Abstract
Global warming is expected to cause wet seasons to get wetter and dry seasons to get drier, which would have broad social and ecological implications. However, the extent to which this seasonal paradigm holds over land remains unclear. Here we examine seasonal changes in surface water availability (precipitation minus evaporation, P–E) from CMIP5 and CMIP6 projections. While the P–E seasonal cycle does broadly intensify over much of the land surface, ~20% of land area experiences a diminished seasonal cycle, mostly over subtropical regions and the Amazon. Using land–atmosphere coupling experiments, we demonstrate that 63% of the seasonality reduction is driven by seasonally varying soil moisture (SM) feedbacks on P–E. Declining SM reduces evapotranspiration and modulates circulation to enhance moisture convergence and increase P–E in the dry season but not in the wet season. Our results underscore the importance of SM–atmosphere feedbacks for seasonal water availability changes in a warmer climate.
Introduction
Changes in surface water availability (defined as precipitation minus evapotranspiration, P–E) over land have widespread consequences for human and natural systems in a warmer climate^{1,2}. For example, alterations in the seasonal patterns of precipitation and evapotranspiration may enhance flood and drought risks^{3,4}, and pose great challenges to local populations, food security, and sustainable management of water resources. Global warming increases water vapor in the atmosphere. This increase is generally expected to amplify the existing spatial as well as seasonal patterns of P–E, leading to wet regions/seasons getting wetter, and dry regions/seasons getting drier, which is referred to as “wet get wetter, dry get drier” (WWDD) mechanism^{5,6,7,8}. While evidence for the seasonal pattern of WWDD, reflected in both precipitation and P–E, has been found at global and regional scales, mainly over extratropical regions^{6,9,10,11}, over some subtropical dry regions, observations and model projections point to the opposing pattern of seasonal change, with wet seasons becoming drier and dry seasons becoming wetter (WDDW)^{9,11}. This unexpected WDDW pattern, if demonstrated to be mechanistically plausible, would have important implications for the reliability of water resources and the sustainability of terrestrial ecosystems, particularly in dry regions. It is therefore crucial to identify the extent to which the WDDW pattern is robust over land and determine the underlying mechanisms involved.
In this work, we examine projected seasonal pattern of P–E changes from the Coupled Model Intercomparison Project Phase 5 (CMIP5)^{12} and Phase 6 (CMIP6)^{13} and identify the thermodynamic and dynamic mechanisms responsible for seasonal P–E changes using land–atmosphere coupling sensitivity experiments from CMIP6. We find a robust pattern of increasing dry–season P–E and decreasing wet–season P–E over subtropical regions and the Amazon, which is dominated by seasonally varying soil moisture–atmosphere feedbacks, as drying of the soil reduces evapotranspiration and modulates atmospheric circulation to enhance moisture convergence and increase water availability in the dry season but not in the wet season. These results advance our understanding of seasonal shifts in water availability and underscore the need for more in–depth assessments of hydrological changes over subtropical dry regions.
Results
Projected seasonal pattern of water availability changes
We use multi–model simulations from CMIP5 and CMIP6 to investigate seasonal changes in P–E between historical (1971–2000) and future (2071–2100, high–end forcing) periods (Methods). For each model, we define the dry and wet seasons as the three consecutive months with the lowest and highest climatological mean P–E in the historical period, respectively (Supplementary Fig. 1). The spatial patterns of seasonal changes in P–E between CMIP5 and CMIP6, as shown in Fig. 1, are highly correlated (r > 0.83). Over many extratropical regions, both CMIP5 and CMIP6 project increasing wet–season P–E and decreasing dry–season P–E, thereby leading to enhanced P–E seasonality (Fig. 1a–f). The seasonality of P–E is also enhanced over tropical Africa and Southeast Asia, mainly due to wet–season P–E increases, with a larger amplitude evident in CMIP6 relative to CMIP5. The enhancement of the annual P–E range is consistent with previous studies indicating that global warming increases water vapor in the atmosphere and amplifies horizontal water vapor transport to strengthen the seasonal cycle of water availability^{5,11,14}, while changes in atmospheric dynamics may also modify seasonal shifts in P–E^{6,9}. However, CMIP5 and CMIP6 also project some regions with wet–season P–E decreases and dry–season P–E increases, mostly over the subtropics and the Amazon. Although the magnitudes of seasonal P–E changes are uncertain and model dependent, the signature of the drying of wet seasons and the wetting of dry seasons (i.e., WDDW) is significant (p value < 0.05, see Methods) for ~20% of global land area (excluding Antarctica and Greenland), leading to reduced seasonality of P–E (Fig. 1a–f). Further, comparison of seasonal cycles of P–E between historical and future periods also demonstrates opposing seasonal shifts in P–E over the reduced and enhanced seasonality regions (Fig. 1g, h). We have also tested the sensitivity of our results to defining the dry and wet seasons as the three consecutive months with lowest and highest climatological mean P–E in each of the historical and future periods for each model (Supplementary Fig. 1). Overall, while this method yields a smaller percentage (~10% of global land area) of reduced seasonality regions (Supplementary Fig. 2), the spatial patterns of the recalculated seasonal changes of P–E are similar to those in Fig. 1, with spatial correlations of 0.94 and 0.98 for the dry and wet seasons, respectively.
