Introduction

The latest Intergovernmental Panel on Climate Change (IPCC) report, concludes that the global mean temperature increase by the end of the century is poised to be up to 5.4°C [1], whereas preliminary results from the Coupled Model Intercomparison Project phase 6 (CMIP6) projections predict an even higher increase for the business-as-usual scenario [2]. Precipitation is expected to increase at high latitudes, with the opposite effect projected in large parts of the subtropics. Moreover, extreme weather events, such as consecutive days with high temperatures, droughts and extreme precipitation events, are expected to become more common in certain parts of the world [1]. Such events have a detrimental effect on human health. For instance, the World Health Organisation (WHO) estimates that on a global level, excess deaths attributable to direct and indirect effects of climate change (including heat stress, malnutrition, diarrhea, and vector-borne diseases) will exceed 250,000 per year in the period 2030–2050 [3].

The Eastern Mediterranean and Middle East (EMME) is one of the regions most affected by climate change. It comprises Bahrain, Cyprus, Egypt, Greece, Iran, Iraq, Israel, Jordan, Kuwait, Lebanon, Oman, Palestine, Qatar, Saudi Arabia, Syria, Türkiye, the United Arab Emirates, and Yemen [4]. The EMME is considered a climate change hotspot, with warming rates significantly exceeding the global average. As a result, heat extremes and drought events are expected to become increasingly frequent [5–7].

Human health is dependent on a delicate interplay of lifestyle, environmental conditions, access to clean water and an array of other factors. Therefore, the profound changes in local climate which are expected to affect the EMME, are bound to have significant impacts on the health status of region’s population. Factors such as heat stress, water scarcity, and air pollution are projected to impose significant strain on the health systems of EMME nations [4].

High ambient temperature can disrupt the body’s physiological ability to maintain homeostasis by dissipating excess heat, leading to both acute and chronic health effects [8,9]. Human thermoregulation relies primarily on evaporative cooling, which becomes less effective as heat and humidity increase. Under hot and humid conditions, the body progressively loses its ability to dissipate excess heat. Exposure can become lethal when the wet-bulb temperature, which reflects the combined effect of air temperature and humidity rather than air temperature alone, exceeds approximately 35°C. However, adverse and potentially severe health effects may also occur at lower wet-bulb values under prolonged exposure or among physiologically vulnerable populations [10,11]. In this context, heat extremes pose significantly higher risks for vulnerable groups, including elderly and people with preexisting health conditions or disabilities [12,13].

Exposure to low temperatures can also adversely affect human health through direct physiological responses it generates ([14]) and altered biological processes that promote the transmission of infectious diseases [15–18]. In addition, temperature-related health risk is believed to be often compounded by moisture, both in hot and cold conditions. Quantifying the synergistic effects of ambient temperature and moisture on human health is essential for understanding the broader impacts of climate change. These effects are expected to be highly localised, due to the significant spatial variability of both temperature and especially moisture or humidity.

In this direction, several epidemiological studies use humidity metrics, such as relative humidity and wet bulb temperature, to explore the interaction with temperature. However, the results often remain inconclusive [19]. The strong inverse relationship between temperature and relative humidity (RH) can obscure the impact of temperature on health outcomes, leading to ambiguous results in the assessment of heat-health exposure. To overcome these limitations, recent studies recommend using simple mass-based variables, such as water vapour pressure (WVP), instead of RH in exposure–response analyses. Mass-based measures more directly influence sweat evaporation, whereas RH primarily reflects temperature-driven diurnal cycles that do not accurately capture variations in heat strain [19]. As a measure of absolute atmospheric moisture, WVP enables a more direct assessment of the combined effects of temperature and humidity on human health. It quantifies the absolute water content in the air and is less affected by temperature fluctuations [20]. The approach aligns with recent regional findings highlighting that mortality risk increases under compound environmental stressors, such as thermal stress and air pollution, in the Mediterranean cities like Thessaloniki [21].

In this study, we focus on Cyprus, an island located in the Eastern Mediterranean region, and explore the synergistic effects of temperature and atmospheric moisture on human mortality, as expressed by RH and WVP. The climate in Cyprus displays patterns typical of the Mediterranean region, with hot-dry summers, and winters with variable rainfall patterns [22]. A few studies have investigated the effects of temperature on human mortality in Cyprus [23–27] and generally found that the aggregate temporal effect of very high temperatures (e.g., a heat wave lasting for a few days) increases the risk of mortality. These studies also point to increased health risks associated with extremely low temperatures experienced 2–3 weeks in the past. Other reports have considered the additive effect of air pollution on mortality risk from temperature, but found little evidence of an impact for the island [28].

