Tuberculosis (TB) is a chronic respiratory communicable disease caused by Mycobacterium tuberculosis (MTB)1. Despite a reduction of about 5.9% and 2% per year in TB incidence and mortality, respectively, from 2015 to 2022 worldwide1, TB remains a major cause of ill health and one of the leading causes of death except for COVID-191. Globally, there were an estimated 10.6 million people who developed TB and 1.3 million deaths associated with TB in 20221. Disappointingly, progress on track to ending the TB strategy made in the past years has been stalled, and even worse showing a recurring risk in some countries or regions, partly attributable to the COVID-19 outbreak, the rapidly increased burden of drug-resistant TB (DR-TB), and the fast-evolving epidemiological trends of TB determinants (e.g., undernourishment, poverty, HIV-TB co-infection, smoking, and diabetes-TB comorbidity)1,2. Therefore, to achieve the goals of ending TB epidemic, with a reduction of 90% and 95% in TB morbidity and mortality, respectively, in 20353, it is essential to examine the determinants and construct an accurate forecasting model for informing vaccine and effective interventions.

In the context of climate change, meteorological factors are likely to become a major contributor to the spread of communicable diseases as they are capable of impacting the causative agents, hosts, and transmission environment, and they may be a proxy of reliable early warning predictor for estimating the epidemics of many infectious diseases (e.g., scarlet fever, TB, and human brucellosis)4,5,6. Literature has emerged that climate variability is related to TB incidence2,7,8,9,10. For example, Wang et al. found that temperature and precipitation played a pivotal role in TB among people living with HIV/AIDS in Guangxi, while wind had no effect7. Li et al. indicated a combined impact of temperature and humidity on TB epidemic8. Chang et al. showed a non-linear and delayed effect of temperature, relative humidity, and wind speed on TB epidemic10. Diez et al. suggested a negative relation between TB morbidity and sunshine at a 4-month lag in Spain9. But there are gaps in the past studies. (1) Little study has simultaneously investigated the effects of temperature, air pressure, rainfall, sunshine, relative humidity, and wind velocity on TB. However, these six common weather parameters are co-exposed, which may result in complex interplays between positive and negative changes in these variables on TB; (2) Prior studies ignored the strong autocorrelations between dependent variables that must be considered in time series analysis11, which often leads to a biased estimate; (3) Meteorological variables may have long-term (this effect is essential for understanding the sustained influence of climate variability on TB epidemic. By analyzing this effect, the stakeholders can gain insights into the underlying trends and structural changes in the data, allowing for more robust policy recommendations and forecasting) and short-term (this effect pertains to the immediate response of climate variability to disease epidemic12. This captures the transient fluctuations and adjustments in TB incidence following the variation in climatic factors. Understanding short-term effects is crucial for assessing the immediate impacts of climate variation on the TB transmission and informing timely decision-making) asymmetric dynamic effects on TB epidemic (i.e., an increase or decrease in meteorological variables may result in different effects on TB). By considering the interplay between these two types of effects, a comprehensive understanding of the dynamic relationships within the data can be developed.

Currently, the autoregressive distributed lag (ARDL) model has been used to capture the relationships between dependent and independent variables not only at the same time point but also across lagged values13, but the assumption of the ARDL framework is that the long-run relationship is a symmetric linear combination of regressors, which may not align with the nonlinearity and asymmetry often observed in environmental and medical domains7,8. In light of this, a nonlinear autoregressive distributed lag (NARDL) model was thus introduced because of its advantages below12,14,15,16: (1) This framework accounts for both short-run and long-run nonlinearities by modeling them as positive and negative partial sum decompositions of the explanatory variables; (2) Compared to other cointegration models, this framework relaxes the restriction for the same orders of integration among variables and has a higher power of test, particularly in a small sample; (3) This framework favors addressing the endogenous problem between variables; (4) This framework allows automatically identifying the optimal autocorrelations among independent and dependent variables. This holistic approach enables a more nuanced analysis of complex phenomena and helps in uncovering hidden patterns and nonlinearities that traditional linear models may overlook. However, few studies have used the NARDL model to investigate the long- and short-term effects of weather factors on TB. Consider that the TB incidence is a major public health in Henan as it frequently ranked the second among the 26 class B notifiable infectious diseases17. Therefore, this study aimed (1) to investigate the long- and short-run asymmetric dynamic association between meteorological variables and TB in Henan by use of the NARDL model, and (2) to clarify whether the NARDL model can improve the forecasting accuracy of TB epidemic as compared to the ARDL model. By incorporating both dimensions of effects into the analysis, the accuracy and reliability of the forecasts and policy recommendations can be enhanced, which is invaluable for policymakers and analysts seeking to design effective interventions and anticipate the consequences of their actions over varying time horizons14,18.

