Modelling the impact of climate variability on malaria morbidity in the Tamale Metropolitan Area: a time series analysis Authors: Abdul-Ganiu Zakaria, Shamsu-Deen Ziblim, Yakubu Amadu Published: 2026 ABSTRACT BackgroundMalaria remains endemic in northern Ghana, with seasonal outbreaks placing significant strain on health facilities in the Tamale Metropolitan Area. Although climatic variability is known to influence malaria transmission, the specific short- and long-run relationships between temperature, rainfall, relative humidity, and malaria outcomes in this setting remain poorly understood. This study models the impact of climate variability on population-adjusted malaria morbidity in the Tamale Metropolitan Area using a time series approach.MethodsMonthly time series data on laboratory-confirmed malaria morbidity (converted to rates per 1,000 population), as well as temperature, rainfall, and relative humidity, were obtained from the District Health Information Management System and the Ghana Meteorological Agency for the period January 2014 to December 2020 (84 observations). The Autoregressive Distributed Lag (ARDL) bounds testing approach was selected for its ability to handle variables with mixed orders of integration (I(0) and I(1)) and to simultaneously estimate short- and long-run dynamics. No logarithmic transformation was applied, preserving direct interpretation of coefficients as changes in malaria rates per 1,000 population.ResultsThe ARDL cointegration test confirmed a long-run equilibrium relationship between each climatic variable and malaria morbidity. In the short run, increases in relative humidity, temperature, and rainfall were significantly associated with higher malaria morbidity (p < 0.05 to p < 0.001). In the long run, however, temperature showed a significant inverse relationship: a one-unit (1 °C) increase in temperature was associated with a 0.37 case per 1,000 population decrease in malaria morbidity (p = 0.038). Model diagnostic tests (Ljung-Box and ARCH-LM) indicated that residuals were white noise, supporting model validity.ConclusionClimatic variables, particularly temperature, play a significant but complex role in malaria transmission in the Tamale Metropolitan Area, with opposing short- and long-run effects. These findings support the development of climate-informed early warning systems tailored to northern Ghana. To strengthen local malaria control and climate adaptation strategies, future efforts should integrate intervention coverage and health system data, improve surveillance, and apply advanced time-series methods to better capture non-linear and delayed climate effects.Clinical trialNot applicable. Introduction Malaria remains one of the most significant global public health threats. Although global malaria morbidity and mortality declined steadily over the past three decades, recent disruptions, notably the COVID-19 pandemic, have reversed some of these gains, with morbidity rising from 56.3 to 59.0 cases per 1,000 population at risk between 2019 and 2020 [1]. Sub-Saharan Africa bears a disproportionate burden, accounting for approximately 95% of cases and 96% of deaths globally, with 80% of all regional deaths occurring among children under five years of age [2]. This high transmission intensity is largely driven by ecological and climatic conditions that favor the proliferation of Anopheles mosquitoes. In Ghana, malaria is endemic and persistent, with marked seasonal fluctuations. The northern region experiences a single rainy season from May to September (peaking in August), resulting in a three- to seven-month transmission season and a maximum number of cases between July and November. Ghana accounts for 2.1% of global malaria morbidity and 1.9% of global mortality, and despite considerable control efforts, malaria imposes a heavy socioeconomic burden: malaria expenditure accounts for 34% of the income of impoverished households compared with only 1% among wealthy households [3–5]. Pregnant women and children under five remain at increased risk of malaria mortality [6]. The growth and maturity of the Plasmodium parasite and the ecology of mosquito vectors are highly sensitive to climatic factors. Temperature influences parasite development rates and mosquito survival [7, 8]; rainfall creates breeding sites, although extreme