Here, we extend previous analyses by modelling 16 years of mortality data from Cyprus and quantify the (non-additive and non-linear) compound effects of temperature and humidity-related variables on mortality risk, something that has not yet been studied for this region.

Data

Meteorological variables

Meteorological data were obtained from a novel human thermal bioclimate dataset, which was developed based on the Copernicus European Regional Reanalysis (CERRA) [30]. The dataset is publicly available (https://zenodo.org/records/10893914; accessed on 12 May 2025) and has been described in detail by Giannaros et al. [31]. Briefly, hourly air temperature and relative humidity data were obtained from CERRA at 5.5×5.5km spatial resolution [30]. The hourly WPV values were calculated using the following formula (1) from [32]:

(1)

The meteorological data were spatially aggregated for each of the five administrative districts of the Republic of Cyprus. To better represent population-level exposure, the aggregation was a population-weighted average, thereby aligning the thermal metrics with the actual distribution of residents in each district [31]. Subsequently, hourly estimates of population-weighted temperature, relative humidity (RH) and water vapour pressure (WVP) were temporally aggregated to daily values to align them with the resolution of mortality data. Specifically, for each district, the daily maximum temperature (Tmax) and RH were computed, while for WVP, the value corresponding to the time of Tmax was selected to best represent concurrent exposure.

Fig 1 illustrates the monthly climatology of Tmax, RH and WVP across all districts for 2004–2019. A strong seasonal pattern is evident, with summer months (June–August) characterised by elevated Tmax values and suppressed RH, reflecting their inverse relationship. Maximum temperatures consistently exceed 25°C from April through October, peaking in July and August. In contrast, cooler conditions prevail from November to March, delineating a well-defined thermal contrast. This consistent annual pattern provides an objective criterion for defining April–October as the warm season and November–March as the cold season – a classification aligned with known seasonal variations in temperature-related health risks. This seasonal division is important for capturing the potentially quite different response of health outcomes to temperature-stress in each season. The uniformity of this pattern across diverse geographic regions further supports the validity of this seasonal segmentation for subsequent analyses.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Monthly averages of maximum temperature (top), mean relative humidity (middle) and WVP (bottom), respectively, for each district in Cyprus, for 2004–2019.

The shaded area around the lines represents ±1 standard errors.

https://doi.org/10.1371/journal.pclm.0000821.g001

Regarding atmospheric moisture, coastal districts (e.g., Famagusta, Larnaca, Limassol, and Paphos) consistently exhibit higher WVP values than the inland district of Nicosia, which displays both lower absolute values and reduced seasonal variability. This spatial heterogeneity highlights divergent environmental exposure profiles and suggests potentially distinct thermo-physiological and health impacts between coastal and inland populations. To capture this heterogeneity, we stratified the analysis into two geographic categories. Nicosia, which comprises roughly one-third of the population, was classified as ‘non-coastal’, while the remaining four districts were grouped as ‘coastal’. This classification reflects both climatic variation in thermal exposure and underlying demographic distribution patterns across regions.

Distributed lag models

For daily mortality counts M_t (for day t) and daily temperature T_t, a DLM can generically be defined as

(2)(3)

where L = 9 days for warm and L = 20 days for cold season, is the maximum lag we consider in this work, a value consistent with other studies looking into the seasonal effects of temperature on mortality [20,27,37,38]. The Negative Binomial distribution is a conventional choice in epidemiological analyses, and extends the Poisson distribution to allow for over-dispersion (extra variance). The natural logarithm of the population, P_t, is used as an offset to allow for background population variability (for each district and calendar year) so that the effects act on the mean mortality incidence rate per unit population (rather than the mean count). The framework can be extended to allow for non-linear effects at each lag, via:

(4)

where h(l, T_t−l) is a 2D function of temperature and lag l. If we assume that h(l, T_t−l) is smooth then this can be interpreted as a GAM, which as implemented in the R package mgcv [34,39] includes penalisation for optimal estimation of h(l, T_t−l).

Penalisation is a way to avoid over-fitting (over-explaining) the data, which also reduces sensitivity to the choice of L, as long as this is large enough. Furthermore, interactions of the temperature-lag effect can be straightforwardly defined via

(5)

Function is defined using tensor product interactions of regression splines [34], a flexible way of constructing smooth multi-dimensional functions. All models presented in the subsequent section are implemented in the R package mgcv, which uses penalised maximum likelihood to do the estimation of the smooth functions.

