Projecting global soil moisture droughts under climate change: characteristics, agricultural exposure, and adaptation insights

In recent years, the intensification of global warming has led to an increase in the frequency and severity/intensity of extreme prolonged drought events^1,2,3, such as: the Millennium Drought in Australia in 1997–2010⁴; the 2010 droughts in the Amazon⁵, southwestern China⁶, and Russia⁷; the 2011 droughts in Texas-Mexico⁸ and East Africa⁹; and the USA summer droughts of 2012 and 2016^10,11. These events have had devastating consequences for agriculture¹², economies and societies^13,14, exacerbating food insecurity^15,16,17, water scarcity¹⁸, and socio-economic vulnerabilities^19,20,21. Addressing drought impacts is critical, not only for climate resilience, but also for advancing sustainable development goals (SDGs), particularly those related to^22,23: Zero Hunger (SDG 2), Climate Action (SDG 13) and Life on Land (SDG 15). A comprehensive understanding of future drought evolution is essential to developing sustainable water management strategies, mitigating risks, and fostering adaptive resilience in vulnerable regions^24,25.

A literature review on drought types highlights the complexity of drought classification, typically divided into meteorological, hydrological, agricultural, and socio-economic droughts²⁶. Meteorological drought refers to prolonged periods of below-average precipitation²⁷, while hydrological drought is characterised by reduced water availability in surface and groundwater resources²⁸. Agricultural drought, particularly soil moisture drought, occurs when insufficient soil moisture limits crop growth and productivity²⁹. Studies on soil moisture droughts frequently utilise indices such as the Standardised Soil Moisture Index (SSMI) to assess drought severity/intensity, frequency and duration³⁰.

Given the critical role of soil moisture in agricultural systems, numerous studies have explored its variability and long-term trends, aiming to improve drought monitoring and management strategies³¹. Extensive research has also explored historical and projected future drought patterns using observational datasets and climate model simulations^32,33. Long-term trends in drought conditions under climate change have been assessed through various drought indices³⁴. Many studies have analysed droughts by examining their magnitude, duration, frequency, and spatial patterns^35,36,37. Additionally, uncertainties in drought assessment, such as data selection, hydrological component formulations, and the choice of drought indices, have been widely discussed^38,39,40,41.

While these studies have provided valuable insights into drought trends and characteristics, they have often been case-specific⁴² and have not captured the long-term implications for sustainable management and climate resilience on a global scale. A significant issue is the reliance on physical drought indices that fail to account for the dynamic nature of regional vulnerabilities, leading to a disconnect between hazard forecasts and actual impacts⁴³. This highlights the necessity of developing methods that focus on impacts rather than on merely forecasting hazards⁴⁴. Additionally, significant gaps remain in quantifying the future impacts of droughts on a global scale and in understanding how emission pathways shape these outcomes⁴⁵. The complex and multi-sectoral nature of drought impacts, coupled with challenges in characterising their spatio-temporal dynamics, impedes the integration of impact information into drought projections⁴³. Bridging these gaps is critical for developing strategies to mitigate drought impacts, enhance resilience, and ensure sustainable resource management under climate change.

The 2023 report by the National Oceanic and Atmospheric Administration (NOAA) highlights the critical need for enhanced drought assessments that explicitly account for the impacts of climate change⁴⁶, thereby underscoring the urgency and relevance of this study. In response, this research aims to project the global agricultural impacts of climate change-induced soil moisture droughts under varying future scenarios, recognising the growing threats posed by climate change to soil moisture regimes^47,48. By identifying the regions most at risk, the study seeks to generate actionable insights that can support policymakers and stakeholders in developing targeted adaptation strategies.

In this regard, the primary objectives of this paper are outlined as follows:

• To assess the historical and projected trends of soil moisture droughts on a global scale under different climate change scenarios using climate model simulations.

• To evaluate the spatial and temporal variations of soil moisture droughts and their implications for agriculture.