events may diminish them [9, 10]; and relative humidity affects vector desiccation and the extrinsic incubation period [11, 12]. Empirical evidence has consistently shown that climatic variability can either support or undermine malaria control efforts [13]. Despite this established nexus, current research investigating the climate–malaria relationship through time-series methodologies remains limited, particularly in highly endemic regions such as Sub-Saharan Africa [6, 14, 15]. Where studies exist, they often rely on conventional ARIMA-type models, which, while effective for capturing seasonal patterns, inadequately account for covariance among meteorological variables, thereby limiting explanatory power for complex climate-disease interactions [14, 16]. More dynamic approaches, such as ARDL cointegration, dynamic regression, and machine learning, have been employed elsewhere [6, 17–19], but Ghana remains largely underrepresented in this predictive modelling literature. Specifically, no study has systematically modelled both the short- and long-run effects of climate variability on malaria morbidity in the Tamale Metropolitan Area of northern Ghana. This knowledge gap is actionable. Tamale shares socioeconomic and climatic peculiarities with much of northern Ghana: a single intense rainy season, a fragile health system, and recurrent inundation of under-resourced facilities with malaria cases during peak transmission months. Understanding how rainfall, temperature, and relative humidity drive malaria outcomes in this setting is essential for developing climate-informed early warning systems and relevant preventive and coping strategies. Therefore, the objective of this study was to model the short-run and long-run effects of climate variability (temperature, rainfall, and relative humidity) on laboratory-confirmed malaria morbidity in the Tamale Metropolitan Area from January 2014 to December 2020, using the Autoregressive Distributed Lag (ARDL) bounds testing approach to cointegration. The findings aim to inform climate-resilient prevention and adaptive measures against malaria in northern Ghana. Methods This section highlights the different methods and techniques adopted in this study. Study area The study was conducted in the Tamale Metropolitan Area, the administrative capital of northern Ghana (latitude 9°23′ N, longitude 0°50′ W; approximately 750 km²). This location was selected due to the endemic nature of malaria and the availability of reliable health and climatic data. Data sources Monthly time series data spanning January 2014 to December 2020 (84 observations) were obtained from two sources: Health data Laboratory-confirmed malaria morbidity records were sourced from the District Health Information Management System (DHIMS) of the Ghana Ministry of Health (accessed August 23, 2021). Monthly malaria morbidity counts were converted to incidence rates per 1,000 population using annual population estimates for the Tamale Metropolitan Area. Intercensal estimates were interpolated linearly between the 2010 Ghana Population and Housing Census (371,351) and the 2020 projected population (494,000). Climate data Monthly measurements for temperature (°C), rainfall (mm), and relative humidity (%) were obtained from the Ghana Meteorological Agency (accessed August 30, 2021). Analytical framework Pre-processing and software All analyses were performed in R version 4.2.2 using the packages dynlm (dynamic linear models), tseries (KPSS test), vars (VAR lag selection), lmtest (diagnostics), and strucchange (structural break tests). Data were checked for missing values (none present). Variable transformation and functional form All variables (malaria morbidity rate, temperature, rainfall, and relative humidity) were analyzed in their original level forms (non-logged). This choice preserves direct interpretation of coefficients as marginal effects (change in malaria cases per 1,000 population per one-unit change in a climate variable) and avoids the elasticity interpretation that would be required in a log-log specification. Trend analysis To identify underlying temporal patterns and guide the inclusion of a deterministic trend, three models (linear, quadratic, and log-linear) were estimated for each variable. Model selection was based on the Bayesian Information Criterion (BIC). Stationarity