Introduction

Global climate dynamics have long been the focal point of extensive climatological research, with numerous studies highlighting the profound impacts of external factors such as greenhouse gas concentrations, oceanic circulations, and large–scale atmospheric patterns on regional air temperature changes1,2,3,4,5. These external influences, often operating globally, dictate long-term climate trends and variability, overshadowing the intricate and localized processes that directly affect regional climate. This global-centric view has often left a gap in understanding how local environmental variables contribute to climate variability and change6,7, particularly in regions with complex climatic interactions. Global factors, such as greenhouse gas concentrations and large-scale atmospheric patterns, are critical in shaping regional climates. These drivers often influence changes associated with anthropogenic climate change. However, there is increasing recognition of the significant role that local environmental variables play in determining near-surface air temperature (NST)8,9,10,11. The main drivers of NST can differ locally, and therefore, it is important to understand the drivers for different ecosystem management.

The West African sub-region is a major hotspot for climate change research due to its significant land-atmospheric and ocean-atmosphere interactions12,13,14. This region has been highlighted as particularly vulnerable to climate change impacts, making it crucial to understand its unique climatic dynamics15. West Africa is pivotal in smoothly regulating the Intertropical Convergence Zone (ITCZ), essential for maintaining global climate stability15,16,17. For instance, changes in the Sahara and Sahel influence dust storms and their global radiative effects18,19. At the same time, the Guinea Coast’s experiences with intense precipitation and flooding have regional and local implications20. Previous studies21,22,23 have suggested that changes in mean temperatures largely account for the variations in observed extreme temperatures. For instance, the IPCC’s sixth assessment report (2021) states that the global NST has increased by 1.1 °C above pre-industrial levels from 2011 to 2020. This rise significantly impacts the climate system, causing more frequent and severe extreme events with each increment of warming. Consequently, numerous studies have explored the possible impact of the increase in extreme events across various regions, including West Africa24,25,26. These studies have mainly focused on the globe or other regions, with relatively few studies specifically exploring the changes in West Africa. However, like many other countries in West Africa, Ghana has been experiencing significant increases in NST over recent decades27.

Understanding the characteristics of local climates and their variability is crucial for developing effective adaptation and response strategies, particularly in the context of environmental management and policy development28. Enhancing our knowledge of surface-atmosphere feedback effects at these scales is essential not only for improving future climate projections but also for ensuring preparedness against climate variabilities and extremes28. In Ghana, where climate variability directly impacts agriculture, water resources, and overall socio-economic stability, it is essential to identify and quantify the local climatic factors influencing NST. Existing studies across the globe and in Africa have noted the significant impact of local factors in driving temperature variability and trends, resulting in an increment in extreme events9,29,30,31.

This study diverges from the conventional approach, which primarily attributes NST changes to non-localized external forces32. Instead, it builds upon prevailing research establishing connections between NST and various atmospheric, environmental, and climatic variables such as land surface temperature (LST), total cloud cover (TC), relative humidity (RH), and precipitation. For instance, LST exhibits a direct and strong positive link with NST, as both respond to similar environmental conditions. A complex relationship exists between RH and NST, with high humidity often associated with cooler daytime temperatures due to reduced surface heating and warmer nighttime temperatures due to enhanced longwave radiation trapping33. TC modulates solar and longwave radiation processes, affecting NST differently at day and night26. This dual effect explains the complexity of cloud cover’s impact on temperature. Precipitation also plays a significant role, with dry conditions leading to higher temperatures and wet conditions cooling them34. These findings emphasize the multifaceted drivers behind NST variability, shedding light on how atmospheric composition, land use changes, and global climate systems influence local temperature patterns35.

While most research employs pairwise correlation to examine the relationships between NST and other climatic variables32,36,37, our work introduces a novel methodological approach by combining pairwise correlation and multivariate causality analysis. The Information flow (IF) based causality tool quantitatively evaluates the cause-effect relationship in time series38. This method has been successfully implemented in environmental science39, climate science40,41, and finance42. These studies demonstrate the usefulness of the multivariate Liang-Kleeman (LK) causality in practical applications. Therefore, the study combines pairwise correlation and the Liang-Kleeman Information Flow (LKIF) method for multivariate causality analysis, providing a new quantitative framework to evaluate complex interactions and causality among climatic variables, which is rarely done in existing studies. This dual approach seeks to answer the questions: What role do local environmental and climatic variables play in determining NST changes in Ghana? Which variables are primarily responsible for driving this change?

The study enriches the broader scientific dialogue surrounding regional climate dynamics by delving into these intricacies. Importantly, it provides policymakers and stakeholders with a vital reference point for crafting climate adaptation frameworks and management tailored to the unique characteristics of local climate systems. These strategies align with the United Nations Sustainable Development Goals (SDGs), particularly those related to climate action (SDG 13), sustainable cities (SDG 11), life on land (SDG 15), and zero hunger (SDG 2). We offer a novel methodological approach that evaluates complex interactions and causality among climatic variables by integrating pairwise correlation and multivariate causality analysis. These focused studies form an indispensable foundation for constructing resilient and impactful climate action strategies to safeguard our planet’s future.

