Attribution of air temperature and precipitation to the future global drought events

Quantifying the contributions of air temperature and precipitation changes to drought events can inform decision-makers to mitigate the impact of droughts while existing studies focused mainly on long-term dryness trends. Based on the latest Coupled Model Intercomparison Project (CMIP6), we analyzed the changes in drought events and separated the contributions of air temperature and precipitation to the risk of future drought events. We found that drought frequency, duration, severity, and month will increase in the future (56.4%, 63.5%, 82.9%, and 58.2% of the global land area in SSP245, and 58.1%, 67.7%, 85.8%, and 60.5% of the global land area in SSP585, respectively). The intermediate scenario has a similar pattern to the most extreme scenario, but low emission was found to mitigate drought risk. Globally, we found that air temperature will have a greater impact than precipitation on intensifying drought. Increasing precipitation will mitigate drought risks in some middle and high northern latitudes, whilst the trend in increasing air temperature will counter the effects of precipitation and increase the impact of droughts. Our study improves the understanding of the dynamics of future devastating drought events and informs the decision-making of stakeholders.


Introduction
Drought reduces gross primary production, incurs vegetation mortality, and therefore affects the terrestrial carbon cycle Green et al 2019, Van der Molen et al (2011), resulting in multifaceted challenges in drinking water security, agriculture, and the economy (Park et al 2016). Devastating drought events, therefore, have attracted widespread attention and discussion across Australia (Araujo et al 2022), the Horn of Africa (Funk et al 2019), and southwest China (Song et al 2019 amongst others. A critical goal within the scientific community is to advance knowledge of drought events to inform decision-making regarding drought events. Unravelling the drivers of drought events and how these drivers affect droughts under ongoing climate warming is therefore a compelling scientific question of widespread importance.
The frequency, duration, and intensity of global drought events are expected to increase under future predicted warming (Dai et al 2018, Zhang et al 2019, Christian et al 2021. The main cause of the drought was often assumed to be the lack of precipitation (Ukkola et al 2020, Mckee et al (1993)). It is widely reported that precipitation will increase in the central Sahel, eastern Russia, northern China, and northern high latitudes (Ukkola et al 2020, Zhao andDai 2021). In theory, the increasing precipitation should ease the droughts in these regions. Precipitation, however, does not represent the only climatic factor controlling drought. Temperature is also a key determinant, given that climate warming exponentially stimulates potential evapotranspiration (PET) (Mcvicar et al (2012) Thornthwaite 1948, Allen et al 1998. It, therefore, seems to be necessary to consider the effect of climatic warming on drought events (Vicente-Serrano et al 2010, Frierson and Scheff 2014, Zhao and Dai 2015, Zhao and Dai 2021. The precipitation and temperature variability are the main factors that directly affect the change trends of dryness/wetness (Alexander et al 2006, Hance et al 2007, Liu et al 2009, Wu and Chen 2019. The idea of evaluating the relative contributions from changes in moisture supply (precipitation) versus evaporative demand (PET), mainly depending on temperature (Arora and Boer 2001, Sun et al 2016, Hui-Mean et al 2018, to drought index is not new. For instance, Feng and Fu (2013) examined the contributions of precipitation and PET to projected changes in the P/PET ratio. By using different drought indexes, they found a widespread dryness trend in the twenty-first century because of increased PET and reduced precipitation over subtropical areas (Cook et al 2014, Zhao and Dai 2015, Cook et al 2020, Zhao and Dai 2021. A key conceptual issue is to distinguish drought events and long-term aridity trends (Sherwood and Fu 2014). Existing studies focused mainly on long-term aridity trends, rather than drought events.
In this study, we focus on agricultural drought events, which often defined as soil water deficiency caused by long-term meteorological drought (Vicente-Serrano et al 2010, Beguería et al 2014, Sherwood andFu 2014). Unravelling the drivers of drought development not only improves our understanding about future drought dynamics but also informs the decision-making of stakeholders to reduce the losses caused by droughts. To date, few works have quantitatively separated the contributions of precipitation and PET to future drought event changes (Zhang et al 2019. These progress in drought events attribution researches are via past and regional analyses, a key scientific question remains: To what extent will climate change affect global drought events in the future? To address this question, we characterize the spatiotemporal dynamics of future agricultural drought events and separate the contributions of air temperature and precipitation on these dynamics.