Mechanisms of seasonal water availability changes
Long–term P–E changes are driven by both thermodynamic and dynamic processes, which have been widely investigated^{15,16,17}. Global warming, from a thermodynamic perspective, is expected to increase atmospheric water vapor and horizontal moisture transport, favoring increased PE over wet regions of the tropics and extratropics and reduced PE over subtropical dry regions^{5,17,18}. This thermodynamic effect should also apply for seasonal changes in water availability, with wet seasons getting wetter and dry seasons getting drier^{6,9,11}. On the other hand, atmospheric dynamic processes driven by ocean and land region warming (and their contrasts) may also drive PE change. For example, the potential expansion of the Hadley cell shifts the descending branches poleward and causes subtropical drying^{19,20,21,22}. This dynamic effect may contribute to the drying of wet seasons, particularly to the extent that associated poleward displacements of storm tracks shift the locus of rainproducing synoptic disturbances poleward, although it is unclear that this should account for the wetting of dry seasons. Overall, the existing thermodynamic and dynamic mechanisms appear to be insufficient to explain the projected WDDW pattern over subtropical regions and the Amazon.
A recent study indicates that soil moisture (SM) also plays an important role in regulating longterm PE changes^{23}. Over subtropical regions and the Amazon, SM is projected to decrease (Fig. 2), which strongly limits evapotranspiration and reduces moisture recycling for precipitation^{24,25}. However, by shifting the surface turbulent flux partitioning toward sensible heating, i.e., increasing the surface Bowen ratio, reduced SM may enhance lowlevel flow convergence and associated moisture convergence, thereby contributing to weaker declines in precipitation than in evapotranspiration, resulting in an increase in PE and a negative SM feedback on PE^{23}. This implies that increasing dryseason PE may be associated with local drying of the soil.
To examine whether SMatmosphere feedbacks can explain the WDDW pattern, we first compare future minus historical SM changes between the wet and dry seasons. In both CMIP5 and CMIP6, we find declining SM in the WDDW regions for both wet and dry seasons, with small interseasonal differences (Fig. 2). This indicates that the WDDW pattern is not due to seasonally asymmetric SM changes. On the other hand, it has been demonstrated that the SM limitation on evapotranspiration is stronger under drier conditions^{24}, and the SM regulation of precipitation is also intrinsically linked to the SM effect on evapotranspiration^{26}. We thus hypothesize that the SM effects on evapotranspiration, and hence PE, may vary seasonally, contributing to the seasonal PE changes and the WDDW pattern over subtropical regions and the Amazon.
Seasonally varying soil moisture effects on water availability
To investigate the hypothesized SM effect on evapotranspiration and seasonal PE changes, we take advantage of a multimodel CMIP6 ensemble from the Land Feedback Model Intercomparison Project with prescribed Land Conditions (LFMIPpdLC)^{27}. LFMIPpdLC is identical to the historical and future (SSP585) simulations throughout the simulation period 1980–2100, except that SM is prescribed as the mean seasonal cycle over 1980–2014 from the historical simulation in each of the five participating models (CESM2, CNRMCM61, ECEarth3, IPSLCM6ALR, MPIESM12LR; see Methods). We isolate the SM effect on seasonal PE changes between the historical (1980–2000) and future (2080–2100) periods as the fivemodel mean difference between CMIP6 and LFMIPpdLC.
Our comparison of the CMIP6 and LFMIPpdLC simulations reveals distinct SM effects on evapotranspiration and PE between wet and dry seasons (Fig. 3 and Supplementary Fig. 3). As expected, the SM limitation on evapotranspiration is stronger in the dry season than the wet season over subtropical regions and the Amazon (Supplementary Fig. 3d, e). In the wet season, PE changes in CMIP6 and LFMIPpdLC are comparable over most land area (Fig. 3a, d), and the SM effect on PE changes is relatively small (Fig. 3g), especially in the Northern Hemisphere, compared to PE changes induced by other processes, such as anthropogenic climate change, in LFMIPpdLC (collectively, we term these the nonSM effect). The drying of the wet season over subtropical regions and the Amazon is therefore mainly caused by the nonSM effect (Fig. 3a, d, g), as anthropogenic warming reduces precipitation but enhances evapotranspiration with prescribed SM in LFMIPpdLC (Supplementary Fig. 4a, d). However, the SM effect is opposite, and of similar magnitude, to the nonSM effect on PE changes in the dry season (Fig. 3e, h). Over subtropical regions and the Amazon, the nonSM effect reduces PE and SM in both seasons (Figs. 3d, e, 2g, h). SM drying and associated landatmosphere processes strongly increase dryseason PE, which cancels out PE reductions induced by the nonSM effect, resulting in an increase in the net PE and the wetting of the dry season (Fig. 3b, e, h). Although the magnitude of SM drying is similar in both seasons (Fig. 2g, h), the negative SM feedback on PE is much weaker in the wet season than the dry season (Fig. 3g, h), contributing to the WDDW pattern and reduced seasonality of PE over 18% of land area in CMIP6 coupled simulations (Fig. 3c). Without the SM effect in LFMIPpdLC, only 9% of land area experiences reduced seasonality of PE (Fig. 3f). In the reduced seasonality regions, 80% of wetseason PE reductions (−0.50 ± 0.24 mm/day) are caused by the nonSM effect, with the remaining 20% from the SM effect. In contrast, the positive SM effect on dryseason PE (0.29 ± 0.11 mm/day) is roughly twice the magnitude of nonSM induced PE reductions (−0.16 ± 0.13 mm/day) (Fig. 3j, k). Overall, the distinct SM effects between wet and dry seasons (−0.39 ± 0.16 mm/day) are responsible for 63% of the reduced annual range of PE, while the nonSM effect (−0.24 ± 0.25 mm/day) accounts for the remaining 37% (Fig. 3l).