Methods

Study area

The study area, delineated in Fig. 1, is located between longitudes 3.5° W to 1.5° E and latitudes 4.5° N to 11.5° N, covering the diverse ecological landscape of Ghana in West Africa. The Gulf of Guinea borders Ghana to the south, Burkina Faso to the north, Côte d’Ivoire to the west, and Togo to the east. The total land area of Ghana is approximately 239,460 km², of which 8,520 km² comprises water bodies. This region features various climatic zones, from the arid north to the humid coastal south, each supporting distinct ecosystems and agricultural practices43. Mount Afadjato, the highest point in Ghana at 878 m, is part of the Akwapim-Togo Ranges. Mean annual temperature variations across the country range from 26.1 °C along the southwest coast to 28.9 °C in the northeast, bordering the Upper East region44. It should be noted that temperatures in the Upper East region occasionally reach up to 40 °C. The diverse climatic conditions delineate two primary seasons: a hot season from December to February and a cooler season from June to August. The agroecological divisions include the Sudan Savannah zone, Guinea Savannah zone, transition zone, moist evergreen, deciduous forest zone, wet evergreen zone, and coastal savannah zone, highlighting the vast agricultural potential that covers 57% of the nation’s land area43. Ghana experiences various precipitation patterns influenced by its geographical and climatic diversity. Mean annual precipitation varies significantly across the country, ranging from approximately 1000 mm in the northeast boundary of the Upper East region to 2,200 mm in the southwest shore of the Western region. The coastal zone, by contrast, is relatively dry, with an annual precipitation average of about 800 mm44,45. These precipitation variations are critical in defining the country’s agroecological zones, water resources, and overall socio-economic activities.

Fig. 1
figure 1

Source: USGS Earth Explorer.

Map of the study area. a Agroecological zones and the 16 administrative regions of Ghana a location of Ghana in Africa. The digital elevation model (DEM) dataset is obtained from the Shuttle Radar Topography Mission (SRTM), with a 90-m spatial resolution.

Full size image

Data

The present study uses monthly gridded surface temperature data from the Climatic Research Unit (CRU) Time-Series version 4.05 (CRU TS v.4.05), produced by the University of East Anglia. This dataset, which extends from 1901 to 2020 on a horizontal grid resolution of 0.5°× 0.5°, is fundamental for understanding long-term climate variability and change46. Furthermore, the study utilizes the gridded temperature dataset from the Climate Prediction Centre (CPC) to assess and confirm the consistency of trends and shifts in mean temperature over the study region. The CPC dataset comprehensively covers global land surface conditions on a 0.5°× 0.5° from 1979 to the present-47. A previous work48 identified CRU and CPC as optimal substitutes for in-situ data in climate change studies in Ghana, supporting our choice of these datasets for analysis.

To investigate the possible relationship between the local climatic variables and the changes in mean temperatures, we use ERA5 datasets provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). ERA5 represents the fifth generation of ECMWF’s global atmospheric reanalysis, from 1950 to the present, on a 0.25°× 0.25° grid. ERA5 provides hourly updates on atmospheric, land-surface, and sea-state parameters49. This study uses LST, TC, and RH. These variables are vital for understanding the energy and moisture fluxes affecting NST. Additionally, we use the Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) dataset as the precipitation dataset to analyze the drivers of NST. A gridded quasi-global merged precipitation dataset from satellite and in-situ observations to analyze precipitation patterns50. This dataset offers high-resolution daily precipitation records at 0.05°× 0.05° from 1981 to the present, which is essential for assessing hydrological impacts on climate. All datasets are processed to ensure consistent spatial and temporal resolutions. All the datasets are resampled to a spatial resolution of 0.5° × 0.5° and a temporal resolution of 1981 to 2020. Table S3 provides a detailed description of all the datasets used, including their spatial resolutions and sources.

Anomalies

Anomalies are calculated relative to the mean and standard deviation of the new climate normal period, following established methodologies23,51. This standardization facilitates comparative analysis across different climatic parameters and datasets, providing a contemporary baseline that reflects recent climatic conditions. Robust statistical methods integrate these diverse data sources, ensuring causality assessments are based on reliable and consistent data inputs. The anomalies are calculated using the following:

Where Ti, j represent the observed temperature for year i and month j, and \(\:\stackrel{-}{{T}_{\text{j}}}\) the climatological mean temperature for month j over the reference period (1991–2020). The temperature anomaly \(\:{A}_{i,j}\) for month j in year i given by:

$$\:{A}_{i,j}=\:{T}_{i,j-\:}\:\stackrel{-}{{T}_{\text{j}}}$$
(1)

For annual anomalies, the monthly anomalies were averaged to obtain the annual temperature anomaly \(\:{A}_{i}\) for the year i.

