Trends and oscillation characteristics of hourly PM_2.5 levels in arid environment using wavelet coherence and lagged correlation

Abstract

Understanding the spatiotemporal dynamics of PM_2.5 in arid urban environments is critical for effective air quality management and public health protection. This study investigates the diurnal variability, long-term trends, and climatic interactions of hourly PM_2.5 levels in Kuwait City from 2017 to 2024 using wavelet coherence, lagged correlation, and non-parametric trend analyses. Results show significant diurnal variation in atmospheric loading, with PM_2.5 concentrations peaking during summer evenings at 7 PM (63.3 µg/m³ in July), greatly reducing total solar irradiance (TSI) available for energy conversion. Monthly data indicate higher atmospheric particulate levels during May, July, and August, coinciding with peak solar energy demand periods. Trend analysis using Theil–Sen slopes indicates an overall decline in PM_2.5 levels (− 4.1671 µg/m³/month in September), suggesting improved conditions for solar energy collection. Cross Wavelet Transform analysis uncovered persistent relationships between PM_2.5 and meteorological factors influencing solar panel performance, including temperature, humidity, and wind patterns at 100–300-day scales. Multivariate regression identified wind speed at 50 m (− 2.05, p < 0.001), rainfall (− 0.24, p < 0.001), and TSI (− 0.004, p < 0.001) as significant predictors of atmospheric clarity for renewable energy systems. Lagged correlations confirmed delayed but strong influences of meteorological variables, peaking at lags of 11–30 h. The study provides evidence of the climatic regulation of PM_2.5 in arid urban settings and underscores the need for integrated, weather-informed mitigation strategies to improve air quality and reduce health risks in desert cities.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36902-7.

Introduction

Air pollution, particularly due to fine particulate matter (PM_2.5), remains one of the leading global environmental health risks. PM_2.5 consists of airborne particles with aerodynamic diameters smaller than 2.5 micrometers that can penetrate deep into the human respiratory system, reaching the alveolar regions of the lungs and even entering the bloodstream. Numerous epidemiological studies have established strong associations between PM_2.5 exposure and increased morbidity and mortality from respiratory and cardiovascular diseases, stroke, and lung cancer^1–3. Short-term exposure to PM_2.5 causes approximately 1 million premature deaths per year globally, whereas long-term exposure is estimated to have contributed to more than 4.9 million deaths in 2023⁴. Urban areas account for nearly a quarter of PM_2.5-related short-term mortality, highlighting the disproportionate burden in densely populated regions. In urban areas, PM_2.5 levels are often elevated due to dense traffic, industrial activity, and the urban heat island effect⁵. Air pollution contributes to 1.6 million COPD deaths, around 500,000 lung cancer deaths, and a significant share of global stroke (21%) and cardiovascular (19%) mortality. Additional evidence links long-term PM_2.5 exposure to elevated risks of cardiovascular disease and several cancers⁶. Furthermore, the economic cost of PM_2.5 related health impacts is estimated to exceed 3% of global GDP, according to World Bank and World Health Organization (WHO) reports. Understanding the behavior and sources of PM_2.5 is thus crucial for designing effective mitigation strategies and protecting public health^7–9.

The behavior of PM_2.5 in arid and semi-arid environments differs markedly from that in temperate regions because pollutant levels are shaped by complex interactions between natural and anthropogenic processes¹⁰. Arid cities like Kuwait experience frequent dust events, high wind speeds, temperature extremes, and sparse vegetation, all of which influence particulate matter dynamics in ways that differ from those observed in temperate climates¹¹. Kuwait is particularly susceptible to dust storms, which can significantly elevate PM_2.5 levels and persist for hours or days. Prior studies reported that daily PM_2.5 concentrations often exceeded 150 µg/m³ during dust events^12–14, which is more than 30 times higher than the WHO’s updated 2021 24-hour guideline of 5 µg/m³. In addition, localized sources such as oil refineries, traffic congestion, and power plants contribute to chronic background pollution levels^15,16. Despite these challenges, most studies in the region have focused on annual or daily averages^17,18, overlooking important diurnal patterns and short-term fluctuations that are critical for exposure assessment and forecasting pollution episodes.

Recent advances in data analysis techniques have highlighted the need for higher temporal resolution and more sophisticated tools to examine pollutant dynamics. Traditional correlation or regression models often fail to capture the nonstationary and multiscale relationships inherent in PM_2.5 data, particularly when the data are influenced by both meteorological and anthropogenic drivers¹⁹. Wavelet coherence analysis, a powerful time–frequency method, has emerged as a valuable tool to examine co-movements and localized phase relationships between PM_2.5 concentrations and meteorological parameters such as wind speed, temperature, and relative humidity^19–21. Several studies have successfully used this method in diverse settings. Aguilar-Velázquez and Reyes-Ramírez²² applied wavelet coherence to analyze air quality-meteorological interactions in Mexico, while Liu et al.²³ demonstrated its utility for exploring seasonal dependencies between PM_2.5 and temperature in China. Similarly, lagged correlation analysis²⁴ can account for temporal dependencies and the delayed effects of meteorological conditions on pollution levels, revealing causal dynamics that are often missed by standard approaches. In parallel, trend analysis methods such as the Mann-Kendall test (MKT) and Sen’s slope analysis are more effective for detecting and quantifying monotonic trends in non-parametric datasets^25,26. These methods are increasingly applied in environmental studies, including long-term air quality assessments in various regions around the world, including India^27,28, China^29,30, and sub-Saharan Africa^31,32, where data irregularities and non-normality are common. However, their use in hourly-resolution datasets, particularly in arid environments like Kuwait, remains limited.