The above SM effect on seasonal PE changes highlights the seasonally varying nature of SM(PE) feedbacks, but also includes the effect of changes in SM climatology (i.e., mean seasonal cycle of SM, Fig. 2g–i). To directly compare the SM(PE) feedbacks between dry and wet seasons, we apply an empirical statistical method (Methods) to CMIP6 models and observationally constrained reanalysis products (ModernEra Retrospective analysis for Research and Applications (MERRA2)^{28} and European Center for MediumRange Weather Forecasts (ERA5)). The five CMIP6 models and two reanalysis products consistently show strong negative SM(PE) feedbacks over subtropical regions in the dry season, when the positive SM effect on evapotranspiration exceeds that on precipitation (Fig. 4a–c, g–i). However, the negative SM(PE) feedback is weak in the wet season, when SM limitation on evapotranspiration is negligible (Fig. 4d–f, j–l). These results support the conclusion from Fig. 3 of the presence in subtropical regions of a strong effect of SM drying on PE increases in the dry season but a weak SM effect in the wet season. Over the Amazon, the SM(PE) feedbacks are mostly positive (though not statistically significant) in the wet season and negative in the dry season (Fig. 4a, d, g, j), consistent with wetseason PE decreases and dryseason PE increases in response to SM drying (Fig. 3g, h). Based on this empirical assessment, we again conclude that seasonally varying SM effects dominate the reduced seasonality of PE over subtropical regions and the Amazon.
Mechanisms of the SMatmosphere feedbacks in the wet and dry seasons
For further mechanistic insight, we perform an atmospheric moisture budget diagnosis of the SM effect between the wet and dry seasons in CMIP6 and LFMIPpdLC. In the reduced seasonality regions, SM drying exerts seasonally varying effects on the atmospheric moisture budget (Fig. 5b, d, f). Dryseason evapotranspiration reductions are largely offset by increased moisture convergence (or decreased moisture divergence due to reduced supply of water vapor through evapotranspiration), resulting in only small decreases in precipitation (Fig. 5d). In the wet season, the SM effect reduces both evapotranspiration and moisture convergence, and induces large reductions in precipitation (Fig. 5b). However, the SM effect on the atmospheric moisture budget is small over the enhanced seasonality regions (Fig. 5a, c, e), consistent with the weak SM effects on evapotranspiration, precipitation, and PE from reanalysis products (Figs. 3, 4).
At the seasonal scale, the change in atmospheric moisture storage is relatively small; thus PE approximately equals moisture convergence. The SM effects on PE and moisture convergence are therefore spatially congruent (Fig. 3g–i and Supplementary Fig. 3g–i), with spatial correlations greater than 0.85 in both seasons. By decomposing moisture convergence changes into thermodynamic and dynamic terms (Methods), we find the difference in the SM effects on moisture convergence (and PE) between wet and dry seasons is mainly associated with the dynamic effect in the reduced seasonality regions (Fig. 5b, d, f). In the dry season, the drying of land surface reduces evaporative cooling and strongly enhances land warming, thereby reducing air pressure over subtropical land and the Amazon relative to the ocean and many moist land regions (Supplementary Fig. 5b, d, f). The strengthened surface pressure gradient favors lowlevel flow convergence and vertical ascent (represented by negative pressure velocity) throughout the troposphere over subtropical land and the Amazon (Fig. 5h and Supplementary Fig. 6e–h). However, the SM effects on land surface warming and atmospheric vertical motion are weaker in the wet season, with the negative pressure velocity slightly enhanced in the lowlevel troposphere but suppressed in the mid and highlevel troposphere (Fig. 5g and Supplementary Fig. 6a–d). Enhanced/suppressed vertical ascent is associated with enhanced/suppressed net moisture flux from the ocean to land, as inferred by the strong spatial correlation between changes in the negative pressure velocity and changes in moisture convergence (or PE) over land (Supplementary Fig. 6i–l).
