIP was estimated for Kerala using the following empirical equation derived from controlled laboratory experiments on vector‐virus interactions (McLean et al., 1974; Mutheneni et al., 2017; Shil et al., 2020): where T denotes the mean ambient temperature in degrees Celsius. For calculating the EIP, we used the mean temperature at a lag of 3 months, in correspondence with the optimal lag for mean temperature with respect to dengue incidence. Climate Change Projections From CMIP6 The current study utilizes historical simulations (1900–2014) and future projections (2015–2100) from Coupled Model Intercomparison Project Phase 6 (CMIP6) models to generate dengue future projections for the state of Kerala. Climate change projections from CMIP6 models provide multi‐scenario estimates of future climate under varying greenhouse gas emission pathways. Future dengue incidence is assessed under three distinct SSPs, which represent alternative scenarios of socioeconomic development and associated greenhouse gas emissions. These include: SSP1‐2.6, a sustainability‐focused low‐emission pathway where radiative forcing is limited to 2.6 W m−2 by the end of the 21st century; SSP2‐4.5, a “middle‐of‐the‐road” scenario with moderate emissions and radiative forcing stabilizing at 4.5 W m−2; and SSP5‐8.5, a high‐emission scenario characterized by fossil fuel‐driven development and radiative forcing reaching 8.5 W m−2 by 2100 (J.‐Y. Lee et al., 2021). While SSP5‐8.5 may be considered as a low‐likelihood pathway and should not be interpreted as the most‐likely future emission scenario (Hausfather & Peters, 2020), its inclusion facilitates comparison with existing literature while also providing an upper‐bound estimate of potential future dengue risk (Feng et al., 2025; Sophia et al., 2025; Y. Wang et al., 2024). Rather than relying on a simple multi‐model ensemble mean of all available models, a subset of CMIP6 models that demonstrated higher skill in simulating Indian summer monsoon rainfall was selected to ensure more robust and regionally relevant dengue projections for Kerala. Sophia et al. (2025) evaluated 15 CMIP6 models for their ability to spatially represent rainfall climatology and intraseasonal oscillations over the Indian region and identified eight models with relatively higher skill. However, two of these models were excluded in the current study due to missing daily maximum and minimum temperature data across all three emission scenarios (SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5). As a result, six models (listed in Table S1 in Supporting Information S1) were selected for dengue projection analysis over Kerala under the three future emission scenarios. Methodology ML models have gained widespread use in disease modeling due to their ability to capture the non‐linear and cumulative effects of multiple predictor variables on disease incidence. Some of the most effective and widely used ML algorithms for health‐related applications include Random Forest Regression (RFR), eXtreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Long Short‐Term Memory (LSTM) networks (An et al., 2023; Kakarla et al., 2023; Sophia et al., 2025; Yanes et al., 2024). RFR is an ensemble tree‐based method that builds multiple independent decision trees using random subsets of data and predictors, and averages their outputs to reduce variance and improve robustness (Breiman, 2001; Segal, 2004). XGBoost also relies on decision trees but differs by adopting a boosting strategy, where trees are built sequentially to correct errors from previous iterations, often achieving higher predictive accuracy at the cost of greater sensitivity to hyperparameter tuning (Bhati et al., 2021; Friedman, 2001; Sahin, 2020). In contrast to these tree‐based models, LSTM networks are designed to explicitly capture temporal dependencies in sequential data through gated memory cells, making them particularly suitable for time‐series forecasting in climate–disease studies (Hewamalage et al., 2021; Hochreiter & Schmidhuber, 1997). SVR takes a fundamentally different approach by constructing an optimal regression hyperplane within an ε‐insensitive margin, allowing it to model both linear and non‐linear relationships using kernel functions (Cortes & Vapnik, 1995; Drucker et al., 1999). In this study, we evaluated all four algorithms using three standard metrics—R 2 score, Normalized Root Mean Square Error (NRMSE), and Mean Absolute Error (MAE) and selected the one with the highest predictive skill as the final dengue model for Kerala. R 2 score indicates the proportion of variance in observations explained by the model while NRMSE provides a normalized measure of overall error, giving greater weight to larger deviations and enabling comparison across data sets. MAE quantifies the average magnitude of prediction errors. In this evaluation, higher R 2 values and lower values of NRMSE and MAE indicate superior predictive performance. While there is no universal threshold for what constitutes “good” performance, previous studies suggest that NRMSE values below 1 can be considered indicative of satisfactory model performance (Long et al., 2025), while values less than 0.1 indicate excellent performance (Amin et al., 2025). The climate‐dengue modeling framework adopted in the current study is based on the approach described in Sophia et al. (2025) and involves multiple steps including the identification of potential dengue predictors using statistical analysis, model training, hyperparameter optimization, k‐fold cross‐validation, and model performance evaluation using test data. The model was trained using climate and health data from 2006 to 2016 (approximately 80% of the total data) and tested on data from 2017 to 2019 (20% of the total data). To account for the temporal changes in the population size, the target variable in the dengue model is dengue cases per lakh population, which was transformed using the natural logarithm to stabilize variance and minimize the influence of outliers. This transformation also helps approximate a normal distribution, thereby