Materials and methods

Study area

Henan province, situated in central China, lies between 31° 23′-36° 22′ north latitude and 110° 21′-116° 39′ east longitude. It serves as a vital comprehensive transportation hub and a major center for population movement in China. Covering an area of 167,000 km2, the province predominantly falls within the warm temperate zone, with its southern regions transitioning to a sub-tropical climate. Henan experiences a continental monsoon climate, ranging from the northern subtropical zone to the warm temperate zone. By 2023, Henan has a registered population of 115 million, making it the largest province with a registered population in China.

Source of TB data

Monthly TB incident data between January 2004 and December 2019 in Henan were obtained from the Data-center of China Public Health Science (DCPHS) (https://www.phsciencedata.cn), managed by the Chinese Center for Disease Control and Prevention. Population data for the corresponding period were obtained from the Henan Statistical Yearbook. All TB cases were confirmed by authorized institutions following the diagnostic criteria for TB (http://www.nhc.gov.cn/wjw/s9491/wsbz.shtml). A total of 192 observations were collected.

Source of Meteorological data

Daily climate factors were collected from the National Meteorological Science Data Center (http://data.cma.cn/). These factors were then collated in a monthly format for further analysis (Table 1).

Table 1 Data description summarizing the study variables, abbreviations, data sources, and their respective units.
Full size table

Statistical analysis

With climate factors and TB cases being expressed as mean ± standard deviation (\(\bar {x} \pm s\)) or interquartile range (IQR) during the data description. The average annual percentage change (AAPC) was used to characterize the epidemic trend of TB over the study period19. Spearman’s rank correlation (rs) was employed to indicate the correlation between weather variables and TB, with rs > 0.8 or variance inflation factor (VIF) > 10 indicating the presence of multicollinearity between these factors20,21. Subsequently, these factors were incorporated into different NARDL models along with other weather parameters to assess their independent effects on TB.

To investigate the impacts of meteorological factors on TB incidence, two modeling techniques were employed. The ARDL model, known for addressing issues of autocorrelations and non-stationarity among variables, served as the baseline model. However, the ARDL model may generate biased results in the presence of non-linear or asymmetric impacts of climate variables on diseases22. To address this limitation, the NARDL model was introduced. The NARDL model allows for the analysis of long- and short-term asymmetries in the relationship between weather parameters and TB incidence14,23. By considering positive and negative changes in each weather parameter, the NARDL model can quantify the responses of TB incidence to these changes14,23. The construction of the NARDL model involves four key procedures14,23,24. Initially, determining the integrated order for all variables. Although the NARDL model relaxes the integration requirement23, it is imperative to note that the order of integration should not exceed one. Moreover, non-stationary independent variables may lead to a pseudo regression scenario. Consequently, the augmented Dickey–Fuller (ADF) test was employed to assess the order of integration and stationarity for all variables25. In cases where a non-stationary series is identified, logarithmic transformation and/or differencing technique is often used to ensure that stationarity is achieved. Subsequently, exploring the long-run asymmetric cointegration effect of weather variables on TB. The bounds test, often represented by the F statistic, serves as a pivotal tool in determining the existence of this long-run asymmetric cointegration relationship26. Upon confirming the presence of such a relationship, further analysis is conducted to investigate the short- and long-term asymmetries using a Wald test10,12. The third step involves estimating effects. This entails quantifying the dynamic multiplier effect of regressors on the dependent variable, including both positive and negative impacts24. By examining these dynamics, a deeper understanding of the relationships between variables can be gained. Lastly, comparison of the forecasting ability among models. The forecasting performance of the ARDL and NARDL models was assessed using a dataset spanning from January 2004 to December 2018 as the training subset. The remaining observations, constituting the testing subset, were utilized for validation. Various error rate metrics such as root mean square error (RMSE), mean absolute deviation (MAD), mean error rate (MER), and mean absolute percentage error (MAPE) were calculated to evaluate the predictive performance of both models18, and the modified Diebold-Mariano (DM) method was used to test whether there was a statistical significance in forecasting performance of both models27.