Discussion
This study demonstrates a robust pattern of wet seasons getting drier and dry seasons getting wetter, from the perspective of PE, over a significant fraction of subtropical land regions and the Amazon in both CMIP5 and CMIP6 projections. Further analyses of CMIP6 landatmosphere coupling sensitivity experiments provide evidence that the drying of the wet season is mainly caused by anthropogenic climate change, while the wetting of the dry season in terms of increased PE is driven by the negative SM feedback on PE. We have also found that the reduced seasonality of PE is predominantly caused by the seasonally varying SM effects on PE, while anthropogenic climate change reduces PE in both dry and wet seasons, and the net effect on PE seasonality is small over the subtropics and Amazon. The resultant drying of the soil in turn reduces evapotranspiration and recycling of evaporated moisture for subsequent precipitation^{24,25}. Reduced evaporative cooling further amplifies land surface warming, and the associated landocean warming contrast strengthens surface pressure differences between ocean and land, which drives anomalous oceantoland moisture transport and enhances moisture convergence over land. Such an increase in SMinduced moisture convergence offsets the decrease in precipitation driven by reduced moisture recycling of evapotranspiration, resulting in a more muted precipitation response than the evapotranspiration response to SM drying and a negative SM feedback on PE. The SM limitation on evapotranspiration and associated SMatmosphere feedbacks, especially those related to atmospheric dynamics, are strong in the dry season but weak in the wet season, contributing to large PE increases in the dry season and slight PE decreases in the wet season. The seasonally varying nature of SMatmosphere feedbacks therefore leads to reduced seasonality of PE over subtropical regions and the Amazon.
While the SM effect on longterm PE changes in subtropical dry regions has been recognized^{23}, our modeling and empirical assessments further show that the negative SM(PE) feedback occurs mainly in the dry season. The negative SM(PE) feedback caused by the SM regulation of atmospheric dynamics and moisture convergence is also expected from observational evidence that the SM limitation on evapotranspiration is strong in the dry season, when precipitation is relatively low and cannot decrease as much as evapotranspiration could in response to SM drying^{24}. On the other hand, reduced evapotranspiration leads to reduced terrestrial supply of water vapor for moisture divergence and therefore curbs the reduction of PE in the dry season. The projected SM declines combined with the negative SM(PE) feedback explains the positive SM effect on dryseason PE increases isolated using modeling experiments over subtropical regions and the Amazon. While the negative sign of the SM(PE) feedback is supported by both reanalysis products and CMIP6 models, the feedback strength and the magnitude of SMinduced PE changes vary across models and products (Figs. 3, 4), which may result from uncertainties in the representation of SMatmosphere feedbacks.
Seasonal variations in PE are closely related to extreme hydroclimate conditions, such as droughts and floods, which are projected to increase in many regions under climate change^{3,4,29}. Future drought risks and associated carbon loss are likely to be especially strong over the subtropics and the Amazon^{29,30}. Increased dryseason PE due to SM feedbacks may somewhat attenuate the potential increase of drought risk expected from the thermodynamic hydrological changes. In the absence of the negative SM feedbacks, the projected extreme hydroclimate events would likely become more frequent and more extreme than coupled climate projections^{23}, which would reduce the capacity of terrestrial ecosystems to absorb CO_{2} and mitigate climate change in the future.
Our findings underscore the importance of soil moistureatmosphere feedbacks in modulating seasonal water availability changes. In particular, while warmingdriven oceanic and atmospheric changes point to declining dryseason water availability in the future, soil moisture feedbacks can be viewed as offsetting the decline over subtropical dry regions that would be realized in the absence of such feedbacks. Given the widespread opposite effects of soil moisture and other climate factors on water availability in the dry season (Fig. 3e, h), soil moistureatmosphere feedbacks may alleviate the negative climate change implications for regional water resources management. It is worth noting that while soil moistureatmosphere feedbacks lead to increases in surface water availability in the dry season, reduced soil moisture itself and associated declining evapotranspiration also indicate an overall drying trend of the land surface system driven by climate change over subtropical dry regions and the Amazon. As the soil moisture feedback on surface water availability is stronger in drier conditions, human activities like irrigation may weaken the negative soil moisture feedback and magnify water scarcity in dry regions, but the scale at which this effect might be detected need to be identified. Our study enables a mechanistic understanding of the role of soil moistureatmosphere feedbacks in regulating the seasonal pattern of water availability in coupled climate models, while a more indepth assessment of regional hydrological changes and associated hydroclimate extremes and vegetation activities based on observations and model projections is needed. As subtropical ecosystems are among the most vulnerable to climate change^{31}, it is crucial to continue to assess seasonal variations in the freshwater resources and terrestrial ecosystems and refine projections of the coupled climatehydrologyecosystem to promote effective conservation actions.