Although the dynamics of PM_2.5 have been extensively studied in other regions, Kuwait remains comparatively underrepresented in high-resolution air quality research, despite being one of the most urbanized cities in the Middle East. Most available studies rely on aggregated data and lack investigation of how hourly PM_2.5 levels respond to meteorological variations. This represents a critical gap, as understanding the interaction between PM_2.5 and climatic variables is vital for interpreting its temporal dynamics and chemical transformations in arid settings. Kuwait’s environmental conditions, marked by intense solar radiation, low humidity, and frequent dust storms, may play a significant role in shaping PM_2.5 formation, resuspension, and transport^13,33. However, the extent and scale of these influences remain poorly quantified. Advanced analytical techniques, such as wavelet coherence and lagged correlation, offer powerful tools for exploring scale-dependent and time-lagged relationships between PM_2.5 and meteorological variables, but have rarely been applied in this context. Addressing this research gap is essential for informing adaptive air quality management in hyper-arid urban environments.

This study aims to address critical knowledge gaps by analyzing both short-term oscillations and long-term trends in PM_2.5 concentrations in Kuwait City. Specifically, it (i) characterizes the diurnal and scale-dependent behavior of hourly PM_2.5 concentrations under varying meteorological conditions using wavelet coherence and lagged correlation analysis, and (ii) identifies statistically significant trends over the period 2017–2024 using the Mann-Kendall test, Sen’s Slope, and Theil–Sen estimator. By combining high-resolution datasets with advanced analytical methods, this study provides a more nuanced understanding of PM_2.5 dynamics in arid climates and contributes valuable insight for public health policy, environmental regulation, and climate adaptation strategies in the Gulf region and beyond.

Methodology

Characteristics of the study area

Kuwait, situated at the northeastern edge of the Arabian Peninsula, is characterized by an extremely hot, arid desert climate with recurrent dust storms and strong seasonal winds, which substantially elevate PM_2.5 concentrations (Fig. 1). The country experiences some of the world’s highest temperatures and frequent dust episodes that can raise daily PM_2.5 concentrations above 300 µg/m³ during severe events¹². The capital city, Kuwait City, is the largest metropolitan area and home to more than 3 million people³⁴. Geographically, the city is situated at approximately 29.3759° N latitude and 47.9774° E longitude. As the country’s political, economic, and cultural hub, Kuwait City plays a critical role in shaping national development and urbanization trends³⁵. The city’s urban structure is characterized by dense residential neighborhoods, commercial centers, industrial areas, and expanding suburban developments. Over the past few decades, rapid population growth and large-scale infrastructure projects, driven by oil wealth and modernization ambitions, have significantly altered the city’s landscape³⁶. The proliferation of private vehicles, construction activities, industrial emissions, and energy-intensive lifestyles has contributed to rising levels of air pollutants³⁷.

Fig. 1.

Geographic location of the study area.

Kuwait City experiences an arid desert climate (Köppen climate classification BWh), characterized by extremely hot summers, mild winters, low annual rainfall, and high levels of solar radiation. Summer temperatures routinely exceed 45 °C, with record highs occasionally surpassing 50 °C^17,38,39. The hottest months, from June to September, are marked by prolonged heatwaves, low humidity, and intense solar irradiance. The winter months, from December to February, feature milder conditions, with average temperatures ranging from 10 °C to 20 °C. Rainfall is sparse, highly irregular, and mostly confined to short-lived storm events during the winter months. Relative humidity varies seasonally, generally remaining low during the summer but can increase sharply in coastal areas during transitional months. The combination of extreme heat, intense solar radiation, occasional dust storms, and coastal humidity creates complex atmospheric conditions that have profound implications for air quality dynamics⁴⁰. Moreover, the city’s location along the Arabian Gulf subjects it to additional environmental pressures, such as sea breezes that can transport industrial emissions inland, and dust storms originating from desert regions, which significantly elevate particulate matter levels. These conditions are compounded by climatic extremes of heat and low rainfall, which reduce natural cleansing mechanisms such as precipitation and thereby enhance the persistence of pollutants in the urban atmosphere.

Data acquisition

This study utilized hourly PM_2.5 data obtained from AirNow (https://www.airnow.gov), which provides real-time PM_2.5 and ozone measurements for numerous cities globally. The dataset spans 2017 to 2024 and captures hourly variations in air quality. Corresponding meteorological data were obtained from the NASA POWER data archive, which provides hourly-resolution climate variables suitable for environmental research. This dataset has been used in previous climate research across the Middle East, including a recent study in Kuwait by AlDousari et al.³⁹, and is considered a reliable source for examining climate research. To ensure consistency with local conditions, the NASA POWER meteorological variables were aggregated to monthly averages and compared with available ground-based station observations in Kuwait during 2017–2023 The comparison revealed strong agreement between the local station observations and the NASA POWER dataset (R² = 0.964 for average temperature (T_avg), 0.9914 for surface pressure, 0.8319 for precipitation, and 0.8699 for wind speed), along with small biases (–2.67 for T_avg, + 1.76 for surface pressure, + 0.3872 for precipitation, and − 0.3749 for wind speed) and relatively low RMSE values (3.419 for T_avg, 1.92 for surface pressure, 2.07 for precipitation, and 1.1893 for wind speed). These results reinforce the suitability of NASA POWER for accurately representing regional meteorological conditions. However, uncertainties may persist during extreme dust events or localized convective rainfall, where satellite retrievals are less reliable.