The annual anomalies for each year within a decade are averaged to compute decadal anomalies. Where \(\:\frac{1}{12}\) represents the averaging factor used to compute the annual temperature anomaly

$$\:{A}_{i}=\frac{1}{12}\:{\sum\:}_{j=1}^{12}{A}_{i,j}$$
(2)

where, Ad, k represent the decadal anomaly for decade d, where k indexes the years within the decade and \(\:\:\frac{1}{n}\)denotes the averaging factor for calculating the decadal anomaly. It is represented mathematically as:

$$\:{A}_{d,k}=\frac{1}{n}\:{\sum\:}_{k=1}^{n}{A}_{d,k}$$
(3)

Climatic breakpoints and shifts

We use the Pettitt test to detect a singular change point in the annual temperature anomaly time series from 1981 to 2020 52. This non-parametric statistical test identifies whether there is a significant change point in the dataset where the statistical properties, such as the mean, shift53. This test is particularly useful for assessing abrupt changes in climatic data, as it does not require any assumptions about the data distribution54. Due to its sensitivity to changes within a time series, Pettitt’s test has become widely used for change point detection53,54. The Pettitt test ranks the data and calculates a test statistic to identify the maximum difference between two sub-series (before and after the change point). If this test statistic exceeds a critical value corresponding to a significance level of 0.05, the null hypothesis (no change) is rejected, indicating a significant change point53.

To validate the results of the Pettitt test, we also utilize the Welch Two Sample t-test and the Kolmogorov-Smirnov test55,56. The Welch Two Sample t-test compares the means of two independent samples, specifically the periods 1981–2000 and 2001–2020, without assuming equal variances, making it more robust under unequal variance conditions55. The Welch Two Sample t-test provides several key outputs. The t-value measures the difference between the sample means relative to the variability within the samples. Degrees of freedom (df) account for differences in sample sizes and variances, ensuring robust statistical testing. Finally, the p-value indicates the probability that the observed difference between sample means occurred by chance55. If the p-value is less than 0.05, the null hypothesis that the means of the two populations are equal is rejected. The Kolmogorov-Smirnov test is another non-parametric test that compares two distributions or a sample with a reference probability distribution57. In this context, the Kolmogorov-Smirnov test compares two independent samples, specifically the periods 1981–2000 and 2001–2020, to determine if they are drawn from the same continuous distribution. The test outputs include the D-statistic, the maximum difference between the two samples’ empirical cumulative distribution functions (ECDFs), and the p-value, which indicates the probability that the observed difference between the ECDFs occurred by chance. If the p-value < 0.05, the null hypothesis that the samples are drawn from the same distribution is rejected.

Trend analysis

Trends are calculated using the Modified Mann–Kendall test (MMK), a robust non-parametric method that detects the presence of a monotonic tendency in a chronological series of variables57. This test does not require any assumptions about the underlying distribution of the data, making it particularly useful for environmental time series analysis58. The MMK test provides three key information. The first is Kendall’s Tau (KT), which is the Kendall rank correlation coefficient, which measures the monotonicity of the slope57. KT ranges from − 1 to 1, with positive values indicating an increasing trend and negative values indicating a decreasing trend. The second one is the Sen’s Slope, which estimates the overall slope of the time series. It is calculated as the median of all slopes between each pair of points in the series, providing a measure of the rate of change. The third one is the significance, which represents the threshold at which the null hypothesis (that there is no trend) is rejected. The trend is considered statistically significant if the p-value < 0.05. We performed the MMK test on all the variables considered in this study at annual and seasonal levels before and after 2000.

Multivariate causality formalism

The causal inference technique used here is from Liang’s IF and causality analysis theory rigorously derived from first principles38,42,59 and expressed in terms of sample covariances. Given two-time series X2 (e.g., NST) and X1 (e.g., LST), the causality from the former to the latter proves to be measurable by the time rate of the flow of information from X2 to X1, which is given by

$$\:{T}_{2\to\:1}\:=\:\frac{{C}_{12}}{{C}_{11}}\:.\frac{-\:{C}_{12}{C}_{1,d1}+{C}_{11}{C}_{2,d1}\:}{{C}_{11}{C}_{22\:}-{{C}^{2}}_{12}}$$
(4)

where Cij is the sample covariance between Xi and Xj, Ci, dj the covariance between Xi and Xdj, and Xdj a derived series from Xj using the formula in Eq. (4a)

$$\text{X}_{\text{dj}}=[\text{X}_{\text{j}}(\text{t}+\text{k}\Delta \text{t})-\text{X}_{\text{j}}(\text{t})]/(\text{k}\Delta \text{t})$$
(4a)

, the Euler forward differencing, where Δt is the timestep size, and k some positive integer; see38 for further details. Based on the discussions in38, we choose k = 1. Similarly, switching the indices can obtain the IF from X1 to X2. By a theorem called the property of causality in59, and later on called the principle of nil causality42, if T2→1 is nonzero, then X2 is causal to X1, and noncausal if it vanishes, and the magnitude of T2→1 measures the strength of the causality. Further, according to Liang38, positive IF from X2 to X1 implies that significant variations in X2 will amplify variations (referred to as amplifications here), thus sustaining anomalous conditions in X1. This also implies increases in the entropy of X1 and, therefore, increased unpredictability38. In contrast, negative IF suggests that variations in X2 reduce or stabilize variations in X1, essentially due to decreases in the entropy of X1. So X1 remains at and returns to equilibrium due to changes in X2. This interpretation differs from the positive and negative paradigms typically used in regression analysis. Liang’s causality formalism has been demonstrated to be effective in deterministic and stochastic systems38,42,59.