improving model performance (D. K. Lee, 2020). For the XGBoost model, hyperparameter optimization was performed using the Optuna framework. The tuned hyperparameters included the number of trees (n_estimators), maximum tree depth (max_depth), learning rate (learning_rate), minimum child weight (min_child_weight), subsampling ratio (subsample), column subsampling ratio (colsample_bytree), and gamma (minimum loss reduction required for a split). Since the dengue data set consisted of monthly time‐series observations, model evaluation during optimization was conducted using a time‐aware five‐fold cross‐validation approach implemented through an expanding‐window scheme. In this approach, the training set progressively increased with each fold, while the subsequent temporal block served as the validation set, thereby ensuring that training data always preceded validation data and preventing temporal information leakage. Model selection was based on minimizing cross‐validation error on the training data set. Random seeds were fixed during optimization and model fitting to ensure reproducibility. Similarly, the hyperparameter optimization for the RFR and Support Vector Regression (SVR) models was performed using Optuna in conjunction with the same time‐aware five‐fold cross‐validation framework. For RFR, the tuned hyperparameters included the number of trees (n_estimators), maximum tree depth (max_depth), minimum number of samples required to split an internal node (min_samples_split), minimum number of samples required at a leaf node (min_samples_leaf), maximum number of features considered at each split (max_features), and the bootstrap sampling option (bootstrap). For SVR, the tuned hyperparameters included the regularization parameter (C), epsilon (epsilon), kernel type (kernel), and gamma parameter (gamma). The LSTM model was also optimized using the Optuna framework, with the number of units in the first and second LSTM layers (units_1 and units_2), dropout rates (dropout_1 and dropout_2), learning rate (learning_rate), and batch size (batch_size) treated as tunable parameters. Early stopping was employed during training to mitigate overfitting. Thus, all ML models considered in this study were systematically optimized to ensure a fair comparison of model performance. To account for lagged effects between climate variables and dengue incidence, cross‐correlation analysis was performed using the training data set (2006–2016). For each predictor, we identified the lag at which its correlation with dengue incidence was maximum (termed the optimal lag). In the final model, predictor variables were provided at their optimal lag along with all preceding lags (i.e., from lag 0 up to the optimal lag). Further, following model development and evaluation, the finalized dengue model was applied to the complete observational data set (2006–2019), and the resulting predictions were analyzed to identify the climatic conditions associated with increased dengue incidence over Kerala. This approach enabled the interpretation of the model predictions and the identification of climatic ranges favorable for higher dengue incidence using the maximum available observational record. Based on the predicted dengue values, months were grouped into high dengue (above the 75th percentile) and low dengue (below the 25th percentile) categories and the climatic conditions corresponding to both categories were studied. Future Dengue Projections Historical simulations (1900–2014) and future projections (2015–2100) from the selected CMIP6 models were spatially averaged over the Kerala region (8°–12.5°N, 74.5°–77.5°E) and provided as inputs to the dengue model to obtain future dengue projections for Kerala. Specifically, climate variables from each selected CMIP6 model for the historical period (1900–2014), along with future scenario simulations (2015–2100), were used as inputs to the Kerala dengue model (XGBoost), and the resulting outputs represent historical and projected dengue incidence for each model. To evaluate the individual contributions of future trends in climate variables to projected dengue incidence, sensitivity experiments were performed by modifying the CMIP6 projections of key climatic predictors. In the first experiment, each predictor was detrended individually while keeping the others unchanged, thereby isolating the influence of long‐term trends in that variable. In the second experiment, extreme rainfall events above the 90th percentile (relative to the reference period 1995–2014) were removed to evaluate their influence on projected dengue incidence. The percentage change in projected dengue incidence resulting from each experiment was then computed relative to the unaltered future dengue projections, providing insights into how long‐term trends in individual climate variables and extreme rainfall influence the future dengue incidence in Kerala. Impacts of ENSO on Dengue in Kerala ENSO is a major driver of large‐scale climate variability over India, and its teleconnection with the Indian summer monsoon has shown strong, time‐varying, and regionally distinct effects (Athira et al., 2023). ENSO is found to modulate rainfall, temperature, and humidity patterns across India—key environmental factors governing dengue transmission (Athira et al., 2023; Chowdary et al., 2014; Kothawale et al., 2010). ENSO acts as a large‐scale climatic modulator that can amplify or suppress local environmental conditions favorable for dengue transmission (Amarakoon et al., 2008; Andhikaputra et al., 2023; Fuller et al., 2009; Tipayamongkholgul et al., 2009; Vincenti‐Gonzalez et al., 2018). The current study investigates the influence of the ENSO on dengue transmission in Kerala using both observational data and outputs from selected CMIP6 models. A preliminary analysis was conducted using observed SST and dengue incidence data from 2006 to 2019. However, due to the limited length of the observed data set, which restricts robust statistical inference, we extended the analysis using historical simulations (1900–2014) from the six selected CMIP6 