Trend analysis

Mann-Kendall test (MKT)

The MKT is a nonparametric statistical method used to identify monotonic trends (increasing or decreasing) in time-series data without requiring the data to conform to any particular distribution. It is widely employed in environmental and climatological studies to detect long-term trends in variables such as temperature, precipitation, and air quality indices^41–44. The test evaluates whether a variable tends to increase or decrease consistently over time, without assuming any specific distribution of the data. For a time series x₁, x₂, ..., x_n the Mann-Kendall statistic S is calculated using Eqs. (1) and (2).

1

2

For samples with n > 10, the statistic S is approximately normally distributed with mean E(S) = 0 and variance Var(S) given by:

3

where g is the number of tied groups and is the number of data points in the th tied group. The standardized test statistic (Z) is then calculated as:

4

A positive Z-value indicates an upward trend, while a negative Z-value denotes a downward trend. The statistical significance of the trend is assessed by comparing Z with the critical values of the standard normal distribution at the chosen significance level (± 1.96 for a 95% confidence level).

Sen’s slope and theil slope estimator

The Sen’s Slope Estimator (also known as the Theil–Sen Estimator) is the most commonly used nonparametric method for estimating the magnitude of a linear trend in time-series data^45,46. It is commonly applied in conjunction with the MKT to quantify the rate of change in environmental variables such as PM_2.5 concentrations, temperature, or precipitation^19,47,48. The method computes the median of all pairwise slopes between data points, making it highly resistant to the influence of outliers and missing values. For a time series x₁, x₂, ..., x_n with corresponding time indices t₁, t₂, ..., t_n, the Sen’s slope Q is calculated using Eq. (5).

5

The Sen’s Slope Estimator is then given by the median of all N = n(n − 1)/2 computed slopes:

6

This value Q represents the estimated rate of change per unit time (µg/m³ per year or per hour). A positive value of Q indicates an increasing trend, while a negative value signifies a decreasing trend in the time series. The Theil’s Slope Estimator is conceptually identical to Sen’s method, which is often used interchangeably in literature, particularly when estimating linear trends without making assumptions about the distribution of residuals. The method also allows the construction of a confidence interval for the slope using non-parametric rank statistics and provides a statistical range for the true trend⁴⁹.

Correlation analysis

Cross wavelet transform (XWT) analysis

To investigate the co-variability between PM_2.5 concentrations and climatic variables over time and across different frequencies, we employed XWT. This method allows for the identification of regions in time-frequency space where two time series exhibit high common power, enabling a detailed exploration of their coupled behavior under non-stationary conditions. XWT is particularly well suited to environmental time series⁵⁰, in which both PM_2.5 and meteorological parameters exhibit transient patterns, seasonal cycles, and irregular fluctuations⁵¹.

Mathematically, the Cross Wavelet Transform between two time series and is defined as:

7

where and are the continuous wavelet transforms of and respectively, is the scale (inversely proportional to frequency), is the time translation, and the asterisk denotes the complex conjugate. The modulus ∣∣ reflects the common power between the two-time series at a specific scale and time, while the phase angle of indicates the phase relationship, that is, the relative timing or lag between them.

In XWT plots, warmer colors (red to yellow) indicate regions of high cross power, suggesting periods and frequencies where the two-time series are strongly coupled. Cooler colors (such as blue) correspond to low cross power and weaker associations. Overlaid on these power spectra are arrows that convey phase information. Arrows pointing to the right (→) indicate that the two series are in-phase, meaning they rise and fall together. Arrows pointing to the left (←) indicate an antiphase relationship, in which one series increases while the other decreases. Arrows pointing diagonally, either upward-right (↗) or upward-left (↖), suggest a lead-lag relationship, where one variable lead or lags the other depending on the orientation. The significance of the observed coherence is typically assessed using Monte Carlo simulations, and the cone of influence (COI) is used to demarcate regions where edge effects may distort interpretation, particularly at the beginning and end of the time series^52–54.

Lagged correlation

Lagged Correlation Analysis is a statistical technique used to examine the delayed or time-shifted relationship between two time series, enabling the identification of cause-and-effect dynamics in which the impact of one variable on another may not be immediate⁵⁵. The nature of such a method will be instrumental in environmental studies⁵⁶, such as assessing how meteorological factors influence air pollutant concentrations like PM_2.5 with a time lag. The lagged correlation coefficient at lag is calculated by shifting one time series forward or backward in time and then applying the Pearson correlation coefficient⁵⁷. For two time series X= {x₁, x₂, ., x_n} and Y= {y₁,y₂,.,y_n}, the lagged correlation at lag τ is given by:

8

Here, and are the averages of the original series X and the lagged series Y, respectively, and τ is the lag in time units (hours or days). A positive value of indicates a direct positive relationship at lag τ, meaning that an increase in x is associated with a future increase in y. A negative value suggests an inverse relationship. The magnitude of reflects the strength of the association, and the lag value at which the peak correlation occurs helps identify the time delay with the strongest influence.