Methods
CMIP5 and CMIP6 model simulations
We used output from 35 CMIP5 models (Supplementary Table 1) and 30 CMIP6 models (Supplementary Table 2) covering the historical (1971–2000) and future (2071–2100) periods. The highend forcing scenarios (RCP8.5 in CMIP5 and SSP585 in CMIP6) were used in future simulations. We used these models because they provide the monthly total soil moisture content (“mrso”), precipitation (“pr”), and latent heat flux (“hfls”) as required for our analyses. For each model, one ensemble member was used (see Supplementary Tables 1, 2 for details). We calculated evapotranspiration from latent heat flux which was available in more CMIP6 models than evapotranspiration (“evspsbl”), and obtained precipitation minus evapotranspiration (PE) in each model. In CMIP5 and CMIP6, we defined the wet and dry seasons as the three consecutive months with the highest and lowest climatological mean PE, respectively, using data from the historical simulation (1971–2000) in each model. Multimodel mean seasonal changes in soil moisture, evapotranspiration, precipitation, and PE between the historical and future periods were calculated.
Soil moistureatmosphere feedback experiments
We used a new multimodel experiment from the Land Feedback Model Intercomparison Project with prescribed Land Conditions (LFMIPpdLC), which was designed to assess land surface feedbacks on climate change in CMIP6^{27}. LFMIPpdLC performed transient coupled atmosphereocean simulations driven by the same forcing, including sea surface temperature and sea ice, from corresponding CMIP6 simulations (the historical simulation during 1980–2014 and the SSP585 scenario during 2015–2100), except that soil moisture (SM) was prescribed as the mean seasonal cycle of 1980–2014 from the historical simulation in each model (https://wiki.c2sm.ethz.ch/LS3MIP/Tier1Experiments). LFMIPpdLC is currently available for five models (CESM2, CNRMCM61, ECEarth3, IPSLCM6ALR, MPIESM12LR). Comparing the fully coupled historical and future (SSP585) simulations in CMIP6 (expressed as CMIP6 simulations below) and LFMIPpdLC, we could assess the SM effect on PE in each model.
We used monthly total soil moisture content, precipitation, and latent heat flux from these simulations. We assessed seasonal changes in the variables (SM, precipitation, evapotranspiration, and PE) between 1980–2000 and 2080–2100 in CMIP6 and LFMIPpdLC, the latter only covers 1980–2100. Correspondingly, the wet and dry seasons are defined using historical data from 1980 to 2000 in CMIP6. Although longterm hydrological changes are usually assessed at 30year time scales (Fig. 1), we compared seasonal changes in precipitation, evapotranspiration, and PE between 1971–2000 and 2071–2100 and between 1980–2000 and 2080–2100 in CMIP6, and found the results are very close (Supplementary Fig. 7).
We used pressure level data from CMIP6 and LFMIPpdLC to assess the thermodynamic and dynamic mechanisms of the SM effect on seasonal PE changes. We used monthly specific humidity, eastward and northward wind on pressure levels, and surface pressure to calculate moisture convergence and decompose it into thermodynamic and dynamic terms (see “Moisture budget decomposition”). We also used nearsurface (2 m) air temperature, pressure velocity on pressure levels, and sea level pressure in the five models that participate in both CMIP6 and LFMIPpdLC for the mechanistic analyses.
Reanalysis datasets
To support the modeling feedback analyses, we identified the SMatmosphere feedbacks using two stateoftheart reanalysis products: the ModernEra Retrospective analysis for Research and Applications, version 2 (MERRA2)^{28} and the European Center for MediumRange Weather Forecasts (ERA5). MERRA2 and ERA5 are constrained by in situ and satellite remote sensing observations, and reasonably capture the relationship between SM and PE^{23}. We used monthly rootzone (0–100 cm) SM, precipitation, evapotranspiration from MERRA2 (1980–2019) and ERA5 (1979–2019) to assess the SMatmosphere feedbacks in the wet and dry seasons (see “Soil moistureatmosphere feedbacks”).
Robustness of the seasonal changes in PE
We made use of ensembles of CMIP5 and CMIP6 models to test the robustness of the seasonal changes in PE (Δ(PE)). As the sign and magnitude of Δ(PE) vary across models, we used the multimodel mean Δ(PE) as the best estimate and tested whether the sign of multimodel means is statistically robust. For each grid cell, if the multimodel mean Δ(PE) is positive, we tested the following hypothesis:

1.
The null hypothesis is that the sign of Δ(PE) is random, so the probability of a positive Δ(PE) is 0.5 (\(p\,=\,0.5\));

2.
The alternative hypothesis is \(p \, > \, 0.5\);

3.
To test the null hypothesis, we construct a test statistic: the number of models of all models (\(n\)) that show positive Δ(PE);

4.
As the sign of Δ(PE) is independent across different models, the number of models with positive Δ(PE) follows the binomial distribution, and the probability of positive Δ(PE) simulated in exactly \(m\) models is given by
$${P}_{m}\,=\,\frac{n!}{m!\left(n\,\,m\right)!}{p}^{m}{q}^{n\,\,m},\,p\,=\,0.5,\;q\,=\,0.5$$(1) 
5.