The equation of the NARDL model can be written as:

$$\begin{aligned} \Delta {Y_t} & ={a_0}+\sum\limits_{{i=1}}^{{{p_i}}} {{p_{1i}}{Y_{t - {\text{1}}}}} +\sum\limits_{{i=0}}^{{{q_1}}} {q_{{1i}}^{+}x_{{t - 1}}^{+}} +\sum\limits_{{i=0}}^{{{q_2}}} {q_{{2i}}^{ - }x_{{t - 1}}^{ - }} +\sum\limits_{{i=1}}^{{{p_2}}} {{p_{2i}}} \Delta {Y_{t - i}} \\ & \quad +\sum\limits_{{i=0}}^{{{q_3}}} {q_{{3i}}^{+}\Delta x_{{t - i}}^{+}} +\sum\limits_{{i=0}}^{{{q_4}}} {q_{{4i}}^{ - }\Delta x_{{t - i}}^{ - }} +{a_1}month+{a_2}t+{\varepsilon _t} \\ \end{aligned} $$
(1)

where Yt denotes the TB incidence series, x refers to the weather variables, x+ and x− represent the positive and negative partial sums of variation in climate factors, respectively, p and q denote the optimal lag orders of TB incidence series and weather factors, respectively, month denotes the seasonal effect, and the first-order difference is represented by \(\Delta\).

In this study, the maximum lag orders were designated as 5 months, which is based on the understanding that there exists an incubation period of approximately 4–8 weeks from MTB infection to symptom onset. Besides, there is an additional 2-month lag from symptom manifestation to clinical diagnosis in China28. Subsequently, the optimal lag orders are determined through a thorough evaluation using criteria such as the Akaike information criterion (AIC), Schwarz criterion (BIC), Hannan-Quinn (HQ) criterion, log-likelihood, and adjusted R2. Autocorrelation in the dependent variable is assessed using the partial autocorrelation function (PACF) plot29, which elucidates the relationship between current values and past values within a defined context. Furthermore, the model incorporates 11 monthly dummy variables and time variables to account for long-term trend and seasonality12. The stability of the NARDL model was scrutinized through the application of cumulative sum (CUSUM) statistic. The statistical analyses were executed using EViews 10 (IHS, Inc., USA) and R 4.2.0 (R Development Core Team, Vienna, Austria), with a significance level set at a two-sided P ≤ 0.05.

Results

Data description

There were 1,120,365 TB cases in Henan during 2004–2019, with an average monthly count of 7,781 cases (5.624 cases per 100,000 individuals) and an annualized figure of 93,364 cases (67.487 cases per 100,000 persons). Throughout the study period, there was a notable decline in TB incidence, with an AAPC of −6.438% [95% confidence interval (CI): −6.438% to −3.730%]. However, there was a sudden surge in cases observed in 2005, with 107,462 notifications (110.014 cases per 100,000 individuals), followed by a subsequent decrease in cases until 2019, when the reported count stood at 47,153 cases (43.054 cases per 100,000 persons). Furthermore, there was a distinct pattern of periodicity and seasonality in TB incidence, with peaks observed between March and May each year, and troughs in November and December.