Results and discussion

Variation in PM_2.5 levels

The temporal analysis of PM_2.5 concentrations from 2017 to 2024 reveals a coherent pattern of variability across diurnal, monthly, and seasonal timescales, reflecting the combined effects of meteorology, dust dynamics, and human activity (Figs. 2, 3 and 4). The diurnal heat map (Fig. 2) shows that PM_2.5 concentrations peak consistently during the warmer months, particularly June through September, with the highest hourly averages observed in the late evening (7–9 PM). July exhibited the highest hourly concentration (63.3 µg/m³ at 7 PM), followed by August (63.2 µg/m³) and June (62.5 µg/m³). These higher PM_2.5 concentration levels during summer months align with prior studies documenting increased PM_2.5 during the summer season in Kuwait, as well as in Middle East due to dust outbreaks, crustal aerosol loading, and Shamal-driven resuspension^12,14,33. Aldekheel et al.¹² reported that summer mean PM_2.5 level exceeded 75 µg/m³ during 2022–2023, far surpassing winter concentrations. Evening peaks identified in this study may also result from reduced boundary-layer mixing and calmer post-sunset winds, both of which inhibit dispersion in arid climates^11,58. It also indicates substantial health risks, especially for vulnerable people.

Fig. 2.

Hourly average PM_2.5 concentration levels across different months.

Fig. 3.

PDF for hourly PM_2.5 levels in Kuwait.

Fig. 4.

Monthly average PM_2.5 concentration levels during the study years.

Probability Density Functions (PDFs) further reinforce these patterns by illustrating how hourly pollution distributions shift throughout the day (Fig. 3). The broad, right-skewed PDFs observed between 7:00 PM and 9:00 PM indicate both higher average concentrations and greater variability during evening hours, likely reflecting a combination of elevated traffic emissions and reduced atmospheric mixing. In contrast, early-morning periods (1:00–5:00 AM) exhibit narrow PDFs centered on lower values, suggesting more stable and cleaner conditions with minimal human activity and favorable dispersion. Midday and afternoon distributions (12:00–6:00 PM) show moderate yet highly variable concentrations, consistent with interactions between solar heating, increased atmospheric turbulence, and daytime activity levels.

Monthly and annual heat map patterns (Fig. 4) complement these findings by revealing broader seasonal and interannual trends. Across the eight-year record, May, July, and August consistently showed the highest monthly concentrations, often exceeding 48 µg/m³, with notable spikes in August 2018 (55.2 µg/m³) and in May and July 2022 (48.0 µg/m³ and 55.8 µg/m³, respectively). These recurring summer maxima mirror the diurnal patterns and reflect intensified dust events, stagnant air masses, and increased anthropogenic activity during warm months. Conversely, the winter months (December–January) consistently recorded the lowest concentrations, frequently ranging from 25 to 40 µg/m³, attributable to stronger winds, cooler temperatures, and deeper boundary layers that enhance pollutant dispersion¹³. Interannual fluctuations are also evident, with several months in 2024 exhibiting lower concentrations than in preceding years, suggesting year-to-year variability driven by meteorological conditions or emission patterns.

Trends of PM_2.5 levels

To identify the trends in PM_2.5 levels in Kuwait City during the study period, we have used MKT, Sen’s Slope Estimator and Theil’s slope. MKT and SS were used to identify the monthly trends. Theil’s slope was used for hourly trend and monthly trend analysis, and outcomes were provided in the supplementary materials.

The outcomes of MKT and SS in Table 1 shows that most months exhibit negative slopes, suggesting a gradual decline in PM_2.5 concentration levels over the study period. However, these decreases are generally not statistically significant at the 95% confidence level. September stands out as the only month with a significant downward trend, reflecting a consistent and measurable reduction in PM_2.5 concentrations during the study period. This is aligned with the steep negative Theil Slope observed for September (Fig. 1S in the supplementary). In arid regions such as Kuwait, September marks the end of the intense summer Shamal wind season, when dust-storm frequency and wind-driven dust emissions sharply decline, resulting in lower PM_2.5 levels than during June–August^14,59. At the same time, early-autumn cooling increases the boundary-layer height and improves atmospheric ventilation, thereby reducing particulate accumulation relative to the stagnant, dust-laden summer months^59,60. The SS values confirm the declining trend observed in the Theil Slope analysis (Fig. 2S in the supplementary material), with monthly declines in PM_2.5 concentration ranging from moderate to steep throughout the year. In February, the SS was calculated − 2.334, and in September, it’s − 3.555, indicating significant monthly declines. Although SS values for other months also show negative tendencies, the lack of statistical significance in the MKT results indicates that these declines remain within the range of natural interannual variability. The supplementary Theil–Sen analysis confirms this overall pattern, with negative but mostly non-significant slopes across the majority of months.

Table 1.

Outcomes of MKT for monthly PM_2.5 trends.