Data
We obtained monthly simulations of total precipitation, surface air minimum temperature (T min ) and maximum temperature (T max ) from the Coupled Model Intercomparison Project phases 6 (CMIP6). Fifteen climate models were used and averaged to eliminate the uncertainty of artificial factors and model selection (table 1). To compare the difference between historical and future global drought events, we used the historical experiment (1850-2014) and two future scenarios (2015-2100) including Shared Socioeconomic Pathways (SSP) 2-4.5 and 5-8.5. SSP 245 represents an intermediate 'middle of the road' future climate scenario and SSP 585 is a high emissions 'fossil-fueled development' future scenario O'neill et al (2016). We re-gridded all data into 2 × 2°through the bilinear interpolation.
We utilized the 2020 MODIS Land Cover Type Yearly Climate Modeling Grid (MCD12C1.006 product) to mask the desert areas, since it is difficult to assess droughts in places where zero-precipitation amounts are the norm. MCD12C1 dataset was acquired from https://lpdaac.usgs.gov/products, which offers aggregated land cover data at a resolution of 0.05°.

Model
Resolution ( Using percentile thresholds to determine drought periods is a common approach in hydrology and climatology (Trenberth et al 2014, Ukkola et al 2020. One advantage of this method is that it does not rely on any assumptions about the data distribution. This makes it more robust to outliers or non-normal data distributions. To better compare with Ukkola's work, precipitation minus potential evapotranspiration (P − PET) was used in our study to quantify the characteristics of drought.
PET was calculated by precipitation, extraterrestrial radiation, T min, and T max based on the modified Hargreaves equation. The modified Hargreaves equation considers multiple relevant variables affecting PET, which is more reliable than the single variable method (Xu et al 2021a). The modified Hargreaves equation has shown similar results to the Penman-Monteith method (Droogers and Allen 2002, Beguería et al 2014, de Streel et al 2022. According to Droogers and Allen (2002), PET is calculated as: where R a is the extraterrestrial radiation (can be estimated from the latitude and the month of the year); T avg is the monthly average temperature; TD is the temperature range, calculated as the difference between daily maximum temperature and minimum temperature; P is precipitation in mm per month. To consider the effects of temperature changes individually, we fixed the change in precipitation at 0 and calculated R a using the latitude and the month of the year in this study. We converted the monthly P − PET time series into 3-month accumulations to smooth out short-term variations. This is analogous to calculating the Standardized Precipitation Evapotranspiration Index (SPEI) at the time-scale of 3 months and reflects changes in seasonal droughts, which can satisfactorily capture the vegetation response to drought and the characteristics of short-term variations in soil moisture (Lloyd-Hughes 2012, Xu et al 2015).
Quantifying drought in percentiles, rather than mean values, provides a better description of drought characterization. Percentile thresholds involve no assumptions about the data distribution. In this study, the 10th percentile thresholds (x 10m ; mm) were defined separately for each month from the 3-month accumulated P − PET to account for seasonality, then any month below these thresholds is classified as drought. The 10th percentile corresponds approximately to severe drought. The monthly 10th percentile thresholds were derived from the period 1980−2014. Subsequently, we calculated three common drought characteristics (duration, frequency and severity), and drought month. Drought month would provide a more realistic and comprehensive estimate of the temporal exposure to drought in different regions than frequency or duration alone. Drought frequency, duration and severity is focus more on characteristics of drought events. Drought events were first identified by the start and end months of the drought, and then their duration and severity were then determined. The number of continuous drought months is defined as the duration of a drought event. Severity (mm month −1 ) was calculated by averaging the difference between the drought threshold and the monthly 3-month accumulation P − PET value (x m ; mm) over all months during the drought event. Frequency refers to the number of drought events over a period. Drought months is the product of duration and frequency.

Attribution of the drought event
We designed some simulated experiments to separate the respective contributions of the air temperature and precipitation for future P-PET (table 2). We first calculated the P 2031-2060 − PET 2031-2060 (S1) during 2031 and 2060 in the two future scenarios (SSP 245 and SSP 585), and defined the S1 as the baseline case. Secondly, the P 2031-2060 − PET 1981-2010 (S2) and P 1981-2010 − PET 2031-2060 (S3) were obtained by replacing with past (1981-2010) temperature and precipitation, respectively. Compared with S1, S2 represents the P − PET with past temperature but with future precipitation, while S3 represents P − PET with past precipitation and future air temperature. Thirdly, we calculated drought characteristics in S1, S2, and S3. Finally, the contribution of the air temperature to the future drought was the difference in drought characteristics between S1 and S2 (S1-S2),  PET 1981S2 2031-20601981P 1981− PET 2031-2060S3 19812031-2060 and the contribution of precipitation to the future drought was defined as the difference in drought characteristics between S1 and S3 (S1-S3). In addition, we defined a positive contribution as increasing drought event characteristics and a negative contribution as decreasing drought event characteristics.