The functions in Equations (4)–(5) are constrained to be centered at zero (=0), so that the term is the overall average mortality rate. As such, function is interpreted as the multiplicative increase or decrease of the mean mortality rate. So, RR(l, T_t−l) > 1 implies mortality risk is greater than average, whereas RR(l, T_t−l)<1 means lower risk. In that sense, RR(l, T_t−l) is the relative risk or RR (relative to the overall mean death rate) of temperature T at lag l. Given the potential influence of coastal proximity on microclimatic variations and its subsequent effects on health outcomes, we quantify the RR estimates for temperature (Tmax), stratified by coastal and non-coastal regions, to better capture spatial heterogeneity in heat-related mortality. This stratified approach enables a more comprehensive assessment of the synergistic effects between temperature and different moisture metrics (RH, WVP) across distinct geographic regions.

A summary risk measure that integrates out the lag dimension is the cumulative risk (CR), defined as:

(6)

which can be interpreted as the risk of a particular temperature value T_t being held constant for L + 1 days in a row. Although probably implausible, it gives an idea for which temperatures values tend to induce most of the risk.

The mgcv package facilitates the implementation of GAMs, enabling Monte Carlo simulations to quantify the uncertainty of smooth functions and derived quantities. This allows for the assessment of the statistical significance of RR and CR by determining whether the value 1 is contained within their respective 95% credible intervals [36]. If 1 is included, it suggests that the risk is not statistically significant (relative to the overall mean mortality rate); otherwise, if 1 falls outside the interval, the effect is considered significant.

Typically, risk visualization is performed by computing RR and CR over a range of lag and exposure values. However, this method generates counterfactual estimates, as it does not incorporate exposure variability. To overcome this issue, alternative measures such as the Attributable Fraction (AF) and Attributable Number (AN) have been introduced [40], providing a way to interpret risk estimates based on the observed time series. Here, we use the forward AF, which considers current exposure in relation to future risk (e.g., how much of today’s temperature is affecting tomorrow’s risk). This method is preferred for estimating health burdens linked to specific exposure events [40]. The forward AF is given by:

(7)

i.e., one minus the risk ratio, representing the proportion of risk associated with the optimal temperature relative to the risk at the current exposure level T_t. The optimal temperature (OT) corresponds to the temperature associated with the lowest cumulative risk (CR), ensuring that AF(T_t) ≥ 0. As a result, AF(T_t) quantifies the fraction of mortality cases that attributable to deviations from the optimal temperature on a given day t with temperature T_t. Multiplying this value by the average number of cases, , over the period [t,t + L] gives the attributable number of cases for day t: . Finally, summing over all days t and normalizing by the total case count provides the overall AF for the entire study period: . For higher order interactions, the OT is a vector of values, e.g., the optimalTmax and WVP.

Results

First, models with a single exposure were fitted and presented (i.e., equation (4)), before fitting models that include the synergistic effect of temperature and moisture metrics (i.e., equation (5)). To more accurately capture the seasonal risk profile, we have divided the time series into warm (April–October) and cold (November–March) periods. The classification of seven months as warm and five as cold period is grounded in a clear climatological pattern evident across all five districts in Cyprus, as shown in Fig 1. We use Tmax as the quantity that represents the temperature exposure, a common choice in health-heat related studies [20,27,41].

The individual effect from temperature and moisture

Fig 2 and Fig 3 show estimates of the mortality relative risk (RR) as a function of maximum temperature (Tmax), relative humidity (RH) and water vapour pressure (WVP) and the temporal lag for the warm and cold periods. The results are stratified by coastal and non-coastal cities to account for potential geographic and climatic differences.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Estimates of mortality relative risk as a function of Tmax (top), RH (middle) and WVP (bottom) for the warm period. Statistical significance is indicated by grey points.

https://doi.org/10.1371/journal.pclm.0000821.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Estimates of mortality relative risk as a function of Tmax (top), RH (middle) and WVP (bottom) for the cold period. Statistical significance is indicated by grey points.

https://doi.org/10.1371/journal.pclm.0000821.g003

Maximum temperature.

The relative risk (RR) surface of Tmax and during the warm season in Fig 2, exhibits a pattern consistent with previous epidemiological studies [23,27,42]. In both coastal and non-coastal areas, elevated maximum temperatures are associated with a relatively immediate increasing effect on the mortality risk, peaking at short lag periods (0–5 days lag). This is followed by a decline of the risk, indicative of mortality displacement (the “harvesting” effect). In non-coastal areas, low extreme temperature values were also linked to increased risk, though with delayed onset, as reflected by elevated relative risk at longer lags (5–10 days).