According to the probability density function of binomial distribution, if positive Δ(PE) occurs in 23 of 35 (66%) CMIP5 models or 20 of 30 (67%) CMIP6 models, we can reject the null hypothesis (p value < 0.05), and the positive sign of multimodel mean Δ(PE) is significantly robust.
$${P}_{{CMIP}5}\left(m\,\ge\, 23\right)\,=\,\mathop{\sum }\limits_{m\,=\,23}^{35}\frac{35!}{m!\left(35\,\,m\right)!}{0.5}^{m}\,\times\, {0.5}^{35\,\,m} \, < \, 0.05$$(2)$${P}_{{CMIP}6}\left(m\,\ge\, 20\right)\,=\,\mathop{\sum }\limits_{m\,=\,20}^{30}\frac{30!}{m!\left(30\,\,m\right)!}{0.5}^{m}\,\times\, {0.5}^{30\,\,m} \, < \, 0.05$$(3)
The similar hypothesis testing was also applied for grid cells with negative multimodel mean Δ(PE). In general, if the sign of Δ(PE) is consistent with the sign of multimodel means for more than 65% of the 35 CMIP5 models and of the 30 CMIP6 models, the sign of multimodel mean Δ(PE) is deemed to be statistically significant at the 95% confidence level.
Soil moistureatmosphere feedbacks
We applied an empirical statistical method to assess the SM(PE) feedbacks in the wet and dry seasons using the two reanalysis products and CMIP6 models. This method establishes a multiple linear regression model between PE and onemonth lagged SM to identify the sign and strength of the SM(PE) feedback^{23}. As the SM effect on PE may persist for weeks to months, the regression model between PE and 1month lagged SM therefore can isolate the SM feedback on PE from the direct PE effect on SM^{23}. In the regression model, the multiyear mean seasonal cycles and the linear trends of SM and PE are removed to focus on the feedback of SM variations on PE variations. The regression model also takes into account the priormonth PE to overcome the potential effect of PE autocorrelation.
where the subscript \(d\) indicates that the seasonal cycle and linear trend of the variable are removed, and the indicator \(t\) represents monthly steps in the wet or dry season. The regression coefficient \({n}_{1}\) represents the partial derivative of PE variations to SM variations in the priormonth \(\left(\frac{\partial {\left(PE\right)}_{d}\left(t\right)}{\partial S{M}_{d}\left(t1\right)}\right)\), and was used to capture the SM feedback on PE in the wet and dry seasons.
We identified the SM feedback on PE as the standardized \({n}_{1}\), or sensitivity coefficient for SM\(\to\)(PE), which corresponds to standardized \({\left(PE\right)}_{d}\) and \(S{M}_{d}\) of zero mean and unit variance in the wet or dry season. In this way, we could better compare the SM(PE) feedback between wet and dry seasons and across different regions/datasets/models. Alternatively, we standardized the entire time series of \({\left(PE\right)}_{d}\) and \(S{M}_{d}\) to zero mean and unit variance and then applied the regression model to obtain the regression coefficient \({n}_{1}\). The spatial patterns of the identified sensitivity coefficients for SM\(\to\)(PE) from the two standardization methods are identical (Fig. 4 and Supplementary Fig. 8).
We used a bootstrap test to determine the significance of the sensitivity coefficients in case the identified SM(PE) feedback may be sensitive to natural variability. In the bootstrap analysis, the time series of the variables are randomly resampled to perform the multiple linear regression and obtain the 95% confidence intervals of the sensitivity coefficients for the wet and dry seasons. According to the bootstrap confidence intervals, the sensitivity coefficients are deemed statistically significant if the 95% confidence intervals do not overlap with zero. The multiple linear regression method and the bootstrap test were also used to obtain the sensitivity coefficients for the SM effects on evapotranspiration and precipitation.
Moisture budget decomposition
According to atmospheric moisture budget, PE equals to moisture convergence (MC), which is defined as the negative divergence of vertically integrated moisture flux over the pressure (\(p\)) from the top of the atmosphere (\(p\,=\,0\)) to the surface (\(p\,=\,{p}_{s}\)).
where \({\rho }_{w}\) is the density of water, \(g\) is the acceleration due to gravity, \(\nabla\) is the horizontal divergence operator, \({{{{{\boldsymbol{u}}}}}}\) is the horizontal vector wind, and \(q\) is specific humidity.
As we focused on climatological seasonal changes of PE, we used monthly data to approximately calculate MC differences between the historical and future periods.
where overbars indicate monthly mean values. The change in MC (\(\triangle {MC}\)) can be decomposed as
where the subscript 0 represents the historical period, and the delta operator represents changes from the historical to future periods. On the right side of Eq. (8), \(\triangle {MC}\) is decomposed into a thermodynamic term due to specific humidity changes (\(\frac{1}{{\rho }_{w}g}\nabla \,\cdot\, {\int }_{0}^{{p}_{s}}\left({\bar{{{{{{\boldsymbol{u}}}}}}}}_{0}\,\cdot\, \triangle \bar{q}\right){dp}\)), a dynamic term due to horizontal wind changes (\(\frac{1}{{\rho }_{w}g}\nabla \,\cdot\, {\int }_{0}^{{p}_{s}}\left({\bar{q}}_{0}\,\cdot\, \triangle \bar{{{{{{\boldsymbol{u}}}}}}}\right){dp}\)), and a nonlinear term due to the product of specific humidity and wind changes (\(\frac{1}{{\rho }_{w}g}\nabla \,\cdot\, {\int }_{0}^{{p}_{s}}\left(\triangle \bar{{{{{{\boldsymbol{u}}}}}}}\,\cdot\, \triangle \bar{q}\right){dp}\))^{15,23,32}.