and unit root testing The stationarity properties of all series were examined using the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test to determine the order of integration and avoid spurious regression. Cointegration and long-run analysis Given the mixed orders of integration among the variables, conventional cointegration tests (e.g., Johansen) were inappropriate. The Autoregressive Distributed Lag (ARDL) bounds testing approach [20] was therefore selected because it (i) accommodates regressors that are I(0), I (1), or mutually cointegrated; (ii) simultaneously estimates short- and long-run parameters; and (iii) provides valid inference with small sample sizes (n = 84). Lag selection Optimal lag lengths (p, q) were selected using the Akaike Information Criterion (AIC) with a maximum lag of 8 (to capture up to two full seasonal cycles). The final model [7, 8] represents the AIC-minimizing specification. Multicollinearity assessment Variance inflation factors (VIF) were calculated for all independent variables; all VIF values were below 5, indicating no problematic multicollinearity. The general ARDL(p, q) model is specified as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y_t} = {C_0} + \mathop \sum \limits_{i = 1}^p {\phi _i}{Y_{t - i}} + \mathop \sum \limits_{j = 0}^q {\beta _j}{X_{t - j}} + {\varepsilon _t}$$\end{document} where YtYt is the dependent variable (malaria morbidity rate), XtXt is a vector of climate variables, ϕiϕ i and βjβj are parameters, and εtεt is the white noise error term. The existence of a long-run relationship was tested using the F-bounds test. Upon confirmation of cointegration, the model was reparameterized into an Error Correction Model (ECM) form:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{gathered}{{\Delta }}{Y_t} = {C_0} - \alpha \left( {{Y_{t - 1}} - \theta {X_{t - 1}}} \right) \hfill \\+ \mathop \sum \limits_{i = 1}^{p - 1} {\psi _i}{{\Delta }}{Y_{t - i}} + \mathop \sum \limits_{j = 0}^{q - 1} {\omega _j}{{\Delta }}{X_{t - j}} + {u_t} \hfill \\ \end{gathered} $$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} is the speed of adjustment coefficient, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:({Y}_{t-1}-\theta\:{X}_{t-1})$$\end{document} represents the error correction term capturing deviations from the long-run equilibrium. Model diagnostics Model diagnostics: Post‑estimation diagnostic tests were conducted to validate model adequacy: the Ljung‑Box test for serial correlation and the ARCH‑LM test for heteroscedasticity. Residuals were considered white noise if p‑values exceeded 0.05. Ethical approval This study received ethical approval from the Committee on Human Research, Publication, and Ethics (CHRPE) of the Kwame Nkrumah University of Science and Technology (reference number CHRPE/AP/442/21). The study was conducted in accordance with the Declaration of Helsinki. The analysis used aggregated, anonymized data with no personal identifiers; therefore, individual consent was not required. This anonymization protocol supports reproducibility while adhering to ethical standards for secondary analysis of routine health data. Results Summary statistics of data Table 1 shows that over the 84‑month study period (January 2014 – December 2020), the Tamale Metropolitan Area recorded an average malaria morbidity rate of 5.84 cases per 1,000 population per month (range: 0.92–19.06). The high variability (SD = 3.65) reflects the pronounced seasonality of transmission in this setting. Table 1Preliminary analysis of malaria morbidity (Rate per 1,000 Population)VariableMalaria Morbidity (Rate)Mean5.84 per 1,000Median5.40 per 1,000Minimum0.92 per 1,000Maximum19.06 per 1,000STD3.65CV0.625Skewness1.081Ex. Kurtosis1.299Source: Authors Construct Table 1Preliminary analysis of malaria morbidity (Rate per 1,000 Population)VariableMalaria Morbidity (Rate)Mean5.84 per 1,000Median5.40 per 1,000Minimum0.92 per 1,000Maximum19.06 per 1,000STD3.65CV0.625Skewness1.081Ex. Kurtosis1.299Source: Authors Construct Preliminary analysis of malaria morbidity (Rate per 1,000 Population) Source: Authors Construct As illustrated in Table 2, the average relative humidity for the study period was 58.86% (range: 20.50–82.50%); the average temperature was 29.48 °C (range: 26.4–34.2 °C); and the average rainfall was 84.29 mm (range: 0–343.6 