The ARH, AR, AT, AWV, AAP, and ASH were recorded at 65.72% (59.83%, 73.61%), 40.89 mm (90.12 mm, 307.41 mm), 16.62 °C (7.22 °C, 24.52 °C), 2 m/s (1.8 m/s, 2.23 m/s), 1006.99 hPa (992.35 hPa, 1006.99 hPa), and 150.14 h ± 43.68 h, respectively (Table 2). Analysis depicted in Fig. 1 revealed a similar changing trend between TB and AP, AT, AWV, and ASH. Conversely, a contrasting trend was observed between TB and ARH, as well as AAP. Moreover, a strong correlation was identified between AAP and AT, with rs exceeding 0.8 and VIF surpassing 10 (Table 3).

Table 2 Statistical summaries for the TB incidence series and weather variables in Henan during 2004–2019.
Full size table
Fig. 1
figure 1

Time series plot showing the changing patterns of the climate factors and TB cases in Henan, 2004–2019.

Full size image
Table 3 Spearman’s correlation and collinearity results between variables in Henan, 2004–2019.
Full size table

Construction of the NARDL and ARDL methods

The ADF test indicated that the dependent and independent variables exhibited non-stationarity: TB cases (t=-0.865, P = 0.34), AT (t = 1.315, P = 0.952), AAP (t=-1.657, P = 0.092), ASH (t=-0.370, P = 0.550), AWV (t=-1.126, P = 0.236), AR (t = 0.734, P = 0.398), and ARH (t=-0.154, P = 0.629). However, after applying differencing, all variables exhibited a P value of less than 0.001, indicating that they achieved stationarity. The PACF plot highlighted significant autocorrelation at a 1-month lag (Fig. 2). Furthermore, the bounds test resulted in F = 10.846, which significantly exceeds the critical upper bounds (I0 = 2.14, I1 = 3.34), providing strong evidence of a long-term cointegrating asymmetric relationship among the variables. Subsequently, after controlling for seasonality, autocorrelation, and long-term trends, a variety of NARDL models were constructed. Following thorough evaluation of various potential specifications (Tables S1 and S2), the NARDL(1, 0, 1, 0, 0, 5, 4, 1, 2, 1, 0) model emerged as the preferred choice (with TB cases at a 1-month delay, ARH(+) at a 0-month delay and ARH (−) at a 1-month delay, AR(+) and AR(−) at a 0-month delay, AWV(+) at a 5-month delay and AWV(−) at a 4-month delay, ASH(+) at a 1-month delay and ASH(−) at a 2-month delay, along with AAP(+) at a 1-month delay and AAP(−) at a 0-month delay) as this model demonstrated improved performance metrics compared to others, with lower values of AIC (16.520), BIC (17.162), and HQ (16.780), along with higher values of adjusted R2 (0.785) and log-likelihood (−1499.358) (Fig. 3). Moreover, the CUSUM values remained within the 95%CI, confirming the stability of the NARDL model. Similarly, the specification of ARDL(1, 5, 5, 0, 0, 0) was determined as the optimal choice among the various candidates (Tables S3 and S4).

Fig. 2
figure 2

Partial autocorrelation function (PACF) plot for the differenced TB incidence series. It was observed that there was a local maximum value at a one-month delay, indicating that the TB incidence series with a 1-month delay should be included in the model.

Full size image
Fig. 3
figure 3

CUSUM graph for the stability test under the NARDL model. The CUSUM values less than 0.05 at different periods substantiated the validity and stability of the model.