Month	Z value	Sen slope	Significance level
Jan	− 0.12	− 0.431	Non-significant
Feb	− 1.86	− 2.334	Non-significant
Mar	− 1.11	− 2.272	Non-significant
Apr	− 1.11	− 1.702	Non-significant
May	− 1.36	− 2.570	Non-significant
Jun	− 0.62	− 1.491	Non-significant
Jul	− 1.11	− 3.055	Non-significant
Aug	− 1.61	− 2.371	Non-significant
Sep	− 2.1	− 3.555	Negative trend at 95%
Oct	− 1.11	− 4.070	Non-significant
Nov	0.00	0.134	Non-significant
Dec	0.12	0.413	Non-significant
Annual	− 1.61	− 2.139	Non-significant

Influence of weather variables on PM_2.5 levels

Figure 5 shows the XWT between PM_2.5 and weather variables, representing the influence of these weather variables on PM_2.5 distribution in Kuwait City. Figure 5a shows significant coherence at scales from 100 to 300 days between temperature and PM_2.5 levels. The strong in-phase coherence at 100–300-day scales reflect shared seasonal cycles rather than short-lived pollution events. These long periods correspond to Kuwait’s annual meteorological regime, such as the summer dust season and winter stagnation⁶¹, which cause PM_2.5 and temperature/humidity to vary in concert. The arrows predominantly point right, suggesting a direct, in-phase relationship. This could reflect enhanced photochemical reactions under warmer conditions that produce more secondary aerosols. The periods with in-phase relationships could suggest that higher temperatures correlate with increased PM_2.5 levels, possibly due to increased emissions from sources such as vehicles and industry, or to reduced pollutant dispersion under hotter conditions. Conversely, anti-phase segments could indicate situations where rises in temperature correlate with decreases in PM_2.5, perhaps due to meteorological conditions like wind patterns that disperse air pollutants more effectively during warmer weather. The XWT between PM_2.5 and dew point temperature in Fig. 5b shows strong coherence at similar scales (100 to 300 days), with arrows pointing right, indicating that higher dew points are correlated with increased PM_2.5 levels, likely due to moisture facilitating the formation and growth of particulate matter. Throughout the observed period, coherence is particularly pronounced at lower frequencies (higher scale values), indicating a significant long-term relationship. This could indicate that seasonal or annual climatic factors influencing dew point are consistently related to variations in PM_2.5 levels. Arrows within the significant areas mainly pointing right (slightly up-right) indicate that the PM_2.5 and dew point temperatures are typically in-phase (rising and falling together). This phenomenon is especially marked in the larger scale zones around the 64–256 band, reflecting a strong relationship over multiple months. This in-phase relationship suggests that periods with higher dew points, indicative of higher air moisture, often coincide with increased PM_2.5 concentrations, which could be due to air moisture enhancing particles’ ability to remain airborne or creating conditions conducive to pollutant retention in the atmosphere. The absence of a significant antiphase relationship (where arrows point left or downward) confirms that dew point and PM_2.5 move together rather than inversely. Another temperature variable, WBT in Fig. 5c revealed coherence at multiple scales, particularly in the 150 to 250-day range. The arrows, mostly pointing right, suggest that higher wet bulb temperatures, which combine heat and humidity, exacerbate PM_2.5 concentrations. During 2017–2023, significant coherence was visible particularly in lower frequency bands, indicating a consistent and long-term relationship between PM_2.5 levels and WBT. This suggests that seasonal or annual patterns in WBT are strongly correlated with changes in PM_2.5 concentration. The arrows within these significant areas predominantly point slightly upward, indicating that PM_2.5 levels and WBT are generally in phase; that is, as WBT increases, PM_2.5 concentrations tend to increase, and vice versa. Such in-phase relationships are particularly notable at larger scales, suggesting that longer-term climatic patterns affecting WBT are mirrored in PM_2.5 variations. Conversely, the absence of significant antiphase patterns (where the variables move inversely to each other) across most scales further confirms the consistent alignment of these environmental factors.

Fig. 5.

XWT between PM_2.5 and weather variables.

XWT between humidity variables in Fig. 5d and e shows significant coherence at longer scales (around 200 to 350 days), with arrows generally pointing right. The results show that at a 4–16-day timescale, the arrows occasionally point downward or to the left, suggesting anti-phase relationships in which an increase in relative humidity may correspond to a decrease in PM_2.5 levels, or vice versa. Similar anti-phase relationships were observed for specific humidity levels (Fig. 5e). This indicates that periods with higher humidity levels are associated with higher PM_2.5 concentrations, potentially due to the hygroscopic properties of particulate matter. Conversely, windspeed variables showed an opposite coherence. At smaller scales (4 to 16 days), the relationship between PM_2.5 and WS10M (Fig. 5g) exhibits greater variability, with both in-phase (arrows pointing right) and anti-phase (arrows pointing left) relationships evident. Figure 5f shows several periods during which the WS10M appears to significantly influence PM_2.5 levels, particularly at larger scales (64 to 256 days), where coherent patterns are evident. During these periods, the arrows mostly point to the right or slightly downward-right, indicating that the WS10M and PM_2.5 are generally in phase or slightly leading/lagging each other. This suggests that an increase in WS10M is closely followed by an increase in PM_2.5 levels, or vice versa, possibly due to the winds carrying particulate matter from other regions or contributing to the suspension of local particulates. The influence of wind speed on PM_2.5 in Kuwait can operate in both directions, as demonstrated in regional dust–meteorology studies. Strong northwesterly Shamal winds are known to elevate particulate levels in the Middle East through dust transport and surface resuspension^60,62,63, while other wind regimes enhance dispersion and reduce pollutant concentrations. The in-phase coherence observed at longer scales, therefore, likely reflects seasonal dust-bearing wind patterns rather than a uniformly positive relationship between wind speed and PM_2.5. Interestingly, for WS50M (Fig. 5h), at larger scales, particularly from 64 days to 256 days, the coherence between WS50M and PM_2.5 is quite pronounced, showing that seasonal and sub-seasonal variations in windspeed significantly correlate with PM_2.5 levels. The arrows pointing right or slightly upwards within these zones suggest that WS50M and PM_2.5 are largely in-phase, meaning that increases in windspeed at this elevation are contemporaneous with increases in PM_2.5 concentrations. This could be attributed to the wind’s ability to transport particulate matter from distant sources or resuspending local particulates. Similar to WS10M, at smaller scales (4 to 16 days), WS50M exhibited both in-phase (arrows pointing right) and anti-phase (arrows pointing left) interactions. Antiphase relations at these scales could indicate situations in which increases in wind speed may lead to decreases in local PM_2.5 levels due to dispersion effects.