models. To isolate the impacts of ENSO on dengue incidence, all climatic input variables from the CMIP6 models were detrended (to remove the long‐term climate change signal) and provided as inputs to the dengue prediction model, and the resulting dengue projections were analyzed. In addition, we examined the local weather conditions over Kerala—specifically temperature, rainfall, and RH—during El Niño and La Niña years. El Niño and La Niña years were identified based on the Niño 3.4 SST anomaly, calculated from the HadISST data set (1870–2023) for observational analysis and from the CMIP6 historical simulations (1900–2014) for model‐based analysis. In our analysis, a year is classified as an El Niño year if the 3‐month running mean of the Niño 3.4 SST anomaly exceeds +0.5°C for at least five consecutive overlapping seasons. Similarly, a year is classified as a La Niña year if the 3‐month running mean of the Niño 3.4 SST anomaly remains below −0.5°C for at least five consecutive overlapping seasons, following the NOAA Oceanic Niño Index (ONI) criteria. Results Seasonal Patterns and Climatic Predictors of Dengue in Kerala Dengue occurrence in Kerala exhibits a distinct seasonal pattern, with most cases reported during the summer monsoon months from late May to September, peaking in June. These months account for approximately 60% of the total dengue cases reported in the state. Figure 1 presents a comparison between monthly rainfall and dengue cases over the study period (2006–2019), showing that the seasonality of dengue closely aligns with the rainfall climatology of Kerala, with strong concurrence to the monsoon rainfall cycle. Monthly cumulative rainfall (blue line) and dengue cases (black line) in Kerala from 2006 to 2019. Dengue cases show a clear seasonal pattern that corresponds with the rainfall cycle. To identify the key predictors of dengue in Kerala, we considered a broad set of variables from previous studies and only those that improved the model skill were retained in the final dengue model. Selected climatic predictors included mean, maximum, and minimum temperature, diurnal temperature range, rainfall, RH, number of active and break days of monsoon rainfall, Niño 3.4 SST, EIP, and month of the year (Kakarla et al., 2023; Sophia et al., 2025). Field studies worldwide have identified varied lags from 0 to 6 months between different climatic variables and dengue incidence. Table 1 presents the optimal lags identified between each predictor and dengue incidence in Kerala. These lags range from zero for rainfall to up to 5 months for variables such as diurnal temperature range and RH. Except for RH, all other variables show a positive correlation with dengue incidence in Kerala at their optimal lags, indicating that increases in these variables are associated with higher dengue incidence. In contrast, RH exhibits a negative relationship, with higher humidity reducing dengue incidence and lower humidity increasing it, and its influence on dengue incidence is strongest at a lag of 5 months. To further understand how the non‐linear and combined interactions of these predictors influence dengue transmission, we developed an ML‐based dengue model for Kerala. Table 1Climatic Predictors of Dengue in Kerala, Along With the Time Lag (in Months) at Which Their Pearson Correlation With Dengue Incidence Is Maximum (Denoted as Optimal Lag) and the Corresponding Correlation ValueClimatic predictorOptimal time lag (months)Correlation at optimal time lagMean temperature (T mean)30.63Maximum temperature (T max)30.58Minimum temperature (T min)20.59Diurnal range in temperature (DTR)50.46Mean rainfall00.43Active days of monsoon rainfall00.3Break days of monsoon rainfall00.26Relative humidity (RH)5−0.52Niño 3.4 sea surface temperature (SST)30.38 Note. All correlations are statistically significant (p < 0.05). Climatic Predictors of Dengue in Kerala, Along With the Time Lag (in Months) at Which Their Pearson Correlation With Dengue Incidence Is Maximum (Denoted as Optimal Lag) and the Corresponding Correlation Value Note. All correlations are statistically significant (p < 0.05). Dengue Model for Kerala For dengue prediction in Kerala, XGBoost emerged as the best‐performing model among the tested ML models, achieving the highest R 2 score (0.72), along with the lowest NRMSE (0.53) and MAE (0.19), indicating strong predictive accuracy. In comparison, SVR, RFR, and LSTM showed similar and comparatively lower performance, with R 2 scores ranging between 0.48 and 0.52 and NRMSE values around 1.0 (Table S2 in Supporting Information S1). The optimal XGBoost configuration obtained through hyperparameter optimization consisted of n_estimators = 411, max_depth = 2, gamma = 0.0025, learning_rate = 0.1867, subsample = 0.5757, colsample_bytree = 0.6466, and min_child_weight = 10. Figure 2 shows the comparison between actual dengue incidence and the dengue incidence predicted by the selected XGBoost model for the test period (2017–2019). The corresponding time series, along with the climatic predictors at their optimal lags, is presented in Figure S3 in Supporting Information S1. Comparison between actual dengue incidence (X‐axis) and the dengue incidence predicted by the XGBoost model (Y‐axis) for the test period from 2017 to 2019. The dotted red line represents the line of perfect agreement where predictions equal actual dengue incidence. To explore alternative model formulations, we conducted supplementary experiments on the selected dengue model, including (a) incorporating R0 (basic reproduction number, defined as the average number of new infections generated by one infected individual in a fully susceptible population) as an additional predictor, calculated following Mordecai et al. (2017), (b) including previous dengue cases as an additional predictor, and (c) using only variables at their optimal lags. The results of these analyses are provided in Table S5 in Supporting Information S1. The model incorporating R0 and the model using only variables at optimal lags showed lower predictive skill than the current dengue model. Although