For the cold period (Fig 3), a positive effect on mortality appears for the lower extreme temperature range, between , and a reduction of risk at high temperatures is also evident. Previous research has also identified peaks in mortality during periods of cold weather, indicating that this effect is at least in part attributable to an increased incidence of transmissible diseases of the respiratory tract [14,42,43].

Relative humidity.

We also examined the association between relative humidity (RH) and mortality (Fig 2, middle row), as RH has previously been proposed as a key contributor to physical heat stress. Interestingly, our findings indicate that during the warm period, the most pronounced increases in relative risk (RR) occur at low RH levels, with the effect being relatively immediate (up to 7 days of lag) in both coastal and non-coastal cities. This elevated risk under low humidity conditions may be partially reflected in the inverse relationship between maximum temperature and RH (Supplementary S1 Fig). High temperatures in Cyprus frequently coincide with low RH levels, creating conditions commonly described as “dry heat,” which can intensify thermal stress on the human body. This is particularly relevant given the island’s climatic context: Cyprus is predominantly characterised by a temperate climate with hot, dry summers, while certain inland regions are classified as hot and arid according to the Köppen–Geiger climate classification system [44]. These dry and heat-prone conditions not only elevate the risk of heat-related health impacts but also underscore the region’s vulnerability to intensifying climate extremes [22,45].

In contrast, statistically significant negative associations are observed at high RH levels (i.e., “blue” regions in the figure), particularly above 75% in coastal areas and between 50–75% in non-coastal areas. These are likely associated with lower ambient temperatures during the warm season, and thus a substantially reduced risk of heat-related mortality.

During the cold season (Fig 3), there little fluctuation in risk from RH with the exception of decreased risk for higher values of RH and increased risk for average value of RH in coastal areas. In non-coastal areas, mortality risk does not seem vary from the baseline with RH. These findings suggest RH has a limited and inconsistent association with mortality and should not be assessed in isolation from temperature, given their closely linked dynamics. Importantly, these results support that RH may not be an optimal indicator for epidemiological assessments of heat stress [19]. However, this does not diminish the importance of atmospheric moisture more broadly. As discueed earlier, metrics such as WVP may offer stronger and more meaningful associations with heat-related mortality outcomes.

Water vapour pressure.

For the warm period, Fig 2 (bottom panel) shows that in coastal areas, the RR surface shows statistically significant excess mortality risk at both low and high WVP levels, with the effects being more pronounced at shorter lags (0–5 days). Elevated WVP values above 20hPa are associated with a statistically significant increase in mortality risk, suggesting a potential health burden of humid and heat conditions. Similarly, lower WVP values (5–10 hPa) correspond to increased risk across multiple lags, which may reflect the compounding physiological stress under dry and hot conditions. These extremes highlight the vulnerability to both ends of the moisture spectrum in coastal areas. In contrast, the pattern in non-coastal areas is less pronounced, with little evidence of increased risk across the WVP range (the RR estimates remain mostly close to baseline). The overall absence of significant associations in non-coastal areas underscores the importance of geographical context, and support the postulation that WVP holds greater epidemiological relevance in coastal cities, where atmospheric moisture levels are higher.

For the cold period, Fig 3 indicates that higher WVP values result in significantly decreased risk for lags between 0 and 20 days for both coastal and non-coastal cities. There is increased risk for very low WVP values at longer lags, a pattern that may reflect the influence of cold-and-dry weather conditions. While the precise mechanism behind this association is not immediately clear, it likely arises from the confounding effect of temperature, which we explore in greater detail in the synergy analysis in Section. Indeed, this pattern is consistent with the estimated association between Tmax and WVP (Supplementary S2 Fig), lending further confidence to this interpretation. As observed during the warm period, the relative risk (RR) patterns for these two variables align: elevated risk occurs under low-temperature and low-moisture conditions, whereas lower risk is observed at high temperatures coupled with high atmospheric moisture. This co-variation supports the notion that “dry-cold” is a climatological feature of Cyprus, particularly during the cooler months, as shown in the climatological analysis in Fig 1, and may help explain the increased risk observed under such environmental regimes.

These findings support the conclusion that WVP – compared to RH – offers a more accurate and physiologically grounded measure of atmospheric moisture’s impact on heat-related health outcomes, particularly during periods with elevated moisture in the atmosphere. The absence of similar effects in non-coastal areas reinforces this interpretation, as WVP levels in these regions remain low, thus attenuating potential moisture-related contributions to heat stress.