mm). Rainfall showed the highest variability (SD = 88.75), followed by relative humidity (SD = 19.57) and temperature (SD = 1.95). Table 2Preliminary analysis of climatic variablesVariableRelative HumidityTemperatureRainfallMean58.8629.4884.29Median64.52951.65Minimum20.526.40Maximum82.534.2343.6STD19.571.9588.75CV0.3320.0661.053Skewness-0.4770.490.918Ex. Kurtosis-1.169-0.75-0.164 No. Data Points = 84 Source: Author’s Construct, 2022 Table 2Preliminary analysis of climatic variablesVariableRelative HumidityTemperatureRainfallMean58.8629.4884.29Median64.52951.65Minimum20.526.40Maximum82.534.2343.6STD19.571.9588.75CV0.3320.0661.053Skewness-0.4770.490.918Ex. Kurtosis-1.169-0.75-0.164 No. Data Points = 84 Source: Author’s Construct, 2022 Preliminary analysis of climatic variables No. Data Points = 84 Source: Author’s Construct, 2022 Trend analysis of malaria morbidity Table 3 presents the trend model selection. For malaria morbidity, the log‑linear model yielded the lowest BIC value (150.8) and the highest adjusted R² (0.324), indicating that malaria outcomes declined exponentially over the study period in Tamale. Table 3Trend analysis of malaria morbidityModelR-squaredAdj R-squaredAICBICMalaria MorbidityLinear0.1940.18414341439Quadratic0.2780.2614271434Log – Linear 0.34 0.324 143.5 150.8 Source: Author’s Construct Table 3Trend analysis of malaria morbidityModelR-squaredAdj R-squaredAICBICMalaria MorbidityLinear0.1940.18414341439Quadratic0.2780.2614271434Log – Linear 0.34 0.324 143.5 150.8 Source: Author’s Construct Trend analysis of malaria morbidity Source: Author’s Construct Fig. 1Trend in malaria morbidity Trend in malaria morbidity From the series plot in Fig. 1, malaria mortality indicates a fluctuating trend that slowly wanders downwards with no obvious outliers. However, there is no sudden shift in the trend and also a seasonal pattern. Trend analysis of climatic variables The results of the trend analysis of the climate variables as indicated in Table 4 shows that the best trend that characterizes relative humidity is the Log-linear model, this is due to the fact that it records the lowest value of Bayesian Information Criterion (BIC). Temperature and Rainfall are also best characterized by the Log-linear Models since the lowest values of the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). Table 4Trend analysis of climatic variablesModelR-squaredAdj R-squaredAICBICRelative HumidityLinear0.001-0.011740.9745.8Quadratic0.002-0.023742.9750.2Log – Linear 0.001 -0.024 88.55 95.84 TemperatureLinear0.009-0.002352.5357.3Quadratic0.017-0.008353.9361.2Log-Linear 0.017 -0.008 -216.7 -209.4 RainfallLinear0.007-0.005994.4999.3Quadratic0.007-0.018996.41004Log – Linear 0.053 0.021 206.7 213.2 Source: Author’s Construct Table 4Trend analysis of climatic variablesModelR-squaredAdj R-squaredAICBICRelative HumidityLinear0.001-0.011740.9745.8Quadratic0.002-0.023742.9750.2Log – Linear 0.001 -0.024 88.55 95.84 TemperatureLinear0.009-0.002352.5357.3Quadratic0.017-0.008353.9361.2Log-Linear 0.017 -0.008 -216.7 -209.4 RainfallLinear0.007-0.005994.4999.3Quadratic0.007-0.018996.41004Log – Linear 0.053 0.021 206.7 213.2 Source: Author’s Construct Trend analysis of climatic variables Source: Author’s Construct Fig. 2Trends in relative humidity Trends in relative humidity As indicated in the series plot in Fig. 2, relative humidity exhibited a consistent seasonal trend over the study period. Fig. 3Trend in temperature Trend in temperature As indicated in the series plot in Fig. 3, the temperature exhibits a fluctuating trend that slowly wanders upward. The series plot indicates seasonality with no obvious outliers and no sudden shift in the trend with the peak temperature recorded around May 2017. Fig. 4Trend of rainfall Trend of rainfall The series plot of rainfall in Fig. 4, for the time period indicates seasonality with no outliers and sudden shifts with no consistent trend. Unit root test The KPSS test (Table 5) showed that relative humidity, temperature, and rainfall were stationary in levels (I(0)), while malaria morbidity was non‑stationary in levels but became stationary after first differencing (I (1); Table 6). This mixed order of integration precluded conventional cointegration tests and justified the use of the ARDL bounds approach. Table 5KPSS test of the series in levelsVariableTest