$$\:{T}_{2\to\:1|\text{3,4},.,d}\:=\:\frac{1}{detC}\:.{\sum\:}_{j=1}^{d}{\varDelta\:}_{2j}{C}_{j,d1\:}.\frac{{C}_{12}}{{C}_{11}}$$
(5)

From (4) and (5), it can also be seen that when c12 = 0, \({T_{2 \to 1}}\) = 0 always; however, the vice versa is not true. Thus, correlation does not imply causation but, fundamentally, association60. On the contrary, causation inherently implies correlation. Although the strong correlation between NST and LST might complicate causal analysis, Liang61 has demonstrated that the formalism remains reliable even in systems with nearly synchronized chaotic oscillators or heavy noise. These attributes are crucial for this study. Further details on the formalism can be found in 43 63.

To rank the importance of the IF results from (5), we normalize them following61. This allows us to fairly compare the IF results from the different emission pathways. Additionally, the normalized IF reflects the importance of the IF of, say, X2 to X1 relative to another process. Liang42, proposed using a normalizer, Z, which indicates the total entropy rate of change due to all the influences acting on the X1, including those beyond X2 and X1. The normalized IF is expressed as T2→1|3,4,…,d./Z.

Test for statistical significance of (4) and (5) is obtained with the Fisher Information Matrix. Given a large sample size, N, of the time series, the IF in (4) and (5) approaches a normal distribution around its true value with a variance obtained from the maximum likelihood estimation. Here, we compute statistical significance at a 5% level. For details, readers are referred to38,62.

Results

Interannual variability and decadal anomalies of mean temperature

Both datasets show an increasing trend in mean temperature, confirming regional responses to global climate change. The CPC dataset shows a rate of increase of 0.25 ± 51.14 °C per decade, while CRU increases by a slightly smaller rate of 0.16 ± 32.77 °C per decade (Fig. 2).

Fig. 2
figure 2

Interannual variability of mean annual temperature anomaly over Ghana from 1981 to 2020 (unit: °C) based on CRU (red) and CPC (blue). Dashed lines represent linear trend lines. The anomalies are calculated relative to the 1991–2020 climate normal period.

Full size image

A positive correlation (0.81) between the CRU and CPC temperature anomalies highlights the strong agreement between these datasets in capturing year–to–year temperature fluctuations (Fig. 2). This agreement enhances the confidence in the results observed. The root mean square deviation (RMSD) between the datasets is approximately 0.230, indicating modest differences in temperature anomalies. Despite these differences, the consistent upward trends and pronounced phases of cooling and warming, especially the significant warming observed in the last two decades, align with the broader global warming trends.

The linear regression lines that illustrate the temperature trends showed consistently rising temperature trends, with the last two decades showing marked increases of approximately 0.6 °C for CRU and 0.7 °C for CPC. This comparison highlights the shift from cooling to warming phases and emphasizes the importance of local climate dynamics as part of the global climate system.

The CRU data for 1981–1990 shows predominantly negative temperature anomalies ranging from − 0.6 °C to -0.2 °C (Fig. 3). During 1991–2000, negative anomalies persist (-0.6 °C to 0 °C), albeit with slightly smaller magnitudes than the previous decade. The 2001–2010 decade marks a significant shift, with temperature anomalies shifting to positive values across most regions, ranging from 0 °C to 0.4 °C. This positive anomaly pattern intensified in 2011–2020, with widespread positive anomalies (0.2 °C and 0.6 °C), suggesting an increase in temperature in the last decade relative to the 1991–2020 climate normal period.

Fig. 3
figure 3

Spatial distribution of change in decadal mean temperature anomaly over Ghana from 1981–2020, relative to the 1991–2020 climate normal period. a 1981–1990, b 1991– 2000, c 2001– 2010, and d 2011– 2020.

Full size image

Similarly, the CPC data for 1981–1990 indicated negative temperature anomalies, predominantly between − 0.9 °C and − 0.3 °C (Fig. 3). The period from 1991 to 2000 maintains this negative anomaly distribution, with values ranging from − 0.8 °C to 0 °C. The 2001–2010 decade sees a noticeable shift towards positive anomalies, with values ranging from − 0.3 °C to 0.5 °C, indicating potential warming relative to the 1991–2020 period. The strongest positive anomalies (0.4 °C to 0.9 °C) appear in 2011–2020, indicating a substantial increase in mean temperature across the country relative to 1991–2020 (Fig. 3).

Both datasets consistently depict a transition from negative to positive temperature anomalies over the four decades. In 1981–2000, both datasets indicated negative temperature anomalies, reflecting a period of relative coolness. The transition period of 2001–2010 marked a notable change characterized by clear shifts towards positive temperature anomalies. This warming trend has become more pronounced in the recent decade (2011–2020), with widespread positive anomalies indicating substantial increases in mean temperatures across Ghana.