Synergistic temperature-moisture effects on mortality: A coastal vs. inland perspective

Although the effect of RH and WVP on mortality is difficult to interpret when considered individually, they may well have a synergistic effect with temperature, in the sense that the temperature effect may vary (possibly non-linearly) with different levels of RH and WVP. In the subsequent section, we present results from the synergistic model (equation (5)) to investigate this.

Joint effect of temperature and relative humidity.

The RR estimates of the joint effect from Tmax and RH for the warm period are shown in Fig 4. Each panel illustrates the temperature-lag RR surface at specific levels of RH, specifically the 5^th, 25^th,50^th, 75^th and 95^th percentiles of the distribution.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. The RR of Tmax and RH for various values of RH (5^th, 25^th,50^th, 75^th and 95^th percentiles) for the warm period. Left: coastal areas, right: inland areas.

https://doi.org/10.1371/journal.pclm.0000821.g004

Fig 4 presents the warm-period relative risk (RR) estimates for coastal and non-coastal regions based on the synergy model incorporating Tmax and RH (equation 5). In both regions, the highest and statistically significant mortality risks are observed under conditions of high Tmax and low RH, particularly in the RH < 44% panels. Notably, as RH increases, the relative risk associated with elevated temperatures declines, with this attenuation being more pronounced and significant in non-coastal areas. This may reflect greater acclimatisation or adaptive capacity of the population to hot and humid conditions.

In addition, across the entire RH spectrum, both areas exhibit significantly increased mortality risk associated with lower Tmax at long lags. This finding highlights the potential health impacts of unexpected “cooler” conditions during the warm season. Such cold anomalies may lead to disproportionate health effects due to limited physiological acclimatisation and lower risk perception, in contrast to more anticipated heat-related impacts.

We hypothesise that the diminished or non-significant estimates in the synergy model arise from the dependence between RH and Tmax – a relationship that appears stronger than the one between Tmax and WVP. As a result, RH may mask the independent contribution of atmospheric moisture to temperature-related mortality. This is supported by the observation that hot-and-dry conditions are associated with elevated mortality risk in both coastal and inland areas, despite the expectation that coastal populations would be more vulnerable to such extremes. This inconsistency is resolved in the WVP-based models, which better capture the synergistic effects of temperature and absolute humidity.

Accordingly, we suggest that RH may not be the most appropriate metric for investigating compound temperature–moisture effects. The cold season results (Supplementary S3 Fig) support this interpretation, indicating that absolute humidity measures may better reflect the relevant environmental stressors in compound exposure scenarios. Notably, we observed no significant differences between the synergistic effect of temperature and RH across varying levels of relative humidity between seasons.

Joint effect of temperature and water vapour pressure: warm period.

The RR estimates of joint effect from Tmax and WVP for the warm period are shown in Fig 5, stratified by WVP level and region. Overall, the plots indicate that the temperature-risk relationship changes visibly across the range of WVP values in a non-additive way – indicating that the synergy is quite non-linear. This suggests that moisture plays an important role in the effect of temperature on mortality, in both coastal and non-coastal cities in Cyprus.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. The RR of Tmax and WVP for various values of WVP (5^th, 25^th,50^th, 75^th and 95^th percentiles) for the warm period. Left: coastal areas, right: inland areas.

https://doi.org/10.1371/journal.pclm.0000821.g005

Coastal risk: In coastal cities, where levels of moisture are generally higher due to marine influences (as seen in Fig 1), a distinct pattern emerges. The left panel of Fig 5 demonstrates that WVP values ≥ 23hPa in combination with high Tmax result in increased risk, particularly at short delays (0–3 days). This increase in mortality risk points to a synergistic hazard, likely linked with reduced thermo-regulatory efficiency in humid conditions, which reduces evaporative cooling and exacerbates physiological stress. Notably, as WVP decreases (e.g., 11–15 hPa), the intensity of the immediate heat related-risk diminishes. A delayed risk emerges for lower temperature conditions at extended lags (5–10 days). This response may reflect cold-related effects associated with cooler transitional periods occurring during April and October.

Non-coastal risk: Non-coastal areas exhibit elevated mortality risk associated primarily with high temperatures over shorter lags (0–5 days), but only when WVP takes very low values (≤ 13hPa). This suggests that hot-and-dry spells constitute the primary threat to health in inland areas. The combination of intense heat and low moisture may promote dehydration and heat stroke, especially in vulnerable populations. In contrast, when WVP is high (>16hPa), even at elevated temperatures, the mortality risk is positive but weak.

Attributable fraction.