with ConstantConstant + TrendTest StatisticCritical ValueTest StatisticCritical ValueRelative Humidity0.0190.4670.0160.148Temperature0.0830.4670.0320.148Rainfall0.0480.4670.0190.148Malaria Morbidity0.7610.4670.1620.148Source: Author’s Construct Table 5KPSS test of the series in levelsVariableTest with ConstantConstant + TrendTest StatisticCritical ValueTest StatisticCritical ValueRelative Humidity0.0190.4670.0160.148Temperature0.0830.4670.0320.148Rainfall0.0480.4670.0190.148Malaria Morbidity0.7610.4670.1620.148Source: Author’s Construct KPSS test of the series in levels Source: Author’s Construct Table 6KPSS Test of the first difference of the seriesVariableTest with ConstantConstant + TrendTest StatisticCritical ValueTest StatisticCritical ValueRelative Humidity0.0350.4670.020.148Temperature0.0180.4670.0120.148Rainfall0.0330.4670.0210.148Malaria Morbidity0.0610.4670.0320.148Source: Author’s Construct Table 6KPSS Test of the first difference of the seriesVariableTest with ConstantConstant + TrendTest StatisticCritical ValueTest StatisticCritical ValueRelative Humidity0.0350.4670.020.148Temperature0.0180.4670.0120.148Rainfall0.0330.4670.0210.148Malaria Morbidity0.0610.4670.0320.148Source: Author’s Construct KPSS Test of the first difference of the series Source: Author’s Construct Autoregressive distributed lag (ARDL) cointegration Because variables were of mixed integration order (I(0) and I (1)), we employed the ARDL bounds test (Table 7). All computed F‑statistics exceeded the upper critical bounds at the 95% confidence level (and 99% for temperature), rejecting the null hypothesis of no cointegration. Therefore, stable long‑run equilibrium relationships exist between each climate variable and malaria morbidity in Tamale, validating the ARDL framework for estimating short‑ and long‑run coefficients. Table 7ARDL cointegration test between malaria morbidity and climate variables90% CL95% CL99% CLp-valueI (0)I (1)I (0)I (1)I (0)I (1)I (0)I (1)Model4.1084.8655.0455.8787.1768.1440.0020.012 Computed F-statistics Decision Malaria Morbidity and Relative Humidity9.694CointegrationSignificantMalaria Morbidity and Temperature11.198CointegrationSignificantMalaria Morbidity and Rainfall9.205CointegrationSignificantCritical bounds (95% CL): I(0) = 5.045, I (1) = 5.878. Source: Author’s construct Table 7ARDL cointegration test between malaria morbidity and climate variables90% CL95% CL99% CLp-valueI (0)I (1)I (0)I (1)I (0)I (1)I (0)I (1)Model4.1084.8655.0455.8787.1768.1440.0020.012 Computed F-statistics Decision Malaria Morbidity and Relative Humidity9.694CointegrationSignificantMalaria Morbidity and Temperature11.198CointegrationSignificantMalaria Morbidity and Rainfall9.205CointegrationSignificantCritical bounds (95% CL): I(0) = 5.045, I (1) = 5.878. Source: Author’s construct ARDL cointegration test between malaria morbidity and climate variables Critical bounds (95% CL): I(0) = 5.045, I (1) = 5.878. Source: Author’s construct Short run effects In the short run, all three climate variables were positively and significantly associated with malaria morbidity in Tamale (Table 8). Specifically: A one-unit (1%) increase in relative humidity was associated with an increase of 1.08 cases per 1,000 population in malaria morbidity (p < 0.001). A one-unit (1 °C) increase in temperature was associated with an increase of 0.97 cases per 1,000 population in morbidity (p < 0.001). A one-unit (1 mm) increase in rainfall was associated with an increase of 0.31 cases per 1,000 population in morbidity (p < 0.05). All error correction terms (ECT) were negative and statistically significant (ranging from − 0.101 to − 0.254), confirming rapid adjustment back to long‑run equilibrium after a short‑run shock. For instance, the ECT of − 0.254 for the relative humidity–morbidity model implies that approximately 25% of any disequilibrium is corrected within one month. Table 8Estimates of short-run coefficients (dependent variable: malaria morbidity rate)VariableEstimateStd ErrorZ-valuePr(> z)Intercept39.386.7175.8631.81 × 10− 7 ***L (∆ Malaria Morbidityt)0.0020.0011.9560.0550.