The outcomes of XWT analysis between PM_2.5 and rainfall in Fig. 5f show several large-scale coherent patches (64–256 days), with arrows pointing right, indicating in-phase relationships where periods of higher rainfall tend to coincide with higher PM_2.5 or vice versa. This may appear counterintuitive, as rainfall generally removes particles through wet scavenging; however, studies in arid and semi-arid regions have reported that elevated humidity prior to rainfall can enhance hygroscopic particle growth and increase PM_2.5 concentrations under stagnant conditions^64,65. Because this mechanism has not been directly demonstrated for Kuwait, this interpretation should be considered cautiously and framed as a potential explanation rather than a definitive causal pathway. The observed long-scale in-phase coherence may therefore reflect the seasonal co-occurrence of humid periods and elevated PM_2.5, a pattern consistent with broader Middle Eastern aerosol–meteorology dynamics⁶⁶. In contrast, at shorter scales (4–16 days), several anti-phase patches appear, where increases in rainfall correspond to decreases in PM_2.5, consistent with the expected short-term cleansing effect of precipitation. The XWT outcomes for TSI in Fig. 5i show various coherence patterns which suggest that solar irradiance exerted a strong influence on PM_2.5 levels, with varying effects depending on the scale of temporal aggregation. In the XWT graph, significant areas of coherence occur primarily at lower frequencies, particularly in the scales longer than 64 days. These areas indicate persistent interactions between TSI and PM_2.5 on a seasonal or sub-seasonal basis. The arrows within these high-coherence zones mostly point to the right or slightly downward, suggesting an in-phase or slightly leading relationship where changes in TSI either coincide with or precede changes in PM_2.5 levels. This could be attributed to the role of solar irradiance in photochemical reactions that produce or modify particulate matter. Overall XWT analysis suggests that at smaller scales, the coherence between PM_2.5 and weather variables are less consistent, while at higher scales, the coherence was consistent. This implies that at shorter scales, association depends on the combination of all-weather variables, which further suggests the importance of multivariate regression analysis.

The regression analysis in Table 2 described the relationship between PM_2.5 concentrations and various weather variables using hourly data from January 1, 2017, to December 31, 2024. The intercept, represented by ‘const’, had a coefficient of 17.4397 but was not statistically significant (p = 0.348), suggesting that the baseline level of PM_2.5 when all predictors were zero was not reliably different from zero given the data. Temperature variables like Tavg and DEW showed positive associations with PM_2.5, with coefficients of 7.8335 and 7.1454, respectively, though these relationships were not statistically significant (p = 0.365 and p = 0.409). WBT was negatively associated with PM_2.5 (− 14.5564), indicating that higher wet bulb temperatures may reduce PM_2.5 levels; however, this association was not statistically significant (p = 0.400). Statistically significant predictors included SH, RH, Rainfall, WS10M, WS50M, and TSI. Result suggests that specific and relative humidity levels increased PM_2.5 with coefficients of 0.3390 (p < 0.001) and 0.0863 (p < 0.001), suggesting humidity factors contributed to higher PM_2.5 concentrations. Rainfall and WS50M showed negative effects on PM_2.5 (− 0.2389 and − 2.0540, respectively; both p < 0.001), implying that rainfall and higher winds at 50 m helped disperse or reduce PM_2.5 concentrations. WS10M and TSI showed contrasting effects. Wind speed at 10 m tended to increase PM_2.5 (1.3265, p < 0.001), while higher TSI slightly influenced decreased PM_2.5 concentration levels (− 0.0040, p < 0.001). These results highlighted the complex interplay of meteorological variables in influencing air quality, specifically PM_2.5 levels. In this regard, we have used Kernal Density Dot plots (Fig. 6) to investigate the range of all the climatic variables where the correlation with PM_2.5 levels were highest.

Table 2.

Multivariate regression analysis between PM_2.5 and weather variables.