We also present estimates of the attributable fraction (AF) of deaths as defined in Equation (7). Table 1 provides a classification of WVP and Tmax into quantiles, stratified by both coastal and non-coastal areas by season. These classifications provide a consistent framework for interpreting the AF estimates. Fig 6 displays the AF estimates for coastal and non-coastal areas during the warm season, stratified by the Tmax and WVP categories.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Classification of WVP quantiles and temperature categories. For WVP, Q1 = driest, Q5 = most humid.

https://doi.org/10.1371/journal.pclm.0000821.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Attributable fraction for coastal and non-coastal cities in the warm period for the five quantiles of WVP (Q1:Q5) defined in Table 1.

https://doi.org/10.1371/journal.pclm.0000821.g006

Coastal: The highest AFs occur in the lowest quantile (Q1), and as atmospheric moisture increases the AFs drop. Moving from Q1 to Q5, the “extreme cold” AF becomes diminished. “Extreme-heat” AFs also decrease but seem to slightly increase again at Q5. Hot-and-dry episodes appear to have the highest AF numbers. Although the RR results showed high risk for warm-and-humid conditions for the coastal regions, the AF results indicate that perhaps such conditions are less frequent than hot-and-dry.

Non-coastal: Inland areas exhibit the same downward gradient: the AF peaks at the lowest values of WVP in Q1, and declines steadily towards Q5. Extreme heat is the single largest contributor throughout. Interestingly, the highest AF for both extreme cold and hot conditions occurs at the lowest WVP levels. Thus, hot-and-dry conditions dominate the inland burden, while increasing moisture confers a clear mitigating effect on both extreme hot and cold-related conditions. Cold-related fractions again collapse with rising moisture levels: extreme-cold falls from 32% to 17%, and mild-cold from 27% to 10.8%.

These findings challenge the conventional “hot-and-humid” narrative by demonstrating that dryness, rather than humidity, amplifies temperature-related mortality in Cyprus during the warm season, particularly at the extremes of the thermal distribution. Inland regions generally exhibit higher AFs compared to coastal areas, where the highest AFs are under dry conditions. As WVP increases, AFs consistently decline, with the steepest reductions observed at lower temperature ranges.

Interestingly, under extreme heat, AFs do not continue to rise as might be expected. This pattern is explained by the U-shaped nature of the exposure–response relationship depicted in Fig 5, where mortality risk increases at both the cold and hot ends of the temperature spectrum. As a result, when dryness intensifies risk across the full range of temperatures, including the cold extremes, it limits the relative contribution of hot conditions alone to overall AF. This dual-end amplification effectively flattens the increase in AF associated with high temperatures, despite the clear elevation in absolute risk.

Joint effect of temperature and water vapour pressure: cold period.

Fig 7 presents the RR surfaces for coastal and non-coastal cities for the cold season, across a range of WVP values. The analysis reveals important distinctions during the cold season compared to the warm period, particularly in terms of risk magnitude, lag structure, and meteorological thresholds at which risk intensifies.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. The RR of Tmax and WVP for various values of WVP (5^th, 25^th,50^th, 75^th and 95^th percentiles) for the cold period. Left: coastal areas, right: inland areas.

https://doi.org/10.1371/journal.pclm.0000821.g007

Coastal risk: In coastal areas, where humidity levels are generally elevated, a clear synergistic effect between low temperatures and low WVP is evident. At WVP levels (6 – 12hPa), significantly elevated mortality risk emerges at relatively low temperatures (5–15°C), with the most pronounced effects occurring at longer lags (5–15 days). These results suggest that cold conditions under relatively low atmospheric moisture substantially amplify short-term mortality risk.

Non-coastal risk: Non-coastal cities exhibit broadly similar patterns, though with lower overall WVP values and slightly attenuated effects. As shown in the right panel of Fig 7, increased mortality risk again emerges under low temperature and relatively lower WVP conditions (≤ 11hPa).

Overall, the cold period analysis indicates that humidity modulates the temperature-mortality relationship in a distinctly non-linear and location-dependent manner. In both coastal and inland areas, low WVP amplifies the adverse health impacts of low temperatures. The RR profiles become flatter and more diffuse compared to the warm period, indicating weaker associations under cold and dry conditions, which may reflect physiological adaptation or reduced viral transmissibility in drier air. The temporal lag structure of the risk also shifts with moisture, with lower WVP values associated with more widely distributed lag effects, although these remain relatively weak.

Attributable fraction

Compared with the warm period, AFs for the cold season shown in Fig 8 are generally higher, particularly in coastal areas. A notable difference between coastal and non-coastal cities lies in the directionality and magnitude of the AF response across the WVP gradient.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Attributable fraction for coastal and non-coastal cities in the cold period for the five quantiles of WVP (Q1:Q5) defined in Table 1.

https://doi.org/10.1371/journal.pclm.0000821.g008

Coastal: In coastal cities, the mortality AF increases consistently with rising WVP levels across all temperature categories. As expected, AF values are generally elevated under cold conditions, with both extreme and moderate cold associated with substantially higher AF values compared to warmer temperatures.