We used monthly specific humidity and zonal and meridional wind velocity on pressure levels, and surface pressure from CMIP6 and LFMIPpdLC to calculate the change in MC between the historical (1980–2000) and future (2080–2100) periods, and its three components (Eq. 8) in the wet and dry seasons. The SM effects on MC changes and the thermodynamic, dynamic, and nonlinear terms were calculated as their differences between CMIP6 and LFMIPpdLC. As the dynamic term (\(\frac{1}{{\rho }_{w}g}\nabla \,\cdot\, {\int }_{0}^{{p}_{s}}\left({\bar{q}}_{0}\,\cdot\, \triangle \bar{{{{{{\boldsymbol{u}}}}}}}\right){dp}\)) is approximately equal to \(\frac{1}{{\rho }_{w}g}{\int }_{0}^{{p}_{s}}\left(\triangle \bar{\omega }\partial q/\partial p\right){dp}\), where \(\omega\) is the pressure vertical velocity, according to the mass continuity equation^{33}, we identify the dynamic effect by analyzing the SM effects on both the horizontal wind velocity and the pressure vertical velocity throughout the troposphere.
Data availability
All data used in this study are available online. The CMIP5 model simulations were downloaded from https://esgfnode.llnl.gov/search/cmip5/, and the CMIP6 (including LFMIPpdLC) model simulations are available from https://esgfnode.llnl.gov/search/cmip6/. The ERA5 reanalysis data are from https://www.ecmwf.int/en/forecasts/datasets/archivedatasets/reanalysisdatasets/era5. The MERRA2 reanalysis data are from https://gmao.gsfc.nasa.gov/reanalysis/MERRA2/data_access/.
Code availability
The R code used for modeling and reanalysis data analyses is publicly available (https://doi.org/10.5281/zenodo.6802965).
References
Rockström, J. et al. Future water availability for global food production: the potential of green water for increasing resilience to global change. Water Resour. Res. 45, W00A12 (2019).
Seddon, A. W. R., MaciasFauria, M., Long, P. R., Benz, D. & Willis, K. J. Sensitivity of global terrestrial ecosystems to climate variability. Nature 531, 229–232 (2016).
Trenberth, K. E. et al. Global warming and changes in drought. Nat. Clim. Change 4, 17–22 (2014).
Hirabayashi, Y. et al. Global flood risk under climate change. Nat. Clim. Change 3, 816–821 (2013).
Held, I. M. & Soden, B. J. Robust responses of the hydrological cycle to global warming. J. Clim. 19, 5686–5699 (2006).
Chou, C. & Lan, C.W. Changes in the annual range of precipitation under global warming. J. Clim. 25, 222–235 (2012).
Greve, P. & Seneviratne, S. I. Assessment of future changes in water availability and aridity. Geophys. Res. Lett. 42, 5493–5499 (2015).
Allan, R. P. et al. Advances in understanding large‐scale responses of the water cycle to climate change. Ann. N.Y. Acad. Sci. 1472, 49–75 (2020).
Chou, C. et al. Increase in the range between wet and dry season precipitation. Nat. Geosci. 6, 263–267 (2013).
Konapala, G., Mishra, A. K., Wada, Y. & Mann, M. E. Climate change will affect global water availability through compounding changes in seasonal precipitation and evaporation. Nat. Commun. 11, 3044 (2020).
Kumar, S., Lawrence, D. M., Dirmeyer, P. A. & Sheffield, J. Less reliable water availability in the 21st century climate projections. Earths Fut. 2, 152–160 (2014).
Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An overview of CMIP5 and the experiment design. Bull. Am. Meteor. Soc. 93, 485–498 (2012).
Eyring, V. et al. Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev. 9, 1937–1958 (2016).
Kumar, S., Allan, R. P., Zwiers, F., Lawrence, D. M. & Dirmeyer, P. A. Revisiting trends in wetness and dryness in the presence of internal climate variability and water limitations over land. Geophys. Res. Lett. 42, 10867–10875 (2015).
Seager, R., Naik, N. & Vecchi, G. A. Thermodynamic and dynamic mechanisms for largescale changes in the hydrological cycle in response to global warming. J. Clim. 23, 4651–4668 (2010).
Roderick, M., Sun, F., Lim, W. H. & Farquhar, G. A general framework for understanding the response of the water cycle to global warming over land and ocean. Hydrol. Earth System Sci. 18, 1575–1589 (2014).