the inclusion of previous dengue cases substantially improved model performance, it also increased dependence on autoregressive terms, reducing the relative contribution of climatic variables and limiting the interpretability of climate–dengue relationships (Table S5 in Supporting Information S1). To interpret the contributions of individual predictors to the model predictions, we used the SHapley Additive exPlanations (SHAP) method. SHAP is a game‐theoretic approach that quantifies the contribution of each feature to a particular prediction by computing Shapley values, ensuring a consistent and locally accurate explanation of the model predictions. Variable importance represents the extent to which a predictor influences or contributes to the model's predictive performance. According to the variable importance calculated using the SHAP method, temperature and rainfall are the most influential predictors for the dengue model (Figure S1 in Supporting Information S1). To account for delayed climatic effects, each predictor was included in the model at several time lags, ranging from lag 0 to its optimal lag. The resulting variable importance ranking therefore reflects the influence of individual lagged predictors (e.g., maximum temperature at lag 3 or rainfall at lag 0), rather than the overall contribution of a given variable. Figure S1 in Supporting Information S1 shows the top 20 most influential lagged predictors. Figure 3 shows box‐and‐whisker plots comparing the climatic ranges at the optimal lags corresponding to high and low dengue months. For example, if July is identified as a high dengue month, the associated climatic predictors are examined at the optimal lag for each variable, such as maximum temperature at a 3‐month lag corresponding to April, minimum temperature at a 2‐month lag corresponding to May, and so on. Box‐and‐whisker plots showing the distribution of climatic predictors at their respective optimal lags corresponding to low and high dengue months in Kerala. Boxes represent the interquartile range (25th–75th percentiles), the horizontal line denotes the median, whiskers extend to the 5th and 95th percentiles, and individual points indicate outliers. The plots compare the ranges of maximum temperature, minimum temperature, mean temperature, diurnal temperature range, mean rainfall, relative humidity, and Niño 3.4 sea surface temperature associated with high and low dengue incidence in Kerala. The high dengue months in Kerala are the monsoon months from June to August, whereas low dengue months include different months across years. Our results show that high dengue incidence during the monsoon season in Kerala is associated with higher monsoon rainfall, and moderate RH, warmer temperatures, and a warmer Niño 3.4 SST in the preceding months at the specific optimal lag periods identified for each climatic predictor (Figure 3). In contrast, lower dengue incidence is linked to reduced concurrent rainfall, along with comparatively cooler temperatures, higher RH, and a cooler Niño 3.4 SST at their respective lags. Warmer Niño 3.4 SSTs indicate El Niño–like conditions, while cooler Niño 3.4 SST values indicate La Niña–like conditions (Figure 3). For Kerala, the optimal lag for temperature variables (t mean, t max, t min, and DTR) ranges from 2 to 5 months, RH shows an optimal lag of 5 months, while rainfall and the Niño 3.4 SST are found to have their optimal influence on dengue incidence at lags of 0 and 3 months, respectively (Table 1). Monthly mean rainfall during monsoon months in Kerala spans a wide range of 81–996 mm, within which a narrower range of 180–457 mm is associated with higher dengue incidence. Similarly, optimal ranges of climatic variables associated with higher dengue incidence in Kerala, along with the corresponding overall mean ranges observed during the monsoon season of the study period (2006–2019) are presented in Table S6 in Supporting Information S1. Overall, the model interpretation suggests that increased dengue incidence over Kerala is associated with warmer temperatures and moderate RH during the pre‐monsoon months (March, April, and May), together with increased rainfall during the monsoon season (Figure 3). In addition, the results suggest the possible role of large‐scale climate variability in influencing regional climatic conditions that affect dengue transmission in Kerala, with El Niño conditions in the Pacific potentially contributing to favorable local weather conditions for increased dengue incidence over Kerala. Impacts of ENSO on Dengue Incidence in Kerala Several studies have identified the El Niño–Southern Oscillation (ENSO) as a key climatic driver of dengue risk (Andhikaputra et al., 2023; Fuller et al., 2009). ENSO influences regional climate by altering variables such as temperature, rainfall, and RH, which in turn affect mosquito dynamics and dengue transmission. In the current study, we find a statistically significant positive correlation (r = 0.4) between monthly dengue incidence in Kerala and Niño 3.4 SST at an optimal lag of 3 months. However, the limited length of the observational data set (2006–2019) restricts our ability to draw robust and conclusive inferences. To strengthen the analysis, we leveraged the dengue model developed for Kerala along with historical climate change simulations (1900–2014) from the six selected CMIP6 models to further investigate the ENSO–dengue relationship in Kerala (sample size = 8,280). Figure 4 presents a scatter plot of normalized dengue predictions against the normalized Niño 3.4 index (representing ENSO variability) from all six selected CMIP6 models. Red and blue markers represent high and low dengue years, respectively. Normalization was applied to both variables to account for differences in the range of values across the individual models. El Niño and La Niña years were identified separately for each model rather than from the ensemble mean. The scatter plot shows that, in Kerala, most years with higher values of Niño 3.4 index correspond to high dengue incidence. However, not all high dengue years