This analysis of decadal climatological mean anomalies from 1981 to 2020 reveals significant shifts in temperature trends, particularly between two distinct periods. The data from the last two decades (2000–2020) exhibit positive mean temperature anomalies compared to the earlier two decades (1981–2000). These observations require further examination of potential shifts in climatic baselines within the study region.

Analysis of climatic breakpoints and shifts in temperature anomalies

The results show that 2000/2001 as a pivotal year, marking a significant transition in climatic patterns (Figure S1). This change point is identified without the interference of multiple change points, which can often complicate trend analysis. This finding is supported by statistical analyses of mean, variance, and skewness for the periods before (BC; 1981–2000) and after the change (AC; 2001–2020) shown in Table 1. In this study, BC refers to the period before the change (1981–2000), and AC refers to the period after the change (2001–2020). These results prove that the mean temperature anomaly increases by 0.266 °C from BC to AC, indicating higher mean temperatures after 2001.

For 1981–2000, the mean temperature anomaly was − 0.133 °C, compared to the mean temperature anomaly of 0.133 °C for 2001– 2020, highlighting a possible warming between these two periods (Table 1). Additionally, the variance of temperature anomaly decreases from 0.024 in BC to 0.017 in AC, suggesting more consistent temperature in the later period. Skewness also decreases from 0.588 to 0.119, reflecting a more symmetrical distribution of temperatures with fewer extreme highs in the latter period.

Table 1 Summary statistics of annual temperature anomalies for the periods 1981–2000 (before change) and 2001–2020 (after change).
Full size table

The Welch Two Sample t-test revealed significant differences in mean temperature anomalies between the two periods, 1981–2000 and 2001–2020 (t = 5.881, degrees of freedom = 36.829, p < 0.0001). Similarly, the Kolmogorov–Smirnov test showed a substantial distribution divergence (D = 0.75, p < 0.0001), highlighting the distinct nature of temperature distributions before and after the change point. Violin plots highlighted the increased frequency and intensity of positive anomalies in the latter period, providing insights into the increased variability and the transition from a modestly right-skewed distribution in BC to a nearly symmetrical distribution in AC (Fig. 4). Therefore, these findings necessitate a detailed exploration of land and atmospheric drivers to understand their interactions and influence on local climate systems.

Fig. 4
figure 4

Violin plots display annual temperature anomalies distribution for 1981–2000 (blue) and 2001–2020 (red).

Full size image

Associations of various climatic variables with near-surface air temperature across seasonal scales

To understand the potential influence of different climatic factors on the evolution of NST, we investigate the relationship between LST, PRE, RH, TC, and NST for DJF (Fig. 5) and JJA (Fig. 6). During DJF, LST depicted a highly significant positive correlation (r = 0.89, Fig. 5a), confirming its substantial and consistent positive relationship with NST. Blue and red data points are closely aligned along the regression line, showing a consistent relationship in both periods. However, the spread of the red data points suggests a more positive association of LST with NST in the later period. Precipitation presents a slightly positive correlation (r = 0.25, Fig. 5b) but no significant relationship NST during this period. In addition, as revealed in (Figure S2a), PRE revealed a weak positive trend (0.042) in 1981–2000 and a negative trend with a KT of (-0.084) in 2001–2020. RH displays a very significant positive correlation (r = 0.57, Fig. 5c), suggesting higher humidity levels have a positive association with increased NST. The red dots are more densely clustered around the positive trend line, suggesting a stronger and more consistent relationship than the blue data points in the recent period. This relationship is likely mediated by the moisture’s capacity to trap and retain heat, enhancing the greenhouse effect and thus warming the atmosphere. TC demonstrates a weak correlation (r = 0.04, Fig. 5d), suggesting a minimal direct impact on NST. This implies that cloud cover’s cooling and warming effects are likely counterbalanced by other climatic factors, resulting in a complex interaction that negligibly alters NST.

Fig. 5
figure 5

Relationship between local climatic variables and changes in mean temperature over Ghana during the December, January, and February (DJF) season. Scatter plot of the standardized mean temperature anomalies versus the standardized mean anomalies of a land surface temperature (LST), b precipitation (PRE), c relative humidity (RH), d total cloud cover (TC). The “r” represents the correlation coefficient, and “*” symbolizes a significant correlation at a 95% confidence level. The color scheme distinguishes between two periods: BC (1981–2000, blue) and AC (2001–2020, red). The distribution of red and blue data points provide insight into how the relationships between various local climatic variables and NST anomalies have evolved.

Full size image

The JJA season was characterized by cool temperatures and high precipitation with distinct patterns in the relationship between NST and various drivers across the two periods. LST maintained a highly significant positive correlation (r = 0.83, Fig. 6a) with NST, indicating that ground heat has a relationship with air temperature, even during the cool months. Both blue and red data points were aligned along a rising trend line, indicating a strong relationship where higher LST values are associated with higher NST values in both periods. The clustering of red points along the positive slope indicated that this relationship has remained consistent or even strengthened after the change. Precipitation (Fig. 6b) exhibits a weak correlation with NST (r = 0.08), highlighting its minimal direct relationship with NST during the JJA season. Also, Trend analysis from (Figure S2 b) indicates that although this season is the main wet season in Ghana, precipitation has experienced a downward trend (-0.063) in the last two decades. RH showed a significant positive correlation (r = 0.39, Fig. 6c). Both blue and red data points cluster along a positive trend line, showing a strong association where higher RH values correspond to higher NST values. The red points indicate a slightly weaker but still significantly positive relationship after the change. TC shows a slight positive correlation (r = 0.26, Fig. 6d), however, has no significant relationship with NST during this season.