• To bridge the gap between hazard-based drought assessments and impact-driven approaches by integrating drought severity with regional vulnerabilities for improved climate resilience strategies.

Unlike many previous studies that primarily rely on precipitation- or temperature-based drought indices, this study directly employs bias-corrected soil moisture outputs from multiple climate models to assess agricultural drought conditions. The novelty of this work lies in three key aspects: (i) the application of a consistent bias-correction and multi-model ensemble framework to soil moisture itself rather than to its climatic drivers, (ii) the use of a non-parametric SSMI formulation that is robust under non-stationary climate conditions, and (iii) the development of a drought exposure index that integrates drought characteristics with global cropland distribution to quantify agricultural drought exposure spatially. This integrated framework enables a shift from conventional hazard-based drought assessments toward a spatially explicit, soil-moisture-driven agricultural drought exposure framework at the global scale.

The findings are expected to guide the formulation of effective adaptation strategies, such as drought preparedness plans, water resource management policies, and agricultural subsidy frameworks, thereby strengthening resilience and supporting the transition towards a more sustainable future. This work contributes to the broader effort to understand the complex interplay between climate change and soil moisture droughts, leveraging advanced modelling techniques and scenario analysis to offer critical insights for advancing sustainable development in the face of a rapidly changing environment.

Reference and CMIP6 climate forcing data

In this paper, monthly 0.25-degree soil moisture data from the Global Land Data Assimilation System (GLDAS-Noah25-M.2.1) provided by the National Aeronautics and Space Administration (NASA) were utilised as the reference dataset^49,50 for the baseline period of 2000–2014. The data were forced with meteorological inputs comprising a combination of observational and modelled sources, and then post-processed with satellite data⁵¹. Previous research has established GLDAS as a reliable dataset for soil moisture records^52,53, providing high-resolution, globally available data for drought assessment and hydrological studies.

GLDAS-Noah v2.1 was selected as the reference dataset due to its globally consistent, gap-free, and physically based simulation of soil moisture at 0.25° resolution, driven by a combination of observational forcing, reanalysis data, and satellite assimilation. Compared to purely remote-sensing soil moisture products, GLDAS provides temporally continuous records that are not affected by cloud cover, retrieval uncertainties, or sensor limitations. In contrast to some reanalysis datasets, GLDAS has been extensively evaluated and widely adopted in hydrological and drought studies as a reliable representation of land surface soil moisture dynamics. Its compatibility with land surface modelling frameworks also makes it particularly suitable as a benchmark for evaluating climate model outputs.

To ensure consistency between GLDAS and soil moisture simulations, total root-zone soil moisture was used in this study. For GLDAS, the Noah model soil layers were aggregated to represent effective root-zone storage. A consistent approach was applied to the climate model outputs. This avoids inconsistencies associated with comparing individual soil layers of varying thicknesses across models and ensures that both datasets represent comparable integrated soil moisture conditions relevant to agricultural drought assessment.

Regarding climate simulations, monthly soil moisture data were used for the hindcast period of 2000–2014 and the projection period of 2015–2050 under the climate scenarios of SSP2-4.5 and SSP5-8.5 (SSP stands for Shared Socio-economic Pathways) from CMIP6 (Coupled Model Intercomparison Project Phase 6), with model selection based on the availability of consistent data across both periods (see Table 1).

CMIP6 is a global climate modelling initiative that provides multi-model simulations to assess past, present, and future climate conditions⁵⁴. SSP2-4.5, or the “middle-of-the-road” scenario, assumes moderate social and technological progress with emissions stabilising and radiative forcing reaching 4.5 W/m² by 2100, reflecting a balance between economic growth and environmental policies⁵⁵. In contrast, SSP5-8.5 describes a “high-emission” future focused on rapid economic growth and heavy reliance on carbon-intensive energy, resulting in radiative forcing of 8.5 W/m² and more severe climate impacts until 2100⁵⁶. By comparing these scenarios, we can assess the effects of different socio-economic pathways on soil moisture, aiding in better planning for future climate risks and sustainable solutions. Some details of the 17 CMIP6 models used are provided in Table 1⁵⁷, and to unify the spatial resolution, all models were reclassified on a 0.25° × 0.25° grid using the bicubic interpolation method⁵⁸, suggested by previous research⁵⁹.