Full size image

The asymmetric and symmetric impacts of climate variation on TB

The findings presented in Table 4 revealed that ASH and ARH exhibited a positive association with TB incidence, whereas AWV and AAP(+) showed a negative association with TB incidence. Specifically, an increase or decrease of 1 h in ASH led to an increment of roughly 32 TB cases, an increase and decrease of 1% in ARH led to an increment of 115 and 152 TB cases, respectively, an increase of 1 m/s in AWV led to a decrement of 3600 TB cases, while a decrease of 1 m/s resulted in approximately a decrement of 5021 TB cases, and an increase of 1 hPa in AAP led to a decrement of 318 TB cases. On the other hand, AT, AR and AAP(−) did not exhibit significant long-term coefficients. However, the results from the ARDL model revealed that AR displayed a significant positive long-term coefficient (Table 4), indicating a linear impact on TB cases. Specifically, when AR increased or decreased by 1 mm, TB decreased by around 38 cases. Furthermore, in the short run, the ∆AT(−) at a 1-month lag, ∆AWV(+) at a 0-month lag, and ∆ASH(−) at a 1-month lag had a significant negative effect on TB, whereas ∆AWV(+) at a 4-month lag and ∆AWV(−) with 2 and 3-month lags had a meaningful positive impact on TB. The results of the Wald test for asymmetry outlined in Table 5 suggested that AWV and AAP played an asymmetric long-run role in TB incidence, a finding further supported by the dynamic multiplier plot (Fig. 4a and b). Notably, no long-run asymmetric relationship was observed for AR, AT, ARH, and ASH. Additionally, there was an indication of short-run asymmetry between AT and TB, as detailed in Table 5; Fig. 4c and f.

Table 4 Long- and short-term effects of climate variables on the TB incidence.
Full size table
Table 5 Wald test results for asymmetry.
Full size table
Fig. 4
figure 4

Multiplier asymmetric impact of weather factors on TB. (a) Dynamic multiplier plot for the impact of AT on TB, (b) dynamic multiplier plot for the impact of AR on TB, (c) dynamic multiplier plot for the impact of ASH on TB, (d) dynamic multiplier plot for the impact of AWV on TB, (e) dynamic multiplier plot for the impact of ARH on TB, and (f) dynamic multiplier plot for the impact of AAP on TB.

Full size image

Model prediction and performance comparison

After constructing the optimal ARDL and NARDL models using the training set and forecasting the testing subset, the fitting and forecasting curves are depicted in Fig. 5. The comparative performance of both models is detailed in Table 6, indicating that the NARDL model exhibited lower error rates in both fitting and predictive analyses compared to the ARDL model. Furthermore, the DM test conducted in the predictive phase yielded significant results, indicating that the predictive accuracy under the NARDL model significantly outperformed that of the ARDL model. This underscores the NARDL model’s superior ability to capture the dynamic epidemic pattern in TB incidence.

Fig. 5
figure 5

Comparison of the fitting and forecasting results with the actual TB incidence series using (a) the ARDL method, and (b) the NARDL method.

Full size image
Table 6 Comparison of the fitting and forecasted abilities under the ARDL and NARDL methods.
Full size table

Discussion

This study unveiled significant findings concerning the connection between climate variations and TB incidence, examining both long- and short-term perspectives. After accounting for seasonality, autocorrelation, and time variables, it was noted that AWV and AAP(+) demonstrated a noteworthy negative nonlinear association with TB, while ASH and ARH exhibited a significant positive nonlinear association with TB in the long run. In the short term, it was found that ∆AWV(+) at a 4-month lag, ∆AWV(−) at 2-3-month lags were positively correlated with TB. Conversely, ∆AWV(+) at a 0-month lag, ∆ASH(−) at a 1-month lag, and ∆AT(−) at a 1-month lag displayed reverse associations with TB. Moreover, the study highlighted that the NARDL model, incorporating meteorological variables, outperformed the ARDL model in forecasting TB epidemic. The study results confirmed the importance of lead time, along with the asymmetric and symmetric effects of meteorological parameters on TB incidence. The effectiveness of the NARDL model in capturing the dynamic structure of TB epidemics by incorporating both dimensions of effects was also underscored. These findings offer valuable insights for predicting the trajectory of TB epidemics, facilitating the development of targeted prevention and control strategies, and enabling timely public health interventions.

The results of the current study revealed a decrease in TB morbidity, aligning with the global and national epidemic trends in China. This can be attributed to the continuous efforts of the government in TB prevention and control, increased funding allocation for essential TB services, and improvements in the public health system30. Additionally, this study uncovered a distinct seasonal pattern in TB morbidity, with peak cases occurring in spring and the lowest numbers in late autumn and early winter, consistent with the seasonal variation at the national level in China31,32. This seasonal variation may be influenced by factors such as population movements, climate changes, intensified health education and interventions during World Tuberculosis Day, the impact of the spring festival, and reporting delays within the monitoring system32.

