The Range of Projected Change in Vapour Pressure Deficit Through 2100: A Seasonal and Regional Analysis of the CMIP6 Ensemble

Abstract

Vapour pressure deficit (VPD) is frequently used to assess the impact of climate change on wildfires, vegetation, and other phenomena dependent on atmospheric moisture. A common aim of projection studies is to sample the full range of changes projected by climate models. Although characterization of model spread in projected temperature and rainfall is common, similar analyses are lacking for VPD. Here, we analyze the range of change in projected VPD from a 15-member CMIP6 model ensemble using the SSP-370 scenario. Projected changes are calculated for 2015–2100 relative to the historical period 1850–2014, and the resulting changes are analyzed on a seasonal and regional basis, the latter using continents based on IPCC reference regions. We find substantial regional differences including higher increases in VPD in areas towards the equatorial regions, indicating increased vulnerability to climate change in these areas. Seasonal assessments reveal that regions in the Northern Hemisphere experience peak VPD changes in summer (JJA), correlating with higher temperatures and lower relative humidity, while Southern Hemisphere areas like South America see notable increases in all seasons. We find that the mean projected change in seasonal VPD ranges from 0.02–0.23 kPa in Europe, 0.04–0.19 kPa in Asia, 0.02–0.16 kPa in North America, 0.15–0.33 kPa in South America, 0.10–0.18 kPa in Oceania, and 0.21–0.31 kPa in Africa. Our analysis suggests that for most regions, no two models span the range of projected change in VPD for all seasons. The overall projected change space for VPD identified here can be used to interpret existing studies and support model selection for future climate change impact assessments that seek to span this range.

1. Introduction

The hydrological cycle plays a fundamental role in maintaining a dynamic equilibrium of Earth’s ecosystem through the process of evaporation, transpiration, and transpiration. Global warming accelerates the water cycle [1,2], leading to more extreme wet and drought events, exhibiting spatial and seasonal variability [3,4]. The increase in temperature has intensified the effect of evapotranspiration, accelerating land drying and leading to drought events in arid regions [5]. Vapor pressure deficit (VPD) is the difference between the actual vapor pressure and the saturated vapor pressure, based on the given temperature and atmospheric pressure. According to the Clausius–Clapeyron equation [6] (Clausius–Clapeyron equation: ${\displaystyle \frac{d(ln{P}^{\mathrm{sat}})}{d(1/T)}}=-{\displaystyle \frac{\mathsf{\Delta }{H}_{\mathrm{vap}}}{R\mathsf{\Delta }{Z}_{\mathrm{vap}}}}$, where $P^{\mathrm{sat}}$ is saturated vapor pressure, T is temperature, ${\displaystyle \frac{H_{\mathrm{vap}}}{R}}$ is the enthalpy of vaporization relative to the gas constant, and $Z_{\mathrm{vap}}$ is the compressibility upon vaporization), it is well-established that an increase in air-surface temperature will result in a rise of saturated vapor pressure, further promoting an increase in VPD. The measurement of VPD is crucial because changes in VPD are highly correlated with natural disasters induced by extreme climate events, including wildfires and agricultural drought events [7,8,9].

Wildfires can cause significant ecological and economic losses [10], and climate change is projected to intensify wildfire activity, leading to increased frequency and more severe fire events [11]. In the context of global warming, the continuous increase in temperature coupled with changes in relative humidity (RH) can lead to an increase in VPD, which is strongly correlated with global fire activity [12]. The drivers of fire include fuel amount and dryness, weather conditions, and ignitions [13]. VPD is a strong predictor of dead fine fuel moisture [14] and the ignitability of litter beds, including dead leaves and twigs [15]. An increase in VPD leads to a decrease in fuel moisture, which then increases both the ignitability and the spread rate of the fire [16]. Increases in VPD not only promote soil moisture evaporation, leading to an increase in broad-scale evapotranspiration, but also aggravate the atmospheric aridity, providing weather conditions that are highly positively correlated with forest fires [17,18]. Studying climate change impacts on VPD is essential given its centrality in understanding and predicting wildfire activity [19].