Projected changes in drought characteristics
The long-term series of percentage of annual average drought area (PAADA) (drought cells divide total cells) from 1981 to 2060 is presented in figure 1. The annual average drought area in the SSP 585 scenario was temporally consistent with the SSP 245 scenario before 2030, whereas significantly higher than the SSP 245 scenario after 2030. In addition, the PAADA showed significant increase trends (0.024 decade −1 and 0.032 decade −1 for SSP 245 and SSP 585, respectively) in the future while the historic trend showed no significant trend. These indicate that the global drought is increasing and will intensify in the future. Figure 2 shows the drought distribution in the past and two future scenarios. As for the historical period (figures 2(a)-(d)), the global drought frequency, duration and month were spatially homogeneous (figures 2(a) (b), and (d)). But the tropical and subtropical regions experienced more severe droughts ( figure 2(c)). In 2031-2060, most tropical and subtropical regions, including Amazon, Australia, Central America, Central Asia, Chile, the Mediterranean, and southern Africa, are projected to experience longer, more frequent, and stronger droughts, especially in the SSP 585 scenario (figures 2(i)-(l)). Although drought frequency has only increased by less than one time, it is noteworthy that the drought months have generally increased by more than two times, and even increased by more than fivefold in the most severe regions.
To directly compare the difference between the past and future cases, we subtracted historical drought features from future ones ( figure 3). Firstly, the strongest frequency increases are expected to occur in most  , duration (the second column), severity (the third column), and month (the fourth column) were employed to quantify the drought event. The historical (the first row) and future cases (the second and third rows for the SSP245 and 585 scenarios, respectively) were calculated with the multi-model averaged precipitation and potential evapotranspiration in CMIP6 simulations. The inset in each subplot represents the corresponding probability distribution. tropical and subtropical regions, with increases in drought frequency from ∼20 during the historical period to ∼40 in the future (figures 3(a) and (e)). By contrast, the drought frequency is projected to decrease by up to 20 in northern high latitudes. As for duration and severity, Amazon, Central Asia, Chile, the Mediterranean, southern Africa, and western Australia showed larger changes than other places (figures 3(b), (c), (f), and (g)). Their duration extended from ∼2 months during the historical period to ∼6 months in the SSP 245 and over 6 months in the SSP 585. Drought month, represents total number of drought months over 30 years, showed similar and more continuous pattern with drought frequency and duration. Although the frequency of droughts has increased by less than 100%, it is noteworthy that the number of drought months has generally increased by more than two times, and in the most severely affected areas, it has even increased by more than five times. This means that the risk of prolonged drought events in hotspots is greatly increased.
There are large global land areas of increased drought frequency (SSP 245 = 56.4%, SSP 585 = 58.1%), duration (SSP 245 = 63.5%, SSP 585 = 67.7%), severity (SSP 245 = 82.9%, SSP 585 = 85.8%), and month (SSP 245 = 58.2%, SSP 585 = 60.5%). This indicates that many regions, including almost all low and middle latitudes, will experience longer and more serious drought events, while droughts in the northern high latitudes will be eased. The overall pattern of drought frequency difference was similar between SSP 245 and SSP 585, but drought duration and severity in SSP 585 showed a larger difference than those in SSP 245.