The study revealed a significant long-term asymmetric negative correlation between AWV and TB, indicating that a decrease in AWV has a more pronounced effect than an increase. This finding aligns with a previous study that showed a protective effect of high AWV on the transmission of TB. This could be attributed to the fact that (1) wind helps in dispersing the concentration of MTB in the long term33; (2) the presence of liquid on the surface of the etiologic agent causes MTB to aggregate into larger particles that increases surface friction and hinders dispersion in the long term; (3) high wind speed improves environmental ventilation, thereby decreasing the likelihood of airborne infections34. Interestingly, a short-term positive effect was observed between TB and AWV(+) with a 4-month lag and AWV(−) with 2-3-month lags. Previous studies have also established a positive association between AWV and respiratory infectious diseases such as brucellosis, scarlet fever, and mumps5,35,36, thereby lending support to the findings in the short term. Moreover, the study suggested that the short-term effect of AWV on TB may help explain the seasonal variation in TB incidence, particularly the higher risk observed during the months of March-May. This could be attributed to the increased wind activity in winter in Henan compared to other seasons. However, the findings fit well with a recent study that reported a negative correlation between AWS at a 0-month lag and TB in Qinghai2. Plausible explanations for the findings in the short term include the following2,35,37: First, strong winds could facilitate the spread of contaminated aerosols in areas with higher population densities over a short term. Second, intense wind conditions might lead to greater evaporation of liquid on the surface of MTB, indirectly influencing the spread of TB in the short term.

Another important finding is that ARH may exert a prolonged and symmetric influence on TB. This finding aligns closely with previous research in the field2,34. Furthermore, this study reflects similarities with various investigations that have highlighted a positive correlation between ARH and other respiratory infectious diseases such as scarlet fever and mumps5,36. However, contrary to the conclusions emerged from this study, some studies have indicated an inverse relationship between ARH and TB29,30. This disparity could potentially be attributed to variations in analytical models, geographical regions, or the absence of adjustments for autoregression in the dependent variable. Of particular significance is the validation of our findings by results from an experimental study38, indicating that elevated ARH levels could potentially cause a decrease in airway mucus secretion and hinder the clearance of external pathogens, potentially increasing susceptibility to MTB infection34. Another plausible explanation could be that MTB has the propensity to bind with ambient particulate matter, forming larger particles under conditions of high ARH, consequently prolonging their airborne presence39.

The third significant discovery revealed that ASH has a long-term symmetric positive impact on TB. This finding aligns well with several recent studies that have demonstrated a positive influence of ASH on TB in Yunnan and scarlet fever5,40,41. The observed long-term positive relationship between ASH and TB could potentially explain the lower-risk seasonality in TB incidence, as reduced sunshine exposure during winter leads to decreased vitamin D levels, thereby weakening host immunity to MTB. However, a recent study in Japan found that reduced sunshine exposure and vitamin D deficiency did not contribute to the seasonality in TB incidence42. Therefore, further investigation into the underlying mechanism of the relationship between ASH and TB is necessary. However, ∆ASH(−) with a 1-month lag was indicated to have a short-term negative effect of on TB. This result is consistent with a recent study that indicated a notable negative correlation between ASH with a 2-month lag and TB in Qinghai2. More recent research has shown that shortened exposure to sunlight and ultraviolet rays favors the growth and reproduction of MTB, as well as restricts vitamin D synthesis in the body, thereby reducing host immunity and potentially affecting susceptibility to TB43,44, this can explain the short-term effect of ASH observed in this study.