CMIP6 data post-processing

CMIP modelling data are prone to biases^60,61,62 due to various factors, including: simplifying assumptions, inaccuracies in model parameterisation, boundary conditions, physical/structural processes, and input variables^63,64. However, high-resolution reliable projections of the Earth’s future climate are crucial for climate-related research and decision-making. Common methods to generate such reliable projection data include statistical downscaling, dynamical downscaling, and hybrid statistical-dynamical downscaling^65,66. In recent years, correcting biases in CMIP models has become a key focus in statistical downscaling studies^{67,68,69,70,71} leading to the development of various bias correction techniques⁷². Compared to traditional downscaling methods, bias correction based on multi-model ensemble means (MMEM) is expected to provide more reliable projections of future climate conditions^73,74.

This study applied a bias correction method⁷⁵ to improve soil moisture data from the selected CMIP6 models based on the GLDAS reference data. The applied bias correction method is a simple yet effective technique used to adjust systematic biases in climate model outputs^75,76,77. This method assumes that the relationship between the monthly observed and modelled data can be corrected using a statistical transformation, ensuring that the corrected values are more consistent with historical observations⁷⁷:

The approach has been thoroughly tested and proven effective for various variables⁷⁶. Further details can be found in previous studies⁷⁸. It is important to note that the linear scaling approach primarily corrects systematic mean bias while preserving the temporal structure of the original model simulations. Because the correction is applied as a monthly ratio between GLDAS and CMIP6 means, the relative anomalies and temporal evolution of soil moisture remain unchanged. As a result, the method does not impose an external trend or reshape the time series, and the historical trends simulated by each model are retained after correction. This property is particularly important for climate change impact studies, where preserving the modelled temporal dynamics is essential for reliable drought trend analysis.

To evaluate the effectiveness of the bias correction method in this study, several statistical metrics have been employed, as presented in Table 2. The choice of evaluation criteria has been recommended by previous studies^79,80.

Assessing soil moisture droughts

To assess soil moisture droughts, this paper calculated the SSMI using the following steps^34,81:

1. 1.

   Time series were created for cumulative bias corrected MMEM soil moisture data.

2. 2.

   Parameter estimation via a non-parametric distribution and the computation of cumulative probabilities: In the present study, the marginal probability distribution of soil moisture data was determined by employing the empirical Gringorten distribution method:

   $$P=(i-0.44)/(N+0.12)$$

   (2)

   while N is the sample count, i is the sample’s position sorted from the smallest, and P is the estimated probability^82,83.

3. 3.

   Then, the probabilities were normalised as follows:

   $${rm{SSMI}}={Phi }^{-1}(P)$$

   (3)

   Φ is the standard normal distribution function⁸¹. The percentiles were standardised using the following equations⁸⁴:

   $${mathrm{SSMI}}=left{begin{array}{l}-left(t-frac{{C}_{0}+{C}_{1}t+{C}_{2}{t}^{2}}{1+{d}_{1}t+{d}_{2}{t}^{2}+{d}_{3}{t}^{3}}right){if},0 < ,P,le 0.5\ left(t-frac{{C}_{0}+{C}_{1}t+{C}_{2}{t}^{2}}{1+{d}_{1}t+{d}_{2}{t}^{2}+{d}_{3}{t}^{3}}right){if},0.5 < ,P,le 1end{array}right.{therefore} left{begin{array}{l}{c}_{0}=2.5155\ {c}_{1}=0.8026\ {c}_{2}=0.0103end{array}right., {&} ,left{begin{array}{l}{d}_{1}=1.4328\ {d}_{2}=0.1893\ {d}_{3}=0.0013end{array}right.$$

   (4)
   $${mathrm{while}}:t=left{begin{array}{l}sqrt{{ln}frac{1}{{P}^{2}}},{if},0 < Ple 0.5\ sqrt{{ln}frac{1}{{(1-P)}^{2}}},{if},0.5 < Ple 1end{array}right.$$

   (5)
4. 4.