∆ Relative Humidityt−21.0840.1248.7741.59 × 10− 12 ***ECT -0.254 0.07 -3.615 0.0006*** Intercept22.274.275.3141.49 × 10− 06 ***L (∆ Malaria Morbidityt)-30.61.58 × 10− 04-0.1630.871∆ Temperaturet−20.9740.1247.8735.95 × 10− 11 ***ECT -0.243 0.126 -1.931 0.058 Intercept11731.43.7220.0004***L (∆ Malaria Morbidityt)0.0139.29 × 10− 031.4010.166∆ Rainfallt−30.3070.1322.3230.0240*ECT -0.101 0.037 -2.77 0.0080** Source: Author’s Construct Table 8Estimates of short-run coefficients (dependent variable: malaria morbidity rate)VariableEstimateStd ErrorZ-valuePr(> z)Intercept39.386.7175.8631.81 × 10− 7 ***L (∆ Malaria Morbidityt)0.0020.0011.9560.0550.∆ Relative Humidityt−21.0840.1248.7741.59 × 10− 12 ***ECT -0.254 0.07 -3.615 0.0006*** Intercept22.274.275.3141.49 × 10− 06 ***L (∆ Malaria Morbidityt)-30.61.58 × 10− 04-0.1630.871∆ Temperaturet−20.9740.1247.8735.95 × 10− 11 ***ECT -0.243 0.126 -1.931 0.058 Intercept11731.43.7220.0004***L (∆ Malaria Morbidityt)0.0139.29 × 10− 031.4010.166∆ Rainfallt−30.3070.1322.3230.0240*ECT -0.101 0.037 -2.77 0.0080** Source: Author’s Construct Estimates of short-run coefficients (dependent variable: malaria morbidity rate) Source: Author’s Construct Long-run effects In the long run, only temperature retained a statistically significant association with malaria morbidity (Table 9). A one-unit (1 °C) increase in temperature was associated with a decrease of 0.37 cases per 1,000 population in malaria morbidity (p = 0.038). Relative humidity and rainfall showed no significant long‑run effects in the final models. Table 9Estimates of long-run coefficients (dependent variable: malaria morbidity rate)VariableEstimateStd ErrorZ-valuePr(> z)L (∆ Malaria Morbidityt)Relative Humidityt−228.517.871.60.115Temperaturet−2-0.370.177-2.110.038**Rainfallt−35.4614.0981.330.187Source: Author’s Construct Table 9Estimates of long-run coefficients (dependent variable: malaria morbidity rate)VariableEstimateStd ErrorZ-valuePr(> z)L (∆ Malaria Morbidityt)Relative Humidityt−228.517.871.60.115Temperaturet−2-0.370.177-2.110.038**Rainfallt−35.4614.0981.330.187Source: Author’s Construct Estimates of long-run coefficients (dependent variable: malaria morbidity rate) Source: Author’s Construct Model diagnosis For all malaria morbidity models, the Ljung‑Box test (p > 0.05) and ARCH‑LM test (p > 0.01) indicated that residuals were white noise, free from serial correlation and heteroscedasticity (Table 10). This supports the validity of the morbidity model estimates. Variance inflation factors (VIF) for all independent variables were below 5 (range: 1.24–3.87), indicating no problematic multicollinearity. Table 10Ljung-box test and ARCH-LM test of models between malaria morbidity, mortality, and climatic variablesModelLjung Box TestARCH-LM TestTest StatisticP-ValueTest StatisticsP-ValueMalaria Morbidity and Relative Humidity6.1880.79922.930.011Malaria Morbidity and Temperature11.220.34118.560.046Malaria Morbidity and Rainfall14.450.15322.590.012Source: Authors’ Construct Table 10Ljung-box test and ARCH-LM test of models between malaria morbidity, mortality, and climatic variablesModelLjung Box TestARCH-LM TestTest StatisticP-ValueTest StatisticsP-ValueMalaria Morbidity and Relative Humidity6.1880.79922.930.011Malaria Morbidity and Temperature11.220.34118.560.046Malaria Morbidity and Rainfall14.450.15322.590.012Source: Authors’ Construct Ljung-box test and ARCH-LM test of models between malaria morbidity, mortality, and climatic variables Source: Authors’ Construct Discussion This study provides the first ARDL‑based evidence from the Tamale Metropolitan Area that climatic variability exerts significant short‑run effects on malaria morbidity, with temperature showing opposing long‑run associations. These findings extend the climate–malaria literature by demonstrating how the same climate variable can have opposite short‑ and long‑run effects in a distinct seasonal setting like northern Ghana. The consistent seasonal pattern of relative humidity observed in Tamale contrasts with reports from Manga, Upper East Ghana, where humidity fluctuated erratically [21]. This discrepancy likely reflects differences in geographical scope and microclimate rather than methodological variation. Temperature in Tamale showed a gradual upward trend without structural breaks, consistent with regional West African warming patterns [22]. Rainfall exhibited pronounced seasonality but no long‑term trend, contrasting with studies that report declining rainfall elsewhere in Ghana [23, 24]. This spatial heterogeneity underscores the importance of local‑level climate