have high Niño 3.4 index values, suggesting that ENSO is an important—but not exclusive—driver of dengue variability. We also find that, on average, 40% of high dengue years coincide with El Niño conditions, whereas none of the high dengue years correspond to La Niña conditions. In contrast, 57% of low dengue years are associated with La Niña conditions, while none of the low dengue years coincide with El Niño (Table S3 in Supporting Information S1). The lack of a one‐to‐one correspondence between ENSO and dengue incidence in Kerala may also reflect the potential role of other large‐scale climate modes, such as the Indian Ocean Dipole (IOD), Pacific Decadal Oscillation (PDO), and Atlantic Multidecadal Oscillation (AMO), which can modulate local weather conditions and potentially affect dengue dynamics in Kerala. Scatter plot of normalized dengue predictions (X‐axis) versus the normalized Niño 3.4 index (Y‐axis) from six selected CMIP6 models. Red and blue markers indicate high and low dengue years, respectively, while gray markers represent all other years. The dotted horizontal line denotes a normalized Niño 3.4 index of 0.5 for reference. To further understand how ENSO is linked to dengue incidence in Kerala, we examined the local climatic conditions in Kerala during El Niño and La Niña years. Figure 5 shows composite anomalies of maximum temperature, mean temperature, rainfall, and RH over Kerala for El Niño years, La Niña years, and the difference between El Niño and La Niña years. Statistically significant regions (p < 0.05), identified using the Monte Carlo method, are marked with black dots. Our results indicate that El Niño years are associated with higher mean and maximum temperatures and lower rainfall and RH across Kerala, according to the selected CMIP6 models. In contrast, La Niña years correspond to lower mean and maximum temperatures, higher RH, and increased rainfall over Kerala. Similar patterns have been reported in previous studies (Athira et al., 2023; Chakravorty et al., 2016; Chowdary et al., 2014; Kothawale et al., 2010). Composites of anomalies of (a, e, i) maximum temperature, (b, f, j) mean temperature, (c, g, k) rainfall, and (d, h, l) relative humidity over Kerala during La Niña years (a–d), El Niño years (e–h), and the difference between El Niño and La Niña years (i–l). Statistically significant regions (p < 0.05), indicating spatial consistency across the six selected CMIP6 models, identified using the Monte Carlo method, are marked with black dots. We also compared these ENSO‐related climatic patterns over Kerala using observed SST, temperature, rainfall, and RH data from 1980 to 2020 (Figure S2 in Supporting Information S1). The observed patterns mostly agree with the model results, although the ENSO impact on rainfall is not statistically significant throughout Kerala. Dengue Future Projections for Kerala Climate change is projected to alter temperature, rainfall patterns, and large‐scale climate variabilities such as ENSO and IOD (Fredriksen et al., 2020; B. Wang et al., 2021; G. Wang et al., 2024), with regionally diverse impacts on dengue transmission dynamics. We studied the future trends (2015–2100) in the key climatic predictors of dengue obtained from the CMIP6 models. The results indicate statistically significant warming trends (p < 0.05) in both temperature and Niño 3.4 SST across all emission scenarios (Table 2). Mean temperature in Kerala is projected to rise by 0.91°C per century under SSP1‐2.6, 2°C per century under SSP2‐4.5, and 4.4°C per century under SSP5‐8.5, during 2015–2100. Similarly, the Niño 3.4 SST shows increasing trends of 1°C, 2.2°C, and 4.2°C per century under SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5, respectively. Rainfall shows a statistically significant positive trend only in the high emission scenario (SSP5‐8.5), with a trend of 31 mm per century, while RH does not exhibit a statistically significant (p < 0.05) trend in any of the emission scenarios (Table 2). Table 2Future Trends (per Century) in Meteorological Predictors of Dengue in Kerala From 2015 to 2100 Under the Shared Socioeconomic Pathways (SSP), SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5VariablesTrend/century during 2015–2100SSP1‐2.6SSP2‐4.5SSP5‐8.5Temperature (°C/century)0.91*2*4.4*Rainfall (mm/century)212431*Relative humidity (%/century)0.40.3−0.2Niño 3.4 SST (°C/century)0.99*2.2*4.2* Note. Trend values are based on the ensemble mean of six selected CMIP6 models. Statistically significant trends (p < 0.05) are marked with an asterisk (*). Future Trends (per Century) in Meteorological Predictors of Dengue in Kerala From 2015 to 2100 Under the Shared Socioeconomic Pathways (SSP), SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5 Note. Trend values are based on the ensemble mean of six selected CMIP6 models. Statistically significant trends (p < 0.05) are marked with an asterisk (*). To understand how the projected changes in the climatic predictors may affect future dengue incidence in Kerala, we utilized the climate change projections from the selected six CMIP6 models along with the dengue model developed for Kerala. Historical simulations (1900–2014) and future projections (2015–2100) of the climatic predictors were provided at different lags from lag‐0 to the optimal lag of the climate predictors as inputs into the dengue model. The resulting future projections of dengue from the dengue model developed for Kerala are presented in Figure 6a. The projections suggest a substantial increase in dengue incidence across all emission scenarios (SSPs) (Figure 6a). In the near future (2021–2040), dengue incidence in Kerala is projected to increase by 16%–20% relative to the reference period (1995–2014) (Figure 6b). This increase becomes more pronounced in the mid‐century (2041–2060), with projected rises of 32% under SSP1‐2.6, 33% under SSP2‐4.5, and 44% under SSP5‐8.5. By the late century (2081–2100), dengue is projected to rise by up to 113% under SSP5‐8.5, while the projected increases under SSP1‐2.6 and SSP2‐4.5 are 46% and 63%, respectively (Figure 