   To end, to assess how well the reference GLDAS data and the bias corrected MMEM data conform to the empirical distribution, the K-S (Kolmogorov-Smirnov) test was employed⁸⁵. This test establishes a level of confidence in using the empirical distribution function as the reference distribution for the dataset.

The proposed framework employs the SSMI to classify dry and wet conditions. In this classification, negative values indicate dry periods, while positive values correspond to wet periods⁸⁶. A SSMI between –0.5 and 0.5 represents normal soil moisture conditions. The SSMI is further categorised into different severity levels: abnormally dry (D₀: –0.50 to –0.79), moderately dry (D₁: –0.80 to –1.29), severely dry (D₂: –1.30 to –1.59), extremely dry (D₃: –1.60 to –1.99), and exceptionally dry (D₄: –2.00 or lower). Similarly, wet conditions are classified as abnormally wet (W₀: 0.50–0.79), moderately wet (W₁: 0.80–1.29), severely wet (W₂: 1.30–1.59), extremely wet (W₃: 1.60–1.99), and exceptionally wet (W₄: 2.00 or higher). This categorisation enables a comprehensive assessment of soil moisture anomalies, facilitating a more accurate evaluation of drought conditions.

Defining drought frequency, duration and severity/intensity

To define the drought characteristic variables (i.e., frequency, duration, and severity/intensity), this study used the definition proposed by previous research³⁵. Once a drought event was identified along with its start and end months, its characteristics were determined (see Fig. 1):

• Drought frequency refers to the number of drought events that occur within a specific period at a given location for a given threshold (e.g., SSMI < -1 in this study). It is typically expressed as the total count of drought occurrences over a predefined time interval (e.g., per decade).

• The duration corresponds to the number of months from the start month (inclusive) to the end month (exclusive).

• Severity is defined as the absolute value of the integral area between the SSMI values and the horizontal axis (SSMI = 0) over the duration of the drought event. It is crucial to differentiate severity from intensity, with intensity referring to the average SSMI value during the drought (unit: value per month). In some studies, intensity is instead defined as the lowest SSMI during the event³⁵, also known as “critical drought severity”⁸⁷, in which case intensity has no associated unit.

Understanding the effects of soil moisture droughts in a changing climate

Soil moisture is a critical factor influencing crop productivity, as it directly affects plant growth, yield, and overall agricultural output². In the context of climate change, future soil moisture droughts are expected to intensify, with varying regional impacts on crop production. This study explores the impacts of future soil moisture droughts on agriculture by using data from the Land Cover Climate Research Data Package (CRDP), which provides high-resolution global land cover information⁸⁸. To comprehensively assess the potential impacts of agricultural droughts, we propose a Drought Exposure Index (DEI), defined as the cube root of the combined normalised drought frequency, total duration, and mean intensity:

This index integrates three fundamental drought characteristics into a single metric, capturing how often droughts occur, how long they persist, and how intense they are on average. The use of the geometric mean provides a balanced representation of these components, preserving proportional relationships among them. By synthesising these dimensions, the DEI offers a more holistic and interpretable measure of drought exposure, suitable for regional comparisons and scenario-based assessments.

Study framework

Fig. 2 illustrates the methodological framework of this study, outlining the step-by-step approach used to assess the impacts of soil moisture droughts on agriculture.

Uncertainties in future projections and CMIP6 post-processing analyses

The performance of climate models varies across different regions and time periods⁸⁹, highlighting their spatial and temporal inconsistencies⁹⁰. Acknowledging the inherent uncertainties in CMIP6 simulations is essential to ensure the credibility and applicability of the results⁹¹. While some degree of uncertainty is inevitable due to model structure and assumptions, employing multi-model ensembles, such as the 17 CMIP6 models used to construct the MMEM in this study, can offer a robust approach to capturing a broader range of plausible future scenarios and enhancing confidence in climate projections⁹².