Contributions of air temperature and precipitation
The temporal curves of PAADA, in different simulations, from 2031 to 2060 are presented in figure 4. The PAADA in P 1981-2010 − PET 2031-2060 (S3) is closer to the PAADA in P 2031-2060 − PET 2031-2060 (S1) than PAADA in P 2031-2060 − PET 1981-2010 (S2) (figures 4(a) and (b)), implying changes in air temperature will have a greater impact on future drought events than precipitation. Close inspection revealed air temperature and precipitation would exert contrasting influences on future drought (negative and positive, respectively), and precipitation and air temperature in SSP 585 contributed more than those in SSP 245 (figure 4(c)).
In SSP 245 and SSP 585, the PAADA in S1 (SSP 245: 0.026 decade −1 , SSP 585: 0.035 decade −1 ) and PAADA in S2 (SSP 245: 0.018 decade −1 , SSP 585: 0.023 decade −1 ) showed significant increase trends. This indicates that the contribution of air temperature for PAADA increases in the future. Specifically, contribution of air temperature for PAADA will increase 0.009 per decade in SSP 245 and 0.012 per decade in SSP 585. While the mitigation effect of precipitation for PAADA will be decreased in the future (0.023 decade −1 in SSP 245 and 0.014 decade −1 in SSP 585).
We also analyzed the distribution of the contributions of air temperature and precipitation for drought frequency, duration, severity, and month. In terms of the drought frequency (figures 5(a)-(d)), air temperature strongly increases the future drought frequency (96.4% and 92.2% of the global land area in SSP 245 and SSP 585, respectively) (figures 5(a) and (b)). While the precipitation decreases drought frequency in Canada, northern China, Russia, and other high latitude regions (in total, 75.1% and 75.9% of the global land area in SSP 245 and 585, respectively) (figures 5(b) and (f)), and increases drought events in Amazon, Australia, Central America, Central Asia, Chile, northwest China, and southern Africa, in line with previous studies (figures 5(c) and (d)).
Air temperature increases generally extend the future drought duration (91.4% and 92.4% of the global land area in SSP 245 and SSP 585, respectively). Precipitation will extend the future drought duration in northern high latitudes while shortening the future drought duration in most low and middle latitudes (62.7% and 64.8% of the global land area in SSP 245 and SSP 585, respectively). In addition, air temperature increases strongly   P 2031-2060 and P 1981-2010 , respectively) and PET (PET 2031-2060and PET 1981 , respectively) were all used to calculate P − PET. The contribution of precipitation to the future drought can be represented by the difference between P 2031-2060 − PET 2031-2060and P 1981− PET 2031-2060 . The contribution of air temperature was calculated similarly but with PET 1981-2010 , which was calculated with historical air temperature. Figure 5. Contributions of air temperature (the first and second columns) and precipitation (the third and fourth columns) to future drought frequency (the first row), duration (the second row), severity (the third row), and month (the fourth row). SSP 245 was used for the first and third columns, and SSP 585 was used for the second and fourth columns. The inset in each subplot represents the corresponding probability distribution. extend drought duration over 2 months in 16.6% of the global land area (figures 5(e)) and 23.5% (figure 5(f)) in Central Asia, Chile, the Mediterranean, southern Africa, and western Australia. These results indicate that precipitation has less influence than the air temperature. The contributions of air temperature and precipitation to drought severity (figures 5(i)-(l)) follow a very similar pattern. (figures 5(e)-(h)).
Drought month provide a more realistic and comprehensive estimate of the temporal exposure to drought in different regions than frequency or duration alone. Although the direction of change is similar, drought month showed much more continuous pattern than drought frequency and duration. Changes in air temperature nearly leads to increase in global drought month (98.5% and 97.4% of the global land area in SSP 245 and SSP 585, respectively), which is more than that in drought frequency and duration. Overall, drought frequency, duration, and severity were increasing globally by air temperature and decreasing by precipitation except in Australia, Chile, the Mediterranean, southern Africa, and the western United States.