6b). However, these projections represent changes in climate‐driven dengue incidence and do not account for other factors such as vector control interventions, population growth, or changes in immunity in the future, which may also influence future transmission. (a) Dengue incidence anomaly projections for Kerala for the historical (1950–2014) and the future (2015–2100) under SSP1 (green line), SSP2 (blue line), and SSP5 (red line). The mean value for the reference period (1995–2014; light gray bar) was subtracted from the projected values. Multi‐model ensemble means were computed using a 20‐year smoothed time series. Shaded regions represent uncertainty as ±1 standard deviation across the selected six models, with the boundaries of model spread marked by circles of corresponding colors on the right side of the figure. (b) Box‐and‐whiskers plot of projected percentage changes in dengue incidence over Kerala for the near future (2021–2040), mid‐future (2041–2060), and far‐future (2081–2100) under each emission pathway (SSP1, SSP2, and SSP5), relative to the reference period (1995–2014). Boxes represent the interquartile range (25th–75th percentiles), the horizontal line indicates the median, whiskers show the 5th and 95th percentiles, and individual points are outliers. The mean percentage changes are annotated above the boxes. Uncertainty in the dengue future projections arises from both inter‐model variability (differences among the six CMIP6 models; Figure 6a shading) and scenario uncertainty (differences between emission pathways; Figure 6b). In the near future (2021–2040), both sources of uncertainty are relatively low, resulting in higher confidence in projections. In the mid‐century (2041–2060), both inter‐model and scenario uncertainties increase compared to the near term, leading to a moderate spread. By the late century (2081–2100), both sources are highest, with scenario uncertainty becoming dominant due to strong divergence between emission pathways, and inter‐model spread also reaching its maximum, resulting in the largest overall uncertainty and reduced confidence in long‐term projections. To investigate the contribution of trends in climatic predictors (Table 2) to the projected dengue incidence in Kerala, we conducted a series of sensitivity experiments. In these experiments, the future projections (2015–2100) of each climate variable were individually detrended, keeping the others unchanged, and the resulting changes in dengue projections were analyzed. The results indicate that increasing trends in mean temperature, maximum temperature, and Niño 3.4 SST contribute significantly to the projected rise in dengue incidence in Kerala (Figure 7). Specifically, the trend in mean temperature accounts for a 1%–4% increase, while the trend in Niño 3.4 SST contributes a 1%–5% increase in projected dengue incidence across the emission scenarios over the period 2021–2100. The trend in maximum temperature has the highest impact, contributing to a 4%–22% increase across low‐to‐high emission pathways (Figure 7). Percentage change in dengue projections for Kerala under the emission scenarios (a) SSP1‐2.6, (b) SSP2‐4.5, and (c) SSP5‐8.5 for the near‐future (2021–2040), mid‐future (2041–2060), and far‐future (2081–2100). Changes were simulated by removing trends in mean temperature (red), rainfall (blue), Niño 3.4 sea surface temperature (yellow), maximum temperature (green), and by removing all extreme rainfall (maroon) from selected CMIP6 model outputs. Extreme rainfall refers to values above the 90th percentile of the reference period (1995–2014). Percentage changes are relative to projections using unaltered climatic predictors. We also conducted an experiment to assess the influence of extreme rainfall events on future dengue incidence, as previous studies suggest that intense rainfall can flush out mosquito eggs and larvae, thereby reducing dengue risk (Benedum et al., 2018; Sophia et al., 2025). In this experiment, the rainfall projections from the selected CMIP6 models were modified by removing all extreme rainfall events—defined as rainfall exceeding the model‐specific 90th percentile of the reference period (1995–2014). The modified rainfall time series was then used as input to the dengue model, keeping other variables unchanged. The results indicate that eliminating extreme rainfall events leads to a 3%–7% increase in projected dengue incidence across 2021–2100 (Figure 7, maroon bar). These findings suggest that future extreme rainfall events can suppress dengue transmission to some extent, thereby potentially offsetting the projected increase in dengue incidence by 3%–7% for Kerala. In contrast, removing the long‐term rainfall trend had negligible impact, indicating that the rainfall–dengue relationship is nonlinear, where extreme rainfall influences dengue transmission differently from non‐extreme rainfall. Recent studies suggest that ENSO is likely to intensify under future climate change scenarios (Fredriksen et al., 2020; Heede & Fedorov, 2023). Using the six CMIP6 models selected in this study, we evaluated projected changes in the magnitude of the Niño 3.4 index relative to historical values across three emission scenarios: SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5. Our results indicate a consistent intensification of both El Niño and La Niña events in the future. During El Niño years, the Niño 3.4 SST anomaly increases from 0.80°C in the historical period to 0.85°C under SSP1‐2.6. This intensification continues under SSP2‐4.5 and SSP5‐8.5, with anomalies reaching 0.91°C and 0.95°C, respectively (Table S4 in Supporting Information S1). Similarly, La Niña events also intensify. The mean Niño 3.4 index during La Niña years is −0.85°C in the historical period, which strengthens to −0.92°C, −0.98°C, and −1.10°C under SSP1‐2.6, SSP2‐4.5, and SSP5‐8.5, respectively (Table S4 in Supporting Information S1). Given that ENSO has been identified as a significant predictor of dengue incidence in Kerala, these findings imply that the projected intensification of ENSO may lead to an increased