Fig. 6
figure 6

Relationship between local climatic variables and changes in mean temperature over Ghana during the June, July, and August (JJA) season. Scatter plot of the standardized mean temperature anomalies versus the standardized mean anomalies of a land surface temperature (LST), b precipitation (PRE), c relative humidity (RH), d total cloud cover (TC). The “r” represents the correlation coefficient, and “*” symbolizes a significant correlation at a 95% confidence level. The color scheme distinguishes between two periods: BC (1981–2000, blue) and AC (2001–2020, red). The distribution of red and blue data points provide insight into how the relationships between various local climatic variables and NST anomalies have evolved.

Full size image

Annual association between climatic variables and NST variability

Each variable uniquely influenced NST, reflecting direct and indirect climatic interactions (Fig. 7). The strong positive correlation between LST and NST (r = 0.89) indicates that land surface heating contributes significantly to NST (Fig. 7a). Both blue and red data points are aligned along a rising trend line, indicating a strong relationship where higher LST values are associated with higher NST values in both periods. The clustering of red points along the positive slope indicates that this relationship has remained consistent or even strengthened after the change. In contrast, precipitation shows a slight negative correlation but no significant relationship with NST (r = -0.13, Fig. 7b). The RH also revealed a highly significant positive correlation (r = 0.63, Fig. 7c). Both blue and red data points cluster along a positive trend line, showing a strong association where higher RH values correspond to higher NST values. The red points indicate a slightly weaker but still significantly positive relationship after the change. Conversely, TC had no significant relationship with NST with a weak positive correlation (r = 0.19, Fig. 7d).

Fig. 7
figure 7

Relationship between local climatic variables and changes in mean temperature over Ghana for the annual time scale. Scatter plot of the standardized mean temperature anomalies versus the standardized mean anomalies of a land surface temperature (LST), b precipitation (PRE), c relative humidity (RH), d total cloud cover (TC). The “r” represents the correlation coefficient, and “*” symbolizes a significant correlation at a 95% confidence level. The color scheme distinguishes between two periods: BC (1981–2000, blue) and AC (2001–2020, red). The distribution of red and blue data points provide insight into how the relationships between various local climatic variables and NST anomalies have evolved.

Full size image

Causal structure between local climatic variables and NST

To illustrate the influences of the various climatic variables and their separate couplings, their causality on each other and NST from 1981 to 2000 is demonstrated in Fig. 8. It shows the absolute value of the time-varying information flow from LST to TC and that of PRE and NST influenced by RH, as indicated by the arrows to these variables. There is no causality between LST, PRE, and NST since there is no preset feedback from LST on these variables, as shown in Fig. 8. All the results are obtained at a 1% significance level. This network emphasizes the complex interactions among these local climatic variables, with LST playing a central role by directly influencing several variables and indirectly impacting changes in NST. This feedback highlights the intricate role of land-atmosphere coupling, where reduced soil moisture from higher LST suppresses PRE and further amplifies warming effects.

Fig. 8
figure 8

Multivariate causality network among key climatic variables LST: land surface temperature, PRE: precipitation, RH: relative humidity, TC: total cloud cover, and their influence on near-surface air temperature (NST) during 1981–2000 before the change (BC). The arrows indicate the direction of information flow, with thickness visually representing the magnitude of influence between variables.

Full size image

In the period after 2001, our multivariate causality analysis revealed a dense and more complex network of interactions and information flow among the local climatic variables (Fig. 9). This increased interconnectivity suggests a more intertwined climatic system with significant implications for NST regulation. The causal structure identified from LST to NST, PRE, and TC indicates that changes in LST significantly impact them in the AC period. Furthermore, PRE significantly influenced NST through an information flow from PRE to NST (Fig. 9), highlighting that NST changes result from the significant LST impact, which directly affects both NST and PRE, with PRE subsequently influencing NST.

Fig. 9
figure 9

Multivariate causality network among key climatic variables LST: land surface temperature, PRE: precipitation, RH: relative humidity, TC: total cloud cover, and their influence on near-surface air temperature (NST) during 2001–2020 after the change (AC). The arrows indicate the direction of information flow, with thickness visually representing the magnitude of influence between variables.

Full size image

Discussions

The findings highlight significant warming trends, shifts in temperature distributions, and the substantial influence of local climatic drivers on NST. These insights contribute to the broader field of climate change research and highlight the importance of incorporating local environmental factors into climate resilience strategies and policy development to effectively address the challenges posed by global warming.