Discussion
Long-term dryness trend and its attribution are an active research area (Cook et al 2014, Frierson and Scheff 2014, Vicente-Serrano et al 2015, Zhao and Dai 2015. The dryness trend, however, does not physically the same as drought events. A drought event is a transient regional phenomenon often defined as deviations from assumed known local climate norms (Sherwood and Fu 2014). Drought events were affected by both the change in mean-state (long-term dryness) and variability (Trenberth et (2011)), resulting in multifaceted challenges in drinking water security, agriculture, and the economy (Park et al 2016). Therefore, Ukkola et al. (2020) based on precipitation investigated changes of drought events globally. However, evaporative demand affected by ongoing air temperature can also play a critical role (Beguería et al 2014, Cook et al 2014, Zhao and Dai 2021. To date, few works have quantitatively separated the contributions of precipitation and PET to future drought event changes (Zhang et al 2019).
This study evaluated the future PAADA and drought event changes including drought frequency, duration, severity and month, and the respective contributions of air temperature and precipitation to drought event changes. If we only consider the contribution from precipitation, drought events would intensify in low and middle latitudes such as Amazon, Chile, the Mediterranean, and southern Africa, while easing in high latitudes (figures 5(c)-(d), (g)-(h), (k)-(l) and (o)-(p)), consistent with Ukkola et al. (2020). Generally, the increasing precipitation would ease the drought events on the global scale (Cook et al 2020, Wu et al 2022. However, the ongoing warming can cause higher PET, which would compensate or even reverse the mediation of drought by increasing precipitation, examples can be observed in central China and central and western United States. By separating the contributions of air temperature and precipitation to the risk of future drought events, our results advance the current understanding of future drought events and their drivers.
There is an unexpected but plausible phenomenon. The patterns of the contribution of air temperature to drought events are not identical to that of the overall air temperature trend. The air temperature will increase substantially at high latitudes, yet moderately at low latitudes under climate change (IPCC 2013). But changes in droughts contributed by air temperature in middle and low latitude regions are stronger than in high latitude regions (figures 5(a)-(b), (e), (f), (i)-(j) and (m)-(n)), which is similar to LIU et al (2020), Vicente-Serrano et al (2015), Cook et al (2014). This is because the relationship between PET and air temperature is non-linear, and the sensitivity of PET to air temperature is slight in cold conditions, e.g., in high latitude regions (Thornthwaite 1948, Hargreaves and engineering, 1994, Droogers and Allen 2002. This means that the droughts most affected by temperature were mainly distributed in the low and middle latitudes, instead of the high latitudes. The CMIP6 model provides more reliable estimates compared with the CMIP5 model for future climate scenarios  (Canturk and Kulaç 2021), and results in this study showed that SSP 245 has a consistent pattern with SSP 585. However, drought events under the SSP 585 scenario showed longer durations with stronger severity than drought events under the SSP 245 scenario (figure 4), indicating that future changes in drought duration and severity will be affected by the intensity of greenhouse gas emissions. Therefore, lower greenhouse gas emissions would mitigate future drought events.
Given the critical role of diverse meteorological and climatological features in the drought development, it is imperative for future studies to employ multiple drought metrics, such as runoff and soil moisture, to comprehensively understand the drivers of drought intensification across the globe. Reasonable assumptions can improve the robustness of the method, making it more resistant to outliers (Guttman 1999, Pieper et al 2020. Additionally, using a distribution function other than the normal distribution expands the method's applicability to non-normal data distributions. In essence, fitting an appropriate distribution function to the time series of precipitation (and water balance) improves the characterization of extreme values in their respective distributions (Hayes et al 2011). Statistically improving the description of the tails of distributions is particularly important for the proper statistical treatment of extreme events. Besides, adhering to the drought index recommendation by the World Meteorological Organization can enhance comparability.
Moreover, there is a need for research that disentangles the complex interactions between drought and socioeconomic impacts. The results of this study can serve as a reference framework for quantifying the sensitivity of droughts. While the drivers of predicted drought are presented, further work is required to advance and validate the risk of drought development.

Conclusions
We compared drought events in the historical period (1981-2010) with two future (2031-2060) climate scenarios (SSP 245 and SSP 585) and attributed the contributions of future air temperature and precipitation to future drought events. Monthly climate data from fifteen CMIP 6 monthly models were applied to model the P − PET across the globe. The results can be summarized as follows: (1)Drought frequency, duration, severity, and month will increase in most regions in the future (56.4%, 63.5%, 82.9%, and 58.2% of the global land area in SSP245, and 58.1%, 67.7%, 85.8%, and 60.5% of the global land area in SSP585, respectively). Importantly, drought events in the SSP 585 scenario showed longer durations and stronger severity than that in SSP 245, therefore, future changes in drought could be mitigated through lower greenhouse gas emissions.
(2)Future precipitation will ease the drought events in Canada, China, and northern high-latitude regions, while intensifying drought events in the Amazon, Australia, Central America, Chile, the Mediterranean, and southern Africa. Air temperature trends would exacerbate these drought events. Low and middle latitudes where precipitation is declining are more vulnerable to drought than areas with increased precipitation.
Overall, we provided a comprehensive and systematic way to quantify the risks and impacts of future drought events. As important drivers of drought, our results highlight how precipitation and air temperature will change drought characteristics. This offers a basis for strategies to cope with the risk of intensifying drought, and enhance water and food security in the future, despite uncertainty in precipitation and air temperature.