Variable	Coefficient	Std. error	t-statistics	P	25% and 95% Confidence Interval
Const	17.4397	18.59	0.938	0.348	(− 18.996, 53.876)
Tavg	7.8335	8.65	0.906	0.365	(− 9.120, 24.787)
DEW	7.1454	8.65	0.826	0.409	(− 9.809, 24.100)
WBT	− 14.556	17.30	− 0.841	0.400	(− 48.464, 19.351)
SH	0.3390	0.082	4.125	0.000	(0.178, 0.500)
RH	0.0863	0.013	6.610	0.000	(0.061, 0.112)
Rain	− 0.2389	0.034	− 6.928	0.000	(− 0.307, − 0.171)
WS10M	1.3265	0.12	11.056	0.000	(1.091, 1.562)
PS	0.0783	0.178	0.439	0.661	(− 0.272, 0.428)
WS50M	− 2.054	0.091	− 22.523	0.000	(− 2.233, − 1.875)
TSI	− 0.004	0.000	− 14.641	0.000	(− 0.005, − 0.003)

Fig. 6.

Density dot plots visualizing the relationship between PM_2.5 levels and weather variables.

The kernel density plots in Fig. 6 illustrate the influence of various weather variables on PM_2.5 levels, with distinct density distributions for different factors. PM_2.5 revealed a prominent density around the temperature range of 10–30 °C, where PM_2.5 levels tended to peak, suggesting that moderate temperatures might facilitate the accumulation of particulate matter. Similarly, the relationship between PM_2.5 and DEW exhibited maximum density at dew points between 5 and 15 °C, indicating that moisture levels within this range may be conducive to higher PM_2.5 concentrations. Specific and relative humidity showed strong associations with PM_2.5 at higher humidity values, notably around 0.6–0.8 gm/m³ for specific and 40–60% for relative humidity, which might reflect the role of humidity in PM_2.5 formation and stability. In terms of windspeed variables (WS10M and WS50M), both exhibited higher densities of PM_2.5 at lower wind speeds, particularly below 10 m/s, emphasizing that lower wind conditions likely reduce the dispersion of particulate pollutants, leading to increased concentrations. PM_2.5 against rainfall showed increased PM_2.5 levels at lower rainfall levels (0–2.5 mm), suggesting that dry conditions favor the persistence and accumulation of airborne particles. The impact of surface pressure was less pronounced, with a slight increase in PM_2.5 observed around pressures of 1010 hPa. Solar irradiance presented a clear impact, with the highest PM_2.5 densities appearing under moderate hourly TSI levels (400–800 W/m²), highlighting the complex interactions between sunlight, atmospheric chemistry, and particulate matter levels.

Figure 7 presents lagged correlation results between hourly PM_2.5 concentrations and various weather variables for the period 2017–2024 in Kuwait, demonstrating how these environmental factors influence PM_2.5 variability. Temperature showed its strongest influence at a 19-hour lag, suggesting a delayed response of PM_2.5 formation and dispersion to thermal variability. DEW and WBT also exhibited notable delayed effects, with peak correlations at 11 h and 18 h, respectively, reflecting the time required for moisture-driven processes such as hygroscopic growth, aqueous-phase reactions, and changes in boundary-layer stability to modify particulate concentrations. Relative humidity exhibited the longest lag, with a peak influence at 30 h. This extended delay indicates a multistage process in which humid conditions initially promote water uptake and particle swelling, followed by enhanced aerosol retention and reduced dry deposition over the subsequent day⁶⁷. In Kuwait’s desert climate, relative humidity typically rises overnight and drops sharply during daytime. Here a 30-hour lag therefore implies that humid nighttime conditions can elevate PM_2.5 concentrations not only the following day but potentially the next night. In contrast, daytime RH changes exert a weaker influence because particles have insufficient time to undergo substantial hygroscopic growth before being dispersed. These lag times likely reflect the delayed thermodynamic processes that govern secondary aerosol formation and pollutant accumulation. Warmer and more humid conditions can enhance gas-to-particle conversion, hygroscopic particle growth, and boundary-layer suppression over several hours, allowing PM_2.5 concentrations to rise well after the initial temperature or moisture change.

Fig. 7.

Lagged correlations between PM_2.5 and weather variables.

Rainfall showed a 10-hour lag, consistent with delayed washout effects or continuing emissions and resuspension during and after rainfall events. WS10M and WS50M exhibited correlations peaking at 18 h and 20 h, respectively, indicating the time needed for synoptic wind patterns to redistribute or ventilate particulate matter. TSI exhibited its maximum effect at 25 h, which reflects the lag between photochemical activity and the formation of secondary aerosol species. These findings highlight the nonlinear and time dependent nature of meteorology-PM_2.5 interactions in arid environments and demonstrate the importance of incorporating lagged atmospheric processes into predictive air-quality models.