The fourth significant finding pertains to the potential asymmetric long-term impact of AAP on TB. This result fits well with a study that also found a significant negative correlation between AAP and TB39, as well as other respiratory infectious diseases such as scarlet fever and mumps5,36. However, a contrasting study focused on Lanzhou revealed a positive association between AP and TB with a lag of 4–6 days45. The suggested explanation for this positive impact of AAP on TB involves higher air pressure enhancing the Brownian motion of MTB, potentially prolonging its airborne duration and increasing the risk of exposure to the bacteria. These findings contradict the results of the present study. Meanwhile, other study has indicated that during winter, low temperatures and high air pressure, in combination with air cooling and reduced molecular movement, can impede the airborne survival of MTB, thereby reducing the risk of human infection34. This aligns with the outcomes of the current study. Given these divergent conclusions, a more thorough investigation into the underlying mechanisms is warranted.

The fifth finding revealed that ∆AT(−) with a 1-month lag had a significant short-term asymmetric negative impact on TB. This implies that a decline in temperature could lead to a decrease in TB cases. This finding aligns with the observed trough in TB incidence during winter in Henan and in studies from various countries2,46. The delayed short-term effect of temperature on TB is influenced by several factors. First, temperature plays a pivotal role in influencing human behavior and contact patterns2. Higher temperatures often prompt more indoor activities, increasing the chances of contact between TB patients and susceptible individuals. Conversely, lower temperatures encourage outdoor activities, potentially reducing opportunities for TB transmission. Second, lower temperatures can impede the replication and viability of pathogens, which is unfavorable for the spread of TB34. Lastly, decreased temperatures are associated with reduced airflow, creating an environment that is less conducive to TB transmission34.

This study focused on analyzing the long- and short-term asymmetric impacts of variations in climate variation on TB incidence. Previous studies have emphasized the significance of considering factors such as changes in population immunity, autocorrelations, various lags and relationship patterns, seasonality, and long-term trends in time series analysis11. These factors were thoroughly considered in this study, with the exception of changes in population immunity due to data unavailability. As a result, the current findings are confident. Fluctuations in climate variation plays a crucial role, both asymmetrically and symmetrically, in TB incidence, and the NARDL model proves to be a valuable tool for forecasting TB epidemics by considering the interplay between these two types of effects. The policy implications emerged from this study findings are that as an active agent, humans have the capacity to mitigate the health impacts associated with climate change through proactive measures. This includes gaining a deeper understanding of climate change patterns and the complex, TB-specific health effects presented in this study, as well as effectively allocating technologies and resources to promote healthy lifestyles and enhance public awareness47. The following adaptation measures are recommended47: (1) Move beyond empirical observations of the link between climate change and TB by developing more scientifically grounded explanations. (2) Enhance the prediction of spatial-temporal processes of climate change and the resulting shifts in TB incidence across various spatial and temporal scales. (3) Establish locally tailored early warning systems to anticipate and address the health implications of projected climate change.

This study also has some limitations. First, under-reporting or under-diagnosis is inevitable in a passive monitoring system. Second, as an ecological trend study, it is unable to investigate individual-based relationships or infer causal effects. Third, daily or weekly data could offer deeper insights into temporal variations between years, but their unavailability hinders further exploration. Lastly, this study did not account for the impact of unmeasured confounders, such as geographic and socioeconomic factors, population density, and host susceptibility.

Conclusions

In summary, this study found that AWV, ASH, AAP, AT, and ARH are significant factors influencing both long- and short-term asymmetric and/or symmetric patterns of TB incidence. It is imperative that meteorological variables are incorporated into public health intervention strategies for TB, especially given the context of ongoing global climate change. The NARDL model has proven to be more effective in capturing the dynamic epidemic structure of TB incidence compared to the ARDL model. Therefore, the NARDL model by considering the interplay between these two types of effects can gain comprehensive insights into the relationships between climate variation and TB and can be considered a valuable tool in making more informed decisions regarding policy, forecasting, or strategic planning. However, considering the spatial heterogeneity in TB incidence and the conflicting results in different studies, it is essential for future research to prioritize addressing this heterogeneity using the non-linear panel autoregressive distributed lag (NPARDL) model, because recent study indicated that the NPARDL model has emerged as a useful statistical method for unearthing the intricate dynamics inherent in heterogeneous panel datasets48. Through the incorporation of non-linear relationships and lag effects, the NPARDL model can adeptly navigate the intricacies associated with spatial heterogeneity in time series analysis48.