Decadal warming and trends

The decadal temperature anomalies from CRU and CPC datasets consistently demonstrate a significant warming trend in Ghana from 1981 to 2020. This warming is evident in the shift from negative anomalies in the early decades (1981–1990, 1991–2000) to markedly positive anomalies in the recent decades (2001–2010, 2011–2020). The magnitude of warming is substantial, with CRU data showing a rise from − 0.6 °C to 0 °C in the earlier decades to 0.2 °C to 0.6 °C in the later decades, while CPC data indicates a more pronounced increase from − 0.9 °C to 0 °C (1980–2000) to 0.4 °C to 0.9 °C (2001–2020). Spatially, this warming trend is observed uniformly across Ghana, although the intensity varies slightly between datasets. The reduction in variance and skewness in temperature anomalies during the latter period suggests more stable and symmetrical temperature distributions, indicating a shift towards more consistent and frequent higher temperature occurrences. This result is consistent with other studies in West Africa. For example, Yaro & Hesselberg63 and Christopher63,64 highlighted a rise in mean annual and seasonal temperatures of about 0.3 to 3 °C in the West African region since the mid-1970s. Similarly, Oduro et al.27 applied Coupled Model Intercomparison Project Phase 6 (CMIP6) datasets for Ghana between 1980 and 2014. They observed warming between 0.3 and 0.5 °C, comparable to what was reported in previous work in Ghana65,66.

Possible local climatic variables association with near-surface air temperature

The LST is strongly associated with NST variability, with a strong positive correlation across all seasons and annual data. This relationship underlines the key role of land surface processes in modulating NST. The influence of LST on NST is likely driven by soil moisture dynamics, surface heat fluxes, and the thermal inertia of the land surface. Higher LSTs increase sensible heat flux, warming the NST9. The critical role of LST in driving temperature variability has been well-documented in the literature, emphasizing its importance in regional climate dynamics13,67. Land surface processes drive heat fluxes and longwave radiation exchanges, affecting local and regional surface temperatures. In Ghana, LST is a primary factor contributing to notable changes in NST, particularly in the northern regions where precipitation is minimal. Research by13 demonstrated that LST substantially influences air temperature more than precipitation. The local energy balance enables positive feedback from land as soil moisture deficit increases due to the high demand for water as evaporative cooling decreases68, leading to a further rise in NST.

The RH significantly correlates with NST, particularly in the annual and DJF season. Higher humidity levels are associated with increased NST, likely due to the enhanced greenhouse effect of water vapor, which traps outgoing longwave radiation and warms the surface9. This relationship is consistent across both periods (BC and AC), indicating a stable climatic influence. The positive impact of RH on temperature is supported by literature, highlighting the role of atmospheric moisture in modulating surface energy balances67.

TC displays varying correlations with NST across different seasons and annual scales. In the JJA season, TC shows a moderate positive correlation, suggesting increased cloud cover is associated with higher NST. This warming effect may be due to clouds trapping longwave radiation from the surface, reducing nocturnal cooling, and increasing minimum temperatures69. However, the annual correlation is weaker, indicating a minor influence of cloud cover on NST. The balanced cooling and warming effects of clouds, which depend on cloud type, altitude, and thickness, can lead to this variability in correlations70.

The findings from this study have significant implications for West Africa, where the interplay of local climatic drivers and NST variability can impact various socio-economic sectors. The dominant role of LST and RH in driving NST highlights the need for integrated land and water management practices to mitigate temperature extremes. For instance, sustainable land use practices that maintain soil moisture and enhance vegetation cover can help regulate surface temperatures13.

Multivariate causality analysis and its implications

Multivariate causality analysis provides a comprehensive framework to understand the intricate and interconnected relationships among local climatic variables that influence NST. The multivariate causality analysis highlights that LST is a dominant driver of NST and indirectly affects PRE and RH. This suggests that changes in surface heating alter atmospheric moisture availability, which, in turn, influences precipitation patterns and local humidity levels, creating a complex and interconnected system. The analysis highlights complex, multi-directional interactions among these variables, with LST emerging as a critical driver of climatic dynamics. LST influences RH through its control over surface evaporation and moisture availability. Higher LST often reduces RH by enhancing evaporation, which reduces surface moisture, contributing to increasing NST. The increased influence of precipitation and TC during AC (2001–2020) emphasizes the growing variability in precipitation patterns and cloud dynamics due to climate change. Understanding these complex relationships and feedback mechanisms is crucial for developing effective climate adaptation strategies in Ghana, as it provides a comprehensive framework to inform land management, water resource management, and climate intervention policies. These shifts and changes pose significant threats to biodiversity and ecosystem functioning. Rising NST and LST and decreasing precipitation alter species distributions, potentially leading to habitat loss and extinction71,72. As temperatures deviate from ambient conditions, more species are required to maintain ecosystem functioning, highlighting the importance of biodiversity in the face of environmental change73,74. The interconnected relationships among these climatic variables impact ecosystems and all levels of biodiversity, from genes to biomes, affecting ecosystem structures and services72,75. As global change continues to intensify at the regional levels, it is essential to consider adaptation and mitigation measures to protect biodiversity and maintain ecosystem services by knowing the driving factors of these changes.