The highest AF, reaching 43%, occurs under extreme cold conditions combined with the highest WVP level (Q5), indicating a strong interaction between cold and humid environments. Conversely, the lowest AF is associated with extreme heat under dry conditions (Q1) – a relatively infrequent combination in coastal regions during the cold period.

The overall AF values during the cold period remain notably higher than those during the warm period (consistent with recent studies [27]), indicating that the combination of cold temperatures and high humidity in coastal areas poses a particularly high mortality risk.

Non-coastal: In contrast, inland areas exhibit an inverse relationship between AF and increasing WVP levels. Specifically, AFs are highest under low WVP (Q1) across all temperature categories, with the greatest AF observed during extreme cold and dry conditions. This pattern suggests that cold-and-dry environments pose a heightened mortality risk in inland regions, where baseline humidity is lower and thermal sensitivity may be greater.

As WVP increases, AF values decline, particularly under colder temperature conditions. The very high AF values observed (reaching almost 50%) are consistent with existing literature highlighting the strong effects of cold conditions or cold spells on mortality in Cyprus [27]. Additionally, in non-coastal areas, a decrease in WVP from Q5 to Q1 is associated with a steeper increase in AF compared to coastal areas: approximately 10% for extreme cold and 15% for extreme heat. These findings highlight the stronger influence of low moisture on mortality risk in warmer environments in non-coastal settings.

Overall, the AF during the cold period is higher than the warm period, emphasising the modifying effect of WVP on temperature-related mortality – especially in cities with more continental climatic characteristics. In summary, for non-coastal areas, the combination of low WVP and extreme cold temperatures appears to have the most significant impact on population mortality.

Discussion

This study provides a novel assessment and exploration of the joint effects of temperature, RH and WVP on mortality risk, with a particular focus on differences between coastal and non-coastal areas in Cyprus across warm and cold seasons. The results, derived from both single-exposure models and models incorporating synergistic effects of temperature and moisture, reveal distinct patterns in mortality risk across different climatic conditions and geographic locations.

As expected, extreme temperatures, both hot and cold, show significant impacts on mortality. RH emerged as a potentially controversial metric, with its effect on mortality risk not fully aligning with expectations and scientific understanding. In the warm period, the most significant positive relative risk values are observed at low RH levels, even for coastal areas, which is contrary to the understanding that higher humidity would exacerbate heat stress.

The association between RH and mortality continues to present significant interpretative challenges in environmental health research. As noted in prior studies, RH effects are highly context-dependent and may vary substantially across different climates, seasons, and population groups. These inconsistencies point to a deeper methodological issue: RH is not an isolated atmospheric factor but is inherently intertwined with other meteorological variables, particularly temperature and absolute humidity. Studies, such as [41], explicitly caution against interpreting RH without accounting for its linearity with other exposures, as doing so can distort effect estimates and obscure the true drivers of mortality risk. The inability to clearly disentangle RH’s independent impact limits our understanding of its physiological and epidemiological relevance and complicates the development of targeted public health interventions [19,46]. Given the increasing frequency of compound weather extremes under climate change, resolving this ambiguity is not merely a statistical concern but a critical scientific priority with direct implications for climate-adaptive health policy and risk communication.

Moreover, the findings here highlight the need for caution when interpreting the effects of RH on mortality in isolation, especially in the warm period. The results suggest that RH is confounded with temperature, e.g., Fig 2 (middle row of panel on the left) indicates lower-than-average risk for high RH on coastal areas in the warm period, which is rather counter-intuitive. This underscores the importance of considering both temperature and moisture levels when assessing heat-related health risks, particularly in regions where both extremes of temperature and moisture are prevalent.

Water vapour pressure, a key indicator of atmospheric moisture, plays a pivotal role in modulating the effect of temperature on mortality risk. While WVP on its own has minimal impact on mortality during the warm period (non-significant but intuitive effects, such as Fig 2 - bottom), its interaction with temperature reveals important insights. The results during the warm season (Fig 5 -Left) suggest that higher WVP levels in coastal areas amplify mortality risk in both extreme low and high temperature conditions. This finding aligns with the hypothesis that high humidity may exacerbate respiratory stress and increase susceptibility to infectious diseases, a known risk factor especially during colder conditions [18]. In contrast, lower WVP levels in non-coastal areas (Fig 5 - Right), especially in dry conditions, seem to elevate mortality risk during extreme heat, particularly in individuals exposed to these conditions for prolonged periods. These findings suggest that areas with more continental climates, such as inland cities, may be more vulnerable to the effects of extreme heat combined with low humidity. This inversion of risk pattern compared to coastal areas highlights geographical variations.