The generated heatmap in Fig. 3 visualises the normalised performance metrics of the CMIP6 models used, ranging from 0 (relatively worst) to 1 (relatively ideal) for the historical period. Each model is represented along the Y-axis; while performance metrics including Bias, CC, IoA, KGE, MAE, ME, and RMSE (as discussed in Table 2) are aligned along the X-axis. The original numerical values, prior to normalisation, are displayed within each cell, enabling direct comparison across models and metrics. Overall, this visualisation offers a comprehensive assessment of the CMIP6 models’ accuracy. The bias-corrected MMEM data demonstrated the best overall performance across the evaluated metrics, achieving the highest normalised values for Bias, IoA, KGE, MAE, ME, and RMSE. This underscores the effectiveness of bias correction in improving model accuracy and reducing systematic errors. Given its superior performance, the bias-corrected MMEM data was selected for drought projections.

Projection of soil moisture droughts

Fig. 4 displays the projected spatial distribution of the 1-month SSMI for January 2030 under two climate scenarios: (a) SSP2-4.5 and (b) SSP5-8.5. It should be noted that this single-month example is presented to illustrate the spatial representation of SSMI under different scenarios and does not serve as a basis for any short-term climate interpretation, as the analysis of this study is based on long-term statistical behaviour and aggregated drought characteristics. The colour scale represents SSMI values, where blue shades indicate wetter conditions and red shades denote drier conditions. The figure visually suggests differences in soil moisture patterns between the two scenarios, with relatively drier conditions appearing more prominently under SSP5-8.5 in parts of Africa and Asia. This visual comparison is intended to provide an example of how SSMI maps reflect scenario-driven spatial variability rather than to infer specific conditions for that month.

Figure 5 presents the projected temporal changes in the number of pixels classified under different drought (D₀-D₄) and wetness (W₀-W₄) categories based on the SSMI from 2015 to 2050. Panels (a) and (c) display the distribution of pixel counts across various drought and wetness categories under the SSP2-4.5 and SSP5-8.5 climate scenarios, respectively. The results indicate a gradual shift towards more extreme conditions, with an increasing number of pixels categorised under severe drought (D₃-D₄) over time, especially under the SSP5-8.5 scenario. Panels (b) and (d) show the corresponding trends in mean SSMI values, revealing a clear declining trend, suggesting worsening soil moisture deficits, particularly under the more extreme SSP5-8.5 scenario. These findings highlight the potential intensification of drought conditions in the future, emphasising the need for climate adaptation strategies.

Figure 6 illustrates the projected spatiotemporal changes in the SSMI across global continents under two future climate scenarios, SSP2-4.5 and SSP5-8.5, for the period 2015–2050. The central map (a), based on GLDAS soil moisture data (2000–2014), visualises the baseline distribution of soil moisture in mm. This map provides important spatial context for understanding regional vulnerability.

Surrounding the map, regional time-series plots (b to g) depict SSMI trends across six continents. Importantly, Australia exhibits considerable interannual fluctuations in soil moisture, suggesting heightened climate variability and the potential for both extreme wet and dry spells. In contrast, regions such as Asia and North America show smoother and more gradual downward trends, implying a more consistent drying trajectory over time. These patterns reflect differences in regional climate sensitivity. South America and Europe show strong signals of drying in both magnitude and consistency, which could indicate increased risk of prolonged droughts. Africa’s trends are relatively stable but still show subtle drying signals, hinting at regional heterogeneity in moisture dynamics.

Additionally, panel (h) complements these observations with box plots summarising the statistical distribution of SSMI values across scenarios and regions. Larger interquartile ranges and more frequent outliers are evident in SSP5-8.5, particularly in Australia, reflecting greater variability and extremes. These findings underscore the need for policy-makers to prioritise adaptive strategies in agricultural planning and water resource management. Importantly, such climate-induced variability directly threatens food security, posing serious challenges to the achievement of SDG 2: Zero Hunger.