risk of dengue in Kerala in the future. Discussion and Conclusion The present study provides a comprehensive assessment of dengue seasonality, climatic drivers, and future risk projections in Kerala under a changing climate. By integrating observational data with ML methods, we demonstrate that regionally tailored models can effectively capture the nonlinear and combined interactions between climatic variables and dengue transmission. Our analysis identifies temperature, rainfall, RH, and ENSO as key drivers of dengue in Kerala, highlighting the potential for climate change to elevate dengue incidence in the future. Previous studies have emphasized the importance of location‐specific climate–dengue associations. For instance, dengue incidence in Delhi was positively correlated with temperatures of 30–39°C (Sahay, 2018), whereas a meta‐analysis of 33 studies reported increased incidence at 23.2–27.7°C (Fan et al., 2014). Other studies found wider optimal temperature ranges, such as 21.6–32.9°C in Guangzhou and 27–35°C in Pune (Sophia et al., 2025; Wu et al., 2018; Xiang et al., 2017). These findings underscore the necessity of developing region‐specific dengue models for early warning and future projections. Our results confirm strong seasonality of dengue in Kerala, with nearly 60% of cases occurring during the southwest monsoon (June–September), closely aligning with regional rainfall climatology. Post‐monsoon dengue shows marked interannual variability, likely reflecting spatial heterogeneity in monsoon rainfall: northern districts receive most precipitation during the southwest monsoon, whereas southern districts experience enhanced rainfall during the northeast monsoon (October–November), influencing localized mosquito breeding and dengue risk (A. Nair et al., 2014). A comprehensive set of predictors—including temperature, rainfall, RH, monsoon active and break days, Niño 3.4 SST, and EIP—was included in the Kerala dengue model. These variables impact the dengue incidence of the state at different time lags ranging from 0 to 5 months. As dengue peaks in Kerala during the monsoon season, mean monsoon rainfall (lag 0), temperature conditions and Niño 3.4 SST during March–April (lag 2–3 months), and RH during January–February (lag 5 months) were found to be associated with dengue incidence during the peak period. Among the ML approaches evaluated (XGBoost, SVR, RFR, LSTM), XGBoost achieved the highest predictive performance, likely due to its capacity to model complex nonlinear relationships, handle correlated predictors, and robustly combine multiple lagged variables. The Kerala dengue model developed in the current study can be further improved through enhanced surveillance, better representation of vector biology and socio‐demographic factors in the model, and continuous validation—essential steps to operationalize climate‐informed health decision‐making. SHAP analysis confirmed temperature and rainfall as the most influential drivers of dengue incidence. Further, our analysis using the dengue model finds that increased dengue incidence over Kerala is associated with warmer temperatures and moderate RH during the pre‐monsoon months, along with increased rainfall during the monsoon season, while El Niño conditions in the Pacific are associated with local weather conditions that favor increased dengue incidence over Kerala. A notable contribution of this study is the incorporation of ENSO as a dengue predictor. Both El Niño and La Niña are found to enhance dengue risk depending on regional climate responses. El Niño has been associated with warmer and drier conditions in Venezuela and Thailand, increasing dengue incidence (Tipayamongkholgul et al., 2009; Vincenti‐Gonzalez et al., 2018), while La Niña elevated dengue risk in the Solomon Islands through warmer local temperatures (Andhikaputra et al., 2023). In India, El Niño is positively correlated with higher surface temperatures (Chowdary et al., 2014; Kothawale et al., 2010), suggesting elevated dengue transmission during El Niño years (F. J. Colón‐González, 2012; Thai et al., 2010). ENSO composites for Kerala indicate that El Niño years are characterized by higher temperatures and reduced rainfall and RH, conditions favorable for dengue transmission. The predictive value of large‐scale oceanic indicators is further supported by a hydroclimate‐based forecasting study from Colombia, which demonstrated that incorporating SST as a predictor can improve dengue prediction skill at 3–6 months lead times (Beal et al., 2025). Integrating ENSO monitoring into the dengue model can therefore provide actionable lead time for public health interventions. Future dengue projections based on the Kerala model indicate substantial increases in dengue incidence across all emission scenarios, with higher incidence under high‐emission pathways (SSP5‐8.5). In the near future (2021–2040), dengue incidence is projected to increase by 16%–20% relative to the reference period (1995–2014), with mid‐century (2041–2060) increases ranging from 32% under SSP1‐2.6 to 44% under SSP5‐8.5. By the late century (2081–2100), projected rises are 46% under SSP1‐2.6, 63% under SSP2‐4.5, and up to 113% under SSP5‐8.5 (Figures 6a and 6b). These results emphasize that strong mitigation under low‐emission scenarios could substantially limit the negative impacts of climate change on dengue and public health. This study has certain limitations. We used climatological factors as exposure variables; however dengue transmission is also influenced by socio‐economic, demographic, geographical, virological, entomological and immunological factors. Changes in the circulating DENV strains could trigger sudden surges in dengue incidence, independent of underlying climatic contributors. Similarly, future projections under SSP scenarios do not explicitly account for localized influences such as urbanization, socio‐economic conditions, or community adaptive capacity. These gaps, common in many CMIP‐based disease projection studies (Wyk et al., 2023; Zheng et al., 2024), need to be addressed in future work. Several