The attributable fraction (AF) analysis indicates that for the warm period, coastal cities experience a higher AF when WVP is low, especially during extreme heat. This contrasts with the relative risk (RR) analysis, which shows that hot-and-humid conditions are associated with the highest instantaneous risk. The discrepancy reflects differences in exposure frequency: hot-and-humid events, although more hazardous, occur less frequently than hot-and-dry conditions in coastal areas of Cyprus and therefore contribute less to the overall mortality burden. Interestingly, extreme cold combined with low WVP is also associated with a high AF, suggesting that cold-and-dry conditions can be particularly detrimental, likely due to factors such as respiratory infections, especially in non-coastal areas [47].

In the cold period, notable differences emerge in the AF patterns between coastal and non-coastal areas. Coastal areas show higher AF values during extreme cold and high WVP conditions (compared to lower WVP conditions), highlighting the heightened vulnerability of populations in coastal regions during cold-and-humid conditions. This result underscores the need for targeted interventions, such as improving indoor heating systems and enhancing public health campaigns focused on respiratory infections during colder months. On the other hand, in non-coastal areas, the AF is higher under cold-and-dry conditions, suggesting that inland populations are more susceptible to cold and dry spells, potentially due to a greater thermal sensitivity and lower baseline humidity.

Collectively, these findings highlight the need to jointly consider thermal and moisture conditions when assessing climate-related health risks. The observed lag structures and interaction patterns support the integration of compound climate metrics into early warning systems, particularly as temperature extremes and atmospheric moisture intensify under climate change. Several limitations should nevertheless be acknowledged. Despite high temporal resolution and detailed demographic information, the small population size and limited geographic scope of Cyprus restrict statistical power for cause and sex-specific mortality analyses. While this limits conclusions regarding specific disease pathways, uncertainty is fully quantified, and the associations remain internally consistent.

Conclusion

Mortality responses to temperature extremes are systematically shaped by their interaction with atmospheric moisture, revealing compound climate-health risks that cannot be captured by temperature alone. Using 16 years of age-stratified mortality data and high-resolution climate observations from Cyprus, we demonstrate that health risks emerge from interacting environmental drivers rather than from isolated exposures, even within a geographically compact region.

By explicitly comparing relative humidity with water vapour pressure, an absolute measure of atmospheric moisture, we show that commonly used humidity metrics can obscure critical health signals. Relative humidity yields inconsistent and often counter-intuitive associations with mortality, particularly during warm periods, whereas WVP provides a more physiologically meaningful and statistically robust basis for quantifying temperature–moisture interactions. This distinction allows clearer identification of harmful compound exposures that remain undetected in single-exposure or RH-based analyses.

Our results reveal pronounced spatial heterogeneity in compound climate–mortality relationships, with distinct vulnerability profiles in coastal and non-coastal environments. Coastal populations exhibit heightened sensitivity to hot-and-dry and cold-and-humid conditions, while inland populations are more vulnerable to hot-and-dry and cold-and-dry extremes based on the AF results. These contrasts emerge over short spatial scales, challenging assumptions of homogeneous climate vulnerability and underscoring the importance of sub-regional analyses in climate–health research.

Quantification of attributable mortality further demonstrates that a substantial proportion of climate-related deaths arises from frequent compound temperature–moisture conditions rather than from rare extremes alone. This finding highlights a critical gap in current risk assessments and suggests that public health burdens may be systematically underestimated when compound exposures are not explicitly considered.

Conceptually, this work establishes atmospheric moisture as a fundamental modifier of temperature–mortality relationships, advancing climate–health research beyond single-exposure paradigms. Methodologically, we introduce a transferable framework for modelling joint, lagged, and non-linear effects of temperature and atmospheric moisture on health outcomes, enabling consistent evaluation of compound climate risks across diverse settings.

Although demonstrated for Cyprus, this framework is directly applicable to other countries in the Middle East and North Africa, where rapid warming coincides with strong moisture gradients, and across Europe, where heat–humidity extremes are projected to intensify under climate change. By integrating absolute humidity metrics into epidemiological analyses, this study provides a generalizable pathway for improving early warning systems, refining climate–health risk assessments, and informing adaptation strategies in a warming world.

Supporting information