Byrne, M. P. & O’Gorman, P. A. The response of precipitation minus evapotranspiration to climate warming: why the “WetGetWetter, DryGetDrier” scaling does not hold over land. J. Clim. 28, 8078–8092 (2015).
Chou, C., Neelin, J. D., Chen, C.A. & Tu, J.Y. Evaluating the “RichGetRicher” mechanism in tropical precipitation change under global warming. J. Clim. 22, 1982–2005 (2009).
Karnauskas, K. B. & Ummenhofer, C. C. On the dynamics of the Hadley circulation and subtropical drying. Clim. Dyn. 42, 2259–2269 (2014).
Lu, J., Vecchi, G. A. & Reichler, T. Expansion of the Hadley cell under global warming. Geophys. Res. Lett. 34, L06805 (2007).
Grise, K. M. & Davis, S. M. Hadley cell expansion in CMIP6 models. Atmos. Chem. Phys. 20, 5249–5268 (2020).
Lau, W. K. M. & Kim, K.M. Robust Hadley circulation changes and increasing global dryness due to CO_{2} warming from CMIP5 model projections. Proc. Natl Acad. Sci. USA 112, 3630–3635 (2015).
Zhou, S. et al. Soil moisture–atmosphere feedbacks mitigate declining water availability in drylands. Nat. Clim. Chang. 11, 38–44 (2021).
Seneviratne, S. I. et al. Investigating soil moisture–climate interactions in a changing climate: a review. EarthSci. Rev. 99, 125–161 (2010).
Dirmeyer, P. A., Schlosser, C. A. & Brubaker, K. L. Precipitation, recycling, and land memory: an integrated analysis. J. Hydrometeor. 10, 278–288 (2009).
Koster, R. D. et al. Regions of strong coupling between soil moisture and precipitation. Science 305, 1138–1140 (2004).
van den Hurk, B. et al. LS3MIP (v1.0) contribution to CMIP6: the land surface, snow and soilmoisture model intercomparison project—aims, setup and expected outcome. Geosci. Model Dev. 9, 2809–2832 (2016).
Gelaro, R. et al. The modernera retrospective analysis for research and applications, version 2 (MERRA2). J. Clim. 30, 5419–5454 (2017).
Zhou, S., Zhang, Y., Williams, A. P. & Gentine, P. Projected increases in intensity, frequency, and terrestrial carbon costs of compound drought and aridity events. Sci. Adv. 5, eaau5740 (2019).
Parsons, L. A. Implications of CMIP6 projected drying trends for 21st century amazonian drought risk. Earths Fut. 8, e2020EF001608 (2020).
Burrell, A. L., Evans, J. P. & De Kauwe, M. G. Anthropogenic climate change has driven over 5 million km2 of drylands towards desertification. Nat. Commun. 11, 3853 (2020).
Zhou, S. et al. Large divergence in tropical hydrological projections caused by model spread in vegetation responses to elevated CO_{2}. Earths Fut. 10, e2021EF002457 (2022).
Bony, S. et al. Robust direct effect of carbon dioxide on tropical circulation and regional precipitation. Nat. Geosci. 6, 447–451 (2013).
Acknowledgements
We acknowledge the World Climate Research Program’s Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Supplementary Tables 1, 2 of this paper) and the LFMIP group for producing and making available their model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. S.Z. acknowledges the NSFC Excellent Young Scientists Fund (Overseas), the Second Tibetan Plateau Scientific Expedition and Research Program (2019QZKK0405), and the Fundamental Research Funds for the Central Universities. P.G. acknowledges support from NASA ROSES Terrestrial hydrology (NNH17ZDA00INTHP) and NOAA MAPP NA17OAR4310127. A.P.W. acknowledges support from the NASA Modeling, Analysis, and Prediction (MAP) program (NASA 80NSSC17K0265). T.F.K. acknowledges support from the RUBISCO SFA, which is sponsored by the Regional and Global Model Analysis (RGMA) Program in the Climate and Environmental Sciences Division (CESD) of the Office of Biological and Environmental Research (BER) in the U.S. Department of Energy Office of Science, and additional support from a DOE Early Career Research Program award (DESC0021023).
Author information
Authors and Affiliations
Contributions
S.Z. conceived and designed the study. S.Z. processed model simulations and reanalysis data. S.Z., A.P.W., B.R.L., K.L.F., T.F.K., Y.Z., and P.G. contributed to data analysis and interpretation. S.Z. wrote the paper. A.P.W., B.R.L., K.L.F., T.F.K., Y.Z., and P.G. edited the paper.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Communications thanks Jiangfeng Wei and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zhou, S., Williams, A.P., Lintner, B.R. et al. Diminishing seasonality of subtropical water availability in a warmer world dominated by soil moisture–atmosphere feedbacks. Nat Commun 13, 5756 (2022). https://doi.org/10.1038/s41467022334739
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467022334739
This article is cited by

Soil moisture–atmosphere coupling accelerates global warming
Nature Communications (2023)

Projected increase in global runoff dominated by land surface changes
Nature Climate Change (2023)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.