data for malaria modelling—a key justification for the Tamale case study. Our analysis confirms that rainfall, temperature, and relative humidity serve as significant short-run predictors of malaria morbidity, though long-run effects are limited to temperature. The sensitivity of malaria to rainfall aligns with findings from the Democratic Republic of Congo [25], South Africa [26], China [27], Rwanda and Uganda [28], and Ghana [29]. The positive short‑run associations we observed are biologically plausible. In Tamale, rainfall creates transient breeding pools for Anopheles mosquitoes [30]. Relative humidity above 60% (typical in Tamale’s wet season) reduces vector desiccation and supports Plasmodium sporogony. Temperature operates as a thermodynamic regulator: the short‑run positive effect (0.97 cases per 1,000 population per 1 °C) aligns with optimal ranges of 25–30 °C, which accelerate parasite development and mosquito biting rates [7, 8]. Not all studies agree. Some report non‑significant [32] or negative [31, 34] climate–malaria associations, including research from Ghana’s forest zone where only humidity was predictive [33]. These discrepancies likely reflect geographical heterogeneity in vector species, baseline immunity, health system capacity, and non‑linear threshold effects. Importantly, climate represents only one component of malaria transmission dynamics. Socioeconomic determinants, including poverty, healthcare access, sanitation, and population density, interact with climatic factors [35, 36]. Intervention coverage (insecticide-treated nets, seasonal malaria chemoprevention, indoor residual spraying) and human mobility further modulate transmission. Effective vector control and emerging vaccination programs may substantially mitigate climate-related transmission risks [37], emphasizing the need for integrated, multi-sectoral approaches. Four main limitations affect interpretation. Omitted variable bias: low R² values for exploratory models indicate missing predictors, notably intervention coverage, health system capacity, and population mobility. These factors were unavailable for inclusion. Causality and generalisability: temporal precedence supports association, but causality cannot be established without quasi‑experimental designs. The single‑region focus limits extrapolation to other ecological zones in Ghana. Overfitting risk: with 84 observations and up to 8 lags, parsimony was prioritized; the final morbidity model uses fewer lags [7, 8]. Readers should interpret coefficients as associations only. Structural breaks: while no significant breaks were detected in the morbidity series, unmeasured breaks (e.g., policy changes, outbreak events) could affect parameter stability. Future studies in Tamale and similar settings should collect and incorporate intervention coverage and health system metrics, apply distributed lag non‑linear models (DLNMs) to capture delayed and threshold effects, use regime‑switching or GARCH models to address volatility clustering, and extend the study period to include post-2020 data to assess pandemic-related disruptions. Despite limitations, three priority actions emerge. Climate‑informed early warning should integrate seasonal climate forecasts (temperature, rainfall) with routine surveillance to trigger pre‑emptive interventions (e.g., vector control, stockpiling) 1–2 months before peak transmission in Tamale’s single rainy season. Health system strengthening is essential because climate adaptation alone is insufficient; investments in universal health coverage, peripheral facility staffing, and malaria commodity security are critical. Finally, improved surveillance with disaggregated data (age, sex, socioeconomic status, treatment delay) and reliable morbidity registration are needed to validate future models. Conclusion Climatic variables significantly influence malaria transmission in the Tamale Metropolitan Area, with differential short-run versus long-run effects. Climate represents an important, but not deterministic, component of malaria transmission ecology. Effective control requires integrating climate adaptation with robust health systems, socioeconomic development, and targeted interventions, while investing in enhanced surveillance and analytical frameworks to better quantify and respond to climate-health linkages.