dengue vaccines are currently in development, and their future deployment could substantially reduce dengue incidence. In addition, this analysis was conducted at the state scale. Given the spatial variability in monsoon characteristics and dengue distribution within Kerala, more localized studies using high‐resolution climate and epidemiological data could provide more detailed insights and improved predictive skill. Despite these constraints, the dengue model demonstrated reasonable forecasting accuracy, providing region‐specific insights that can support early warning systems and public health planning to control dengue in Kerala. To enhance the regional relevance of dengue future projections, the current study performed a model selection procedure and used six CMIP6 models that best represent monsoon variability and mean rainfall over India (Sophia et al., 2025). Finally, this study benefited greatly from the availability of monthly dengue data provided by the Kerala Directorate of Health Services. Their efforts in maintaining timely, open‐access data enabled the development of our predictive model. Such transparency is critical for research and public health preparedness, and other regions are encouraged to adopt similar practices to support the development of disease modeling and early warning systems. At the same time, it is important to acknowledge potential limitations in the case data, particularly the possibility of underreporting due to asymptomatic infections, variability in healthcare‐seeking behavior, and differences in diagnostic and reporting practices over time, which may influence the absolute magnitude of reported dengue cases. Once long‐term data is available, future work could further investigate the contribution of anthropogenic climate change to dengue risk by comparing observed climate with CMIP6 historical‐and‐natural simulations to assess the extent of climate‐driven changes in the observed dengue burden. Conflict of Interest The authors declare no conflicts of interest relevant to this study. Supporting information Supporting Information S1 Acknowledgments The authors acknowledge the Indian Institute of Tropical Meteorology (IITM), Pune, for providing institutional facilities and access to high‐performance computing (HPC) resources through the Pratyush and Arka HPC systems, which supported the computational analyses conducted in this study. The authors also acknowledge the Directorate of Health Services, Kerala, for making dengue surveillance data publicly available, which facilitated this research. Contributor Information Yacob Sophia, Email: sophiayacob123@gmail.com. Mathew Koll Roxy, Email: roxy@tropmet.res.in. Availability Statement Dengue surveillance data used in this study are publicly available from the Integrated Disease Surveillance Programme (IDSP) portal of the Kerala Directorate of Health Services: https://dhs.kerala.gov.in/en/idsp‐2/. Gridded temperature and rainfall data (Pai et al., 2014; Srivastava et al., 2009) were obtained from the India Meteorological Department high resolution gridded data set provided by the National Climate Centre: https://www.imdpune.gov.in/cmpg/Griddata/Rainfall_25_NetCDF.html. Relative humidity data were obtained from the ERA5 reanalysis data set (Hersbach et al., 2020) produced by the European Centre for Medium Range Weather Forecasts and distributed through the Copernicus Climate Data Store: https://cds.climate.copernicus.eu/datasets/reanalysis‐era5‐single‐levels. Sea surface temperature data used to derive the Niño 3.4 index were obtained from the Hadley Centre Sea Ice and SST data set (HadISST; Rayner et al., 2003) provided by the UK Met Office: https://www.metoffice.gov.uk/hadobs/hadisst/. Future climate projections were derived from Coupled Model Intercomparison Project Phase 6 model outputs (Eyring et al., 2016) obtained through the Earth System Grid Federation data portal: https://esgf‐node.llnl.gov/search/cmip6/. All data sources used in this study are publicly available from the links provided above. In addition, the observed climate and health data used in this study, along with details of CMIP6 data download and processing, and the codes required to reproduce the analyses and figures, are available at Zenodo (Yacob & Roxy, 2026) and can be accessed at: https://doi.org/10.5281/zenodo.20096919. References Associated Data Data Citations Supplementary Materials Supporting Information S1 Data Availability Statement Dengue surveillance data used in this study are publicly available from the Integrated Disease Surveillance Programme (IDSP) portal of the Kerala Directorate of Health Services: https://dhs.kerala.gov.in/en/idsp‐2/. Gridded temperature and rainfall data (Pai et al., 2014; Srivastava et al., 2009) were obtained from the India Meteorological Department high resolution gridded data set provided by the National Climate Centre: https://www.imdpune.gov.in/cmpg/Griddata/Rainfall_25_NetCDF.html. Relative humidity data were obtained from the ERA5 reanalysis data set (Hersbach et al., 2020) produced by the European Centre for Medium Range Weather Forecasts and distributed through the Copernicus Climate Data Store: https://cds.climate.copernicus.eu/datasets/reanalysis‐era5‐single‐levels. Sea surface temperature data used to derive the Niño 3.4 index were obtained from the Hadley Centre Sea Ice and SST data set (HadISST; Rayner et al., 2003) provided by the UK Met Office: https://www.metoffice.gov.uk/hadobs/hadisst/. Future climate projections were derived from Coupled Model Intercomparison Project Phase 6 model outputs (Eyring et al., 2016) obtained through the Earth System Grid Federation data portal: https://esgf‐node.llnl.gov/search/cmip6/. All data sources used in this study are publicly available from the links provided above. In addition, the observed climate and health data used in this study, along with details of CMIP6 data download and processing, and the codes required to reproduce the analyses and figures, are available at Zenodo (Yacob & Roxy, 2026) and can be accessed at: https://doi.org/10.5281/zenodo.20096919.