Conclusion

This study presents a comprehensive assessment of the temporal behavior and meteorological influences on PM_2.5 concentrations in Kuwait City, an arid urban environment with complex air quality dynamics. Using hourly data from 2017 to 2024, we employed wavelet coherence, lagged correlation, and trend detection methods to uncover the underlying temporal patterns and climatological interactions of PM_2.5 levels. Our findings revealed clear diurnal variability, with PM_2.5 concentrations peaking during summer evenings, most notably at 7 PM in July (63.3 µg/m³) and reaching their lowest levels in the early hours of winter months, such as 1 AM in December (34.5 µg/m³). Seasonal trends showed elevated PM_2.5 during the hot and dry months (May–August), driven by resuspension of desert dust and stagnation of air masses, whereas the winter months generally exhibited improved air quality. The Theil–Sen slope and the Mann-Kendall test indicated an overall decline in PM_2.5 levels, with September showing a statistically significant reduction at the 95% confidence level. Wavelet coherence analysis identified strong in-phase relationships between PM_2.5 and several meteorological variables, particularly temperature, dew point, and wet bulb temperature at scales of 100 to 300 days, underscoring the seasonal co-variability between climate and aerosol behavior. Crucially, this study emphasizes the lagged effects of meteorological variables on PM_2.5 concentrations, revealing complex and delayed interactions. Temperature, dew point, and wet bulb temperature showed the strongest correlations with PM2.5 at 19, 11, and 18 h, respectively, suggesting that thermodynamic conditions influence secondary aerosol formation and pollutant accumulation over time. Similarly, wind speeds at 10 m and 50 m demonstrated delayed but significant associations (lags of 18–20 h), reflecting both dispersion and resuspension mechanisms. Relative humidity exerted its peak influence at a 30-hour lag, underscoring the role of moisture in enhancing particle hygroscopic growth and atmospheric retention. The influence of solar irradiance was also significant, with a 25-hour lag, suggesting that it affects photochemical processes that drive aerosol transformation. Regression analysis confirmed significant associations with specific humidity, rainfall, wind speed, and total solar irradiance, reinforcing the multi-factorial control of PM_2.5 in arid environments. These findings underline the non-instantaneous, cumulative nature of meteorology-PM_2.5 interactions, which are critical for improving air pollution forecasting systems.

However, this study has certain limitations. First, the analysis is limited to surface-level meteorological variables and excludes vertical atmospheric structures, such as boundary-layer height, thermal stability, and inversion strength. These factors are especially important in arid cities such as Kuwait, where rapid changes in mixing depth can govern the buildup, dispersion, and vertical redistribution of PM_2.5. Second, satellite-derived aerosol optical depth (AOD) was not used, even though it can identify regional dust transport, elevated aerosol layers, and large-scale plume dynamics that surface monitors alone cannot detect features that are particularly relevant during intense dust storms common in the region. Including AOD with boundary-layer information could significantly improve PM_2.5 forecasting by providing a more comprehensive view of aerosol loading and atmospheric mixing. Third, this study does not differentiate between natural and human-made PM_2.5 sources, which limit the ability to attribute observed variations to specific emission categories. Future work could incorporate chemical speciation, emission inventories, and dust-source mapping to separate contributions from traffic, industry, and desert dust. A future spatiotemporal machine-learning framework could integrate ground observations with AOD, lagged meteorological data, and source-related chemical information to better capture nonlinear, time-dependent pollution dynamics and substantially improve early-warning systems for extreme pollution events in hyper-arid environments.

Despite these limitations, the findings of this study have several implications for air-quality management, urban planning, and climate resilience in arid cities such as Kuwait. First, the pronounced seasonal peaks in PM_2.5, especially during summer dust events, highlight the importance of understanding weather patterns. The identified lagged responses to temperature, humidity, and wind conditions emphasize the need for meteorology-informed air-quality forecasting systems. Integrating such forecasts into municipal decision-making can support targeted interventions during high-risk periods, including issuing public health advisories, adjusting traffic flows, and modifying outdoor work schedules to reduce population exposure. Second, the results support Sustainable Development Goal (SDG) 11 (Sustainable Cities and Communities) by demonstrating how hourly-scale PM_2.5 monitoring can guide urban design strategies, such as optimizing ventilation corridors, regulating construction activity during peak dust seasons, and incorporating green or dust-mitigating infrastructure in vulnerable areas. Planning departments can also use these forecasts to better time street cleaning, dust suppression, and emissions controls. Third, this study’s insights advance SDG 13 (Climate Action) by showing that PM_2.5 dynamics are closely linked to climatic variables in hyper-arid settings. Including boundary-layer behavior, dust climatology, and meteorological lag effects into climate adaptation plans can improve preparedness for extreme pollution episodes intensified by climate change. Air-quality forecasts can be integrated into multi-hazard early-warning systems that also monitor heatwaves, dust storms, and stagnant air masses, enabling coordinated public health responses. Finally, the findings contribute to SDG 3 (Good Health and Well-being) by highlighting the importance of exposure-reduction strategies targeting vulnerable groups, such as children, outdoor workers, and individuals with chronic respiratory or cardiovascular conditions. Health agencies can use time-resolved PM_2.5 forecasts to plan clinical outreach, allocate emergency resources, and enhance risk communication protocols.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Abdulla Al Kafy: Conceptualization, Project administration, Data collection, Formal analysis, Data curation, Supervision, Resources, Software, Methodology, Investigation, Writing—original draft, Validation, Writing—review and editing. Wafeek Mohamed Ibrahim: Formal analysis, Data curation, Supervision, Resources, Software, Methodology, Investigation, Funding, Writing—original draft, Validation, Writing—review and editing. Abdullah Al Baky: Data collection, Formal analysis, Data curation, Supervision, Resources, Software, Methodology, Investigation, Writing—original draft, Validation, Writing—review and editing. Tekalign Ketema Bahiru: Formal analysis, Data curation, Supervision, Resources, Software, Methodology, Investigation, Writing—original draft, Validation, Writing—review and editing.

Funding

The authors extend their appreciation to the Deanship of Research and Graduate Studies at King Khalid University for funding this work through small group research under grant number RGP.2/512/46.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.