In the field of plant physiology, VPD is directly correlated to transpiration, water use efficiency, and stomatal acclimation in plants, and an increase in VPD would negatively affect global vegetation growth, crop yields, and productivity [20,21]. During the growth of plants, water plays a key role as a solvent that transports nutrients, which are absorbed from the soil through the roots and distributed throughout plant stems [22]. Stomata are small pores on the surface of leaves that allow plants to regulate water evaporation during transpiration and facilitate gas exchange for photosynthesis [23]. An increase in VPD and temperature can cause soil drying. In response, plants are able to close their stomata to prevent water loss, but this reduces transpiration activities and stem conductivity, further negatively affecting plants’ nutrient absorption and growth [24]. When temperature and VPD increase, the closure of stomata further reduces gas exchange for photosynthesis, ultimately impacting the plant’s synthesis of glucose and growth [25]. In an agricultural context, an increase in VPD is associated with a decrease in crop yields [26]. In addition to fire, studying climate change impacts on VPD is critical to understanding the future of global agriculture and plant health.

Climate projection studies use global climate models to estimate and predict long-term climate changes in the future based on given scenarios. The Coupled Model Intercomparison Project (CMIP) is an international scientific project that allows climate models from different research teams to be contributed and evaluated on a global scale through standardized scenario designs, which provide comparable output results of climate change predictions. Many research teams have been using CMIP models’ outputs to project and study climate change patterns, providing insights for adaptation and other policies to mitigate and reduce environmental and climate degradation [27,28,29]. For example, multiple studies have projected increasing patterns of VPD by the end of the 21st century. This includes a median increase of 51% in summer VPD over the continental U.S. [30], strong VPD increase in arid regions globally [31], and a global land drying trend under high-emission scenarios that is driven much more by VPD than precipitation [32].

A key way to add value and robustness to climate projections is understanding the makeup of the climate models from which the projections are drawn [33,34]. Methods here include evaluating models for their ability to simulate specific variables (generally temperature and precipitation) at given spatial and temporal scales, assessing the independence of models and understanding the range of projected change across all models [35,36,37,38]. Having this knowledge not only makes it possible to interpret the results of studies that use a subset of CMIP6 models—it also facilitates the development of objectively designed ensembles, e.g., regional climate modeling experiments. While there has been some evaluation of models’ ability to simulate VPD and broad patterns in projected VPD [31,32], as yet there have been no global assessments at a regional and seasonal scale of the range of projected changes in VPD from the CMIP6 ensemble. This represents a key barrier to interpreting existing studies and generating new ensembles. In this study, we analyse projected changes in VPD for the period 2015–2100, relative to 1850–2014, on a regional and seasonal basis. We aim to characterise the range of change, i.e., the spread between models for individual regions and seasons. We also aim to explore variation in the drivers of VPD, temperature, and humidity.

2. Methods

2.1. Analysis

In this study, we analyze the projected change in VPD on both seasonal and regional scales. The change in VPD is calculated as the average difference between the VPD of the past period (1850–2014) and that of the future period (2015–2100). For the seasonal scale, we used the standard four seasons, focusing on month identities rather than season identities; e.g., DJF represents December, January, and February, which is summer in the southern hemisphere but winter in the northern hemisphere. All analysis and visualisation was conducted using R version 4.4.2 (R Core Team, 2024), R Foundation for Statistical Computing, Vienna, Austria [39]. In the data processing steps, we used IPCC WGI reference regions (4th edition) as boundary vector data for region identity, which provides an updated reference region to fit climate models for analysis [40]. The entire IPCC WGI reference regions consist of 46 land regions and 15 ocean regions. In this study, our regions of analysis are the six inhabited continents, based on IPCC climate reference regions. In addition to examining the range of projected regional change in seasonal VPD, we examine how these changes co-vary with changes in air-surface temperature and relative humidity.