Policy recommendation

In light of this study’s findings, Ghana must implement adaptive policies that align with the United Nations SDGs, particularly SDG13, SDG11, SDG15 (Life on Land), and SDG2. The significant reduction in precipitation, the strong correlation between RH and NST, and the marked increase in LST over the past two decades highlight the urgent need for a multifaceted approach to climate resilience.

First, enhancing water resource management is crucial. This is because the reduction in water resources directly impacts agriculture, ecosystems, and overall water resources negatively76, making it essential to implement effective water management strategies. To strengthen the government’s one-district, one-dam policy77, localized rainwater harvesting systems should be introduced in the savanna agroecological zones of Ghana, such as the Northern Savanna, Upper East, and Upper West, where precipitation is minimal, and drought occurrences are increasing. Implement community-based reservoirs to store excess rainwater during the wet season for use in dry periods, ensuring year-round water availability for agriculture and domestic purposes. Additionally, promoting climate-smart agricultural practices (CSA). CSA is emerging as a crucial approach to maintain and increase agricultural productivity in the face of climate change78. CSA aims to enhance food security, improve adaptive capacity, and reduce greenhouse gas emissions. Key CSA practices that will be key in the context of Ghana include conservation agriculture, irrigation, agroforestry, and soil conservation structures. These techniques can help crops adapt to rising temperatures, altered precipitation patterns, and help sustain the Ghanaian government’s planting for food and job initiatives79,80. Encourage the adoption of drought-tolerant crop varieties, such as millet, sorghum, and improved maize hybrids, in the Savannah zones where rising LSTs and declining precipitation are severely impacting yields. Support farmers with training programs on conservation agriculture practices, including mulching, minimum tillage, and intercropping with legumes to maintain soil moisture and fertility. Implementing CSA practices can increase crop yields under unfavorable climate conditions while contributing to carbon sequestration81,82.

The success of SDG13 will depend on accounting for the direct, indirect, external, and local influences of climate change that lead to shifting biodiversity distributions and the associated feedback on ecosystems. To effectively address these challenges, the scientific community should engage the public through citizen science and participatory observation approaches. These methods involve community members in data collection and interpretation, addressing gaps in both data and communication while building local capacity75. When carefully tailored to local issues, these approaches can bridge scientific research and its impact at the local level, which goes a long way in fostering practical and effective management interventions. This collaborative approach can enhance public awareness and engagement with environmental issues, making science more accessible and actionable for broader societal benefit.

The centrality of LST highlights the critical role of land surface processes in modulating NST change. Implement green infrastructure projects in urban centers like Accra and Kumasi, such as developing parks, green roofs, and tree-lined streets. These measures will help counter rising LST, reduce urban heat island effects, and improve air quality83. Launch community-driven afforestation programs in deforested and degraded areas of the transition and forest zones, particularly around regions like Atebubu-Amantin. Planting native species such as Acacia and Teak can help stabilize soil moisture, reduce LSTs, and improve carbon sequestration. By integrating these strategies, Ghana can enhance its resilience to climate change, ensuring that environmental management and conservation efforts are adaptive, inclusive, and aligned with local and global sustainability goals.

Conclusions

This study combines pairwise correlation and the LKIF method for multivariate causality analysis, allowing us to move beyond traditional correlation-based approaches. While correlations identify associations, they do not imply causality; the LKIF method enables us to quantify cause-effect relationships between climatic variables and NST. This provides a more physically-grounded understanding of climate dynamics, essential for reliable predictions and informed climate adaptation strategies. The novelty of our work lies in its emphasis on spatially and temporally localized factors, particularly land-atmosphere couplings, to explain changes in NST. Unlike many existing studies focusing on remote and large-scale drivers, such as global climate variabilities, we address a critical gap by highlighting the role of localized causal drivers. This localized perspective complements broader global analyses and demonstrates the importance of comprehensively integrating local and non-local processes to understand climate variability. Improving predictions of local NST fundamentally relies on understanding causal relationships. Incorporating causal knowledge into predictive frameworks helps design robust, evidence-based mitigation and adaptation strategies, reduce reliance on guesswork, and ensure efficient resource use. By providing a clearer understanding of these processes, this research offers valuable guidance for policymakers and stakeholders, helping them to design effective climate adaptation frameworks tailored to the unique characteristics of local climate systems in Ghana. Additionally, integrating public engagement through citizen science and participatory observation approaches is recommended to enhance data collection and community involvement, further strengthening the resilience of environmental management strategies in response to climate change.

While this study offers significant insights, the following limitations are acknowledged. Using coarser spatial resolution data (0.5° × 0.5°) may limit the representation of fine-scale land features, such as open water bodies and orography, which are important for localized analysis. However, this limitation is mitigated by the LKIF method, which inherently accounts for broader influences. Additionally, we recognize that a comprehensive understanding of climate variability requires integrating local and non-local factors. While this study focuses on localized drivers, follow-up studies are planned to explore the interactions between local and large-scale processes, providing a more holistic perspective on climate variability in the region.

Our findings show that LST plays a central role in driving NST variability while interacting with other climatic variables, such as precipitation and RH. These complex relationships highlight the importance of considering land-atmosphere feedback in climate adaptation strategies.