2.2. Model Selection

CMIP6 has a variety of scenarios that simulate future climate change under different levels of Shared Socioeconomic Pathways (SSPs). To improve clarity and meet computational constraints, we chose a single SSP, SSP-370, as the target experiment. SSP-370 is a greenhouse gas emissions pathway without aggressive mitigation of gas emissions, placing it above the lower emissions pathways (SSP-126 and SSP-245), but below the highest emissions pathway (SSP-585). The changes in VPD under SSP-370 are correspondingly significant but not as extreme as SSP-126, SSP-245, and SSP-585, which makes it an informative guide for our research [31]. Focusing on only one scenario helps avoid unnecessarily complicating visualization involving multiple emission scenarios. The selected models required outputs of temperature, humidity, and air-surface pressure for VPD calculation. We also chose ‘r1i1p1f1’ as a filter tag since this is the most common realization across the CMIP6 ensemble, which maximizes the sample size for our study. For seasonal analysis purposes, we limited the timestep to monthly scales to avoid wasting computational resources deriving seasonal values from daily or sub-daily data. The grid spacing of the models used varies from 100 to 300 km. There is some evidence that model differences in physics and surface properties are more important than grid spacing differences [41]. All models were regridded to a common horizontal grid of $1^{\circ }\times 1^{\circ }$. By limiting the selection to the described criteria, we selected 15 models as target models that met our research needs (

Table 1. Selected CMIP6 models that have necessary variables to calculate VPD. All 15 models are listed in alphabetical order from left to right, top to bottom.

List of Selected Models
ACCESS-CM2	ACCESS-ESM1-5	AWI-CM-1-1-MR	BCC-CSM2-MR	CanESM5
CanESM5-1	CESM2-WACCM	CMCC-CM2-SR5	CMCC-ESM2	MIROC6
MPI-ESM1-2-HR	MRI-ESM2-0	NorESM2-LM	NorESM2-MM	TaiESM1

).

2.3. VPD Calculation

VPD is calculated as the difference between the saturated water vapour pressure $\left(e_s\left(T\right)\right)$ and the actual water vapour pressure $\left(e_a\right)$, with units kPa.

$$VPD=e_s\left(T\right)-e_a=e_s\times \left(1-{\displaystyle \frac{RH}{100}}\right)$$

(1)

The saturated vapour pressure $e_s\left(T\right)$ is calculated by using Tetens’ equation with air-surface temperature (T) in degrees Celsius [42], with an adjustment based on the temperature range to avoid errors when temperature is below 0 °C [43].

$$e_s\left(T\right)=\left\{\begin{array}{cc}0.611\times exp\left({\displaystyle \frac{17.27\times T}{T+237.15}}\right),\hfill & \mathrm{if}T>0^{\circ }\mathrm{C}\hfill \\ 0.611\times exp\left({\displaystyle \frac{21.87\times T}{T+265.5}}\right),\hfill & \mathrm{if}T<0^{\circ }\mathrm{C}\hfill \end{array}\right.$$

(2)

The actual vapour pressure is calculated from the relative humidity (RH) and the saturation vapour pressure.

$$e_a=e_s\left(T\right)\times \left({\displaystyle \frac{RH}{100}}\right)$$

(3)

In Formula (3), the relative humidity was calculated based on specific humidity (huss), using a formula adapted from the $wrf\_rh$ function in NCAR Command Language (NCL) [44]. Within this step, we firstly used huss to calculate the mixing ratio (QV). Then we used the result of $\left(e_s\right)$ in Equation (2) to calculate the saturation mixing ratio (QVS). In the last step, QV and QVS are used to calculate relative humidity. The units for both QV and QVS are (kg/kg).

$$QV=huss/(1-huss)$$

(4)

$$QVS=0.622\times \left({\displaystyle \frac{10 \times e_s}{0.01\times ps-(1.0-0.622)\times 10\times e_s}}\right)$$

(5)

$$RH={\displaystyle \frac{QV}{QVS}}\times 100$$

(6)

3. Results

3.1. VPD Difference by Regions

Projected changes in mean annual VPD exhibit regional variability (Figure 1). There is some indication of variation with latitude, with generally high projected changes around 20° S and 20° N, and smaller increases or even decreases at other latitudes. However, there are exceptions. Strong increases in VPD are projected in equatorial South America, whereas equatorial Africa and Southeast Asia are projected to have more moderate increases. At high latitudes (below 40° S and above 50° N, projected increases tend to be smaller. The only region projected to have substantial decreases in VPD is a small area centred on the Himalayas in Asia. In terms of total land area, projected increases in VPD are generally greater in the Southern Hemisphere than the Northern Hemisphere, where substantial parts of Canada, Russia, and Greenland are projected to have relatively little change in VPD.

There were clear seasonal patterns in the three Northern Hemisphere regions, with significant increases in VPD (>0.15 kPa) projected for all during the summer (JJA) season and smaller projected increases (<0.10 kPa) were predicted in the other three seasons (Figure 2). In the Southern Hemisphere, Oceania had the smallest seasonal fluctuations (seasonal range around 0.08 kPa) and the lowest overall magnitude of change (between 0.10 and 0.18 kPa). Oceania’s winter season (JJA) showed the lowest VPD change (0.10 kPa) of all seasons and regions in the Southern Hemisphere. The overall magnitude of seasonal increase in VPD was greatest in Africa (0.21–0.31 kPa), but like Oceania, the seasonal cycle was less pronounced (seasonal range around 0.10 kPa). The projected increase in VPD was largest in the JJA season (0.31 kPa) and smallest in the DJF season (0.21 kPa). In South America, the biggest increase in VPD change was projected to occur during the SON season (0.33 kPa), with more moderate increases in the other seasons (0.15–0.20 kPa). This trend in South America of marked changes during one season and more moderate changes in the other seasons is similar to Northern Hemisphere patterns.

3.2. RH and Temperature Change by Region

In order to understand the drivers of projected changes in VPD, we analyzed the seasonal patterns in its constituent variables, i.e., temperature and RH (Figure 3). Temperature was projected to increase across all regions in all seasons, and RH was projected to decrease in most of the seasons except Asia (SON), Oceania (DJF), and Africa (DJF, JJA & SON). Regions in the Southern Hemisphere had less seasonal variation in projected temperature increase (2.1–2.8 °C) compared to regions in the Northern Hemisphere (3.0–6.3 °C). The winter season (DJF) of each region in the Northern Hemisphere had the highest temperature increase (3.7–6.3 °C). The temperature changes in each season were of broadly similar magnitude in most regions (particularly in the Southern Hemisphere) but RH exhibited pronounced seasonal variability. In the South Hemisphere, South America had the greatest seasonal variation (from −3.4 to −1.3%), but Oceania (from −0.5 to 0.1%) and Africa (from −0.2 to 0.5%) exhibited less seasonal variation. In the Northern Hemisphere, the greatest seasonal variation of RH happened in Europe (from −2.7 to −1.3%), and Asia (from −0.3 to 0.1%) and North America (from −1.4 to −1.2%) had much less seasonal variations. Comparing Figure 2 and Figure 3, the strong Northern Hemisphere temperature increases across all seasons contrast strongly with VPD, which is only projected to increase strongly in summer (0.16–0.23 kPa) in compared with VPD change in other seasons (0.02–0.09 kPa). Note that VPD changes cannot easily be deduced from changes in temperature and relative humidity because of regional differences in baseline temperatures (see Section 3.3 for greater discussion of this point).

3.3. Range of VPD Change Across Models

The contribution of projected changes in relative humidity and temperature to VPD varied by region (Figure 4). Africa, Oceania, and South America had the highest VPD changes (>0.20 kPa), despite having similar magnitude changes in temperature (1.4–4.4 °C) and RH (from −8.4 to 2.7%). In the Northern Hemisphere, multiple models show a lower VPD change (<0.15 kPa) compared with regions in the Southern Hemisphere. Across all seasons, there was greater variation in projected changes in temperature in the Northern Hemisphere compared with the South Hemisphere. For instance, during the SON season the projected change in temperature in the Northern Hemisphere ranged from 2.0 and 7.6 °C, but was 1.4–4.4 °C in the South Hemisphere. The greatest range in projected temperature increase was projected for North America in winter (DJF), from 4.1 to 10.0 °C. The greatest range in projected RH change was projected for South America in SON season, from −8.4 to −1.0%. Regions located near the equator and tropics generally had the highest VPD changes, despite having similar magnitude changes in mean surface temperature and RH compared with other regions. These results can be explained by the VPD formula defined in Section 2.3. The equation for VPD shows that the saturated vapor pressure contains an exponential function with temperature as input. The VPD formula indicates that a unit increase in temperature results in exponential growth in VPD, which explains why Africa, Oceania, and South America experience higher VPD changes. With their higher baseline temperatures, these regions will undergo a greater change in VPD for a given change in temperature. The dependence of VPD changes on baseline temperatures suggests that climate change impacts related to VPD will differ significantly between regions. The outputs of different models exhibited particularly strong differences in projected RH, which led to further differences in VPD.

Examination of individual regions reveals variation between models in terms of the seasonal magnitude of VPD change. In Oceania (Figure 5), the CanESM5 and CanESM5-1 models project the greatest overall change in VPD during DJF and SON, but other models produce higher projected changes in MAM (ACCESS-CM2) and JJA (ACCESS-ESM1-5). MIROC6 projects the lowest increase in VPD in DJF and MAM, but not in JJA or SON (both NorESM2-MM). There are some similarities in these patterns in South America (Figure 6). CanESM5 and CanESM5-1 project the greatest increases in VPD in all four seasons, but the lowest increases vary by season: CMCC-ESM2 in DJF and MAM, MRI-ESM2-0 in JJA, and NorESM2-MM in SON. This highlights the need to take into account region and season when considering models that span the range of projected change in VPD. Readers interested in exploring these results in greater detail are referred to the Supplementary Materials.

4. Discussion

Our research provides a thorough analysis of projected changes in VPD using data from the CMIP6 climate models. By examining the difference between historical and SSP370 future scenarios across the six inhabited continents, the study highlights significant regional and seasonal variability in projected changes in VPD. Regions in the Southern Hemisphere near the equator, such as Africa, Oceania, and South America, exhibit higher VPD changes than other regions due partly to their elevated baseline temperature. This leads to significantly increased VPD across all seasons in these regions. In terms of seasonality, VPD increases are greatest during summer months (JJA) in the Northern Hemisphere. There is more variation in the Southern Hemisphere, with the greatest increase in VPD occurring during spring (SON) in South America, summer (DJF) in Oceania, and winter (JJA) in Africa.

The projected changes in VPD correspond to strongly divergent patterns in projected relative humidity and temperature. Increases in VPD in South America appear to be partly driven by strong declines in relative humidity, particularly in spring (SON). Due to climate change, the length of the dry season in South America is predicted to increase in the future, and the extension of the dry period directly intensifies the VPD change [45]. Europe and North America are also projected to undergo substantial declines in relative humidity. Increases in VPD in Asia appear to be driven more by temperature increases than changes in relative humidity.

The Northern and Southern Hemispheres exhibit pronounced differences in both VPD and temperature changes. Compared with the projected results in the Northern Hemisphere, temperature and VPD changes in the Southern Hemisphere are relatively uniform. One possible source of the relatively pronounced increases in Northern Hemisphere winter temperature, compared to the Southern Hemisphere, is the strong influence of the cryosphere. In the cold season, the Arctic discharges heat that has been absorbed as a result of sea-ice loss during the warm season, contributing to a phenomenon known as Arctic amplification [46]. Climate change impacts on the cryosphere, and downstream impacts on VPD, may therefore be greater in the Northern Hemisphere [47].

CMIP6 models generally have higher equilibrium climate sensitivity (ECS) than CMIP5 models (between 1.8 to 5.6 K) due to their strengthening of cloud feedback [48]. Within our selected models, CanESM5 has high climate sensitivity due to its settings of cloud and sea-ice albedo feedback, which is around 5.6 K [49]. This may explain why CanESM5 models generally exhibit much higher projected changes in VPD than other models. The models’ variability and its inter-model feedback settings also suggest the importance of model selection for projection studies, particularly when assessing multi-models for climate change and extreme events risk analysis. Users interested in sampling from the full range of projected changes in VPD for a given season or region may also want to consider the spread of temperature and relative humidity for their selected models. Users interested in understanding patterns for specific regions, seasons, models, and permutations of temperature and relative humidity are directed to the Supplementary Materials and the freely available data included as part of this article. Our scale of analysis could underestimate the severity of VPD changes in certain regions. A sensitivity analysis using IPCC sub-regions would provide additional information about the robustness of the patterns we have identified for different continents.

This study is limited to the SSP370 scenario, and these results may not be representative of patterns in regional and seasonal VPD change for other SSPs. Without deep and rapid cuts to greenhouse gas emissions, these results will underestimate changes in VPD by the end of the century [50]. Conversely, with a greater uptake in net-zero greenhouse gas emissions pledges, these results may overstate projected changes in VPD [51]. Similarly, our analysis is restricted to the six continents and this scale may not be suitable for users interested in smaller regions or countries. For example, the northern and southern parts of Africa exhibit much higher projected VPD changes than Central Africa. Our scale of analysis could underestimate the severity of VPD changes in certain regions. The same argument applies to the statistics we have used. We show changes in projected seasonal mean VPD and users interested in extreme values may benefit from an analysis based on daily values or higher moments of the distribution (e.g., 90th or 95th percentile VPD).

The projected increases in VPD are likely to have major impacts across the globe. Researchers have studied wildfire data in the southwestern US from 1984 to 2015 to build two statistical models analyzing VPD as a factor for areas burned at high severity. They found that increases in VPD—particularly minimum VPD—were associated with increases in high-severity fire [52]. Using these statistical models with CMIP6 predictions as input, it is estimated that there will be a significantly increased risk for high-severity fire activities around the world. Australia, South America, and Africa are three major regions where natural ecosystems are significantly threatened by wildfires, and these regions have the highest projected increases in VPD. The “Black Summer” fires of 2019–2020 in Australia were some of the most destructive wildfires in human history. The fires burned for more than eight months across multiple states, causing tens of billions of dollars in economic losses, damaging over 4000 houses and facilities, and forcing more than 65,000 people to evacuate [53].

An increase in VPD accelerates soil drying and exacerbates drought, with the severity of this problem varying by region. Food shortages are a long-term, unresolved challenge in Africa, with drought being one of the leading factors [54]. Our results suggest that the highest increases in VPD across multiple seasons are projected to occur in Africa (Figure 2), with likely negative impacts on agriculture yields. These results suggest a potential challenge for ecological and agricultural development in Africa, highlighting the need for applied research to prevent the severity of the food crisis from continuing to increase. Regions like South America may also face food crises due to expanding drought-affected areas and climate change, which require further research and attention. Australia is also one of the major agricultural export nations which benefits from its rich land resources and modern production systems. However, Australia is projected to have relatively high increases in VPD, which may lead to significant economic losses in agriculture.

5. Conclusions

Robust adaptation to climate change requires impact assessments that draw on the full range of projected changes in climate. Until now, there has been no global assessment of seasonal and regional variation in the range of projected change in VPD, a key predictor of wildfire risk, vegetation condition, and agricultural productivity. We characterised the projected change space for seasonal VPD across global land regions using 15 CMIP6 models, identifying a wide range of similarities, differences, and trends. This information can be used to support interpretation and utilisation of previous studies and the development of new assessments that seek to span the range of projected changes in VPD.