Drought frequency, duration and severity/intensity

Figure 7 illustrates the projected 12-month soil moisture drought characteristics under SSP2-4.5 and SSP5-8.5 scenarios for the period 2015–2050. The spatial distribution of drought frequency under both SSP2-4.5 and SSP5-8.5 reveals that most regions worldwide are expected to experience between 5 and 15 drought events during the study period. Under SSP2-4.5, relatively lower drought frequencies are observed across eastern Canada, eastern Europe, parts of central Asia, central South America, and southeastern Australia. Under the high-emission SSP5-8.5 scenario, drought frequency generally declines slightly compared to SSP2-4.5, particularly in central and eastern Canada, parts of the Amazon basin, northern Africa, parts of central Asia, and eastern Europe. Nevertheless, certain regions, such as the Middle East, and southern Europe, maintain a moderate drought frequency under both scenarios.

During this period, regions that are less vulnerable to drought, based on intensity and duration, are identified as follows: under the SSP2-4.5 scenario, central parts of North America, eastern Canada, northwestern South America, parts of Asia, southern and northwestern Africa, central Europe, and southeastern Australia. Under the SSP5-8.5 scenario, the areas include eastern Canada, the eastern and southern regions of North America, northwestern Brazil, northern Europe, parts of Asia, and central Australia.

The global average values of agricultural drought characteristics during 2015–2050 reveal subtle differences between the SSP2-4.5 and SSP5-8.5 scenarios. Under SSP2-4.5, the average drought frequency is approximately 9.09 events, while it slightly decreases to 8.61 events under SSP5-8.5. Total drought severity remains nearly unchanged between the two scenarios, with average values of 98.29 and 98.40, respectively. The mean drought intensity is also consistent, showing a global average of 1.27 in both cases. Similarly, the total drought duration averages around 68.10 months under SSP2-4.5 and 68.14 months under SSP5-8.5. These findings highlight the increasing vulnerability of global land areas to prolonged and severe droughts, reinforcing the need for adaptive water resource management and climate mitigation efforts.

The findings show that, in general, drought events tend to last longer under SSP5-8.5 (higher duration), resulting in fewer distinct drought events within a given time period (lower frequency). Rather than many short droughts, the conditions favour fewer but more prolonged drought episodes. This leads to a reduction in the number of drought occurrences, even though the overall drought impact may be more severe due to extended dry periods.

Agricultural drought impact analysis

Figure 8 illustrates the effects of climate change on agricultural land across different continents under the SSP2-4.5 scenario. Panel (a) maps global agricultural land cover classes, while panel (b) presents the proportion of agricultural land by continent. According to the CRDP dataset, Asia (primarily India and China) accounts for the largest share, comprising 34% of the global agricultural area, followed by South America (22%), Europe (17%), Africa (13%), North America (11%), and Australia (3%).

Panels (c) to (h) depict the percentage of agricultural land affected by drought conditions (D₀ to D₄ levels, based on 1-month SSMI) for each continent over time, revealing a general upward trend towards 2050 with varying rates of increase and seasonal fluctuations. Australia, Asia, North America, and South America show the greatest projected variability and exposure to drought events, while Europe and Africa exhibit a steadier rise in affected cropland. This variability reflects differing regional sensitivities to changing climatic conditions, such as exposure to El Niño-Southern Oscillation (ENSO)-driven climate extremes, and underlying land cover characteristics⁹³. In the broader context of sustainable development, the projected increase in drought-affected agricultural lands jeopardises progress towards SDG 2 (Zero Hunger), while also intersecting with SDG 13 (Climate Action) and SDG 15 (Life on Land).

Figure 9 presents similar information under the SSP5-8.5 scenario. Each subplot illustrates the temporal evolution of affected areas, highlighting increasing trends, particularly in North America, South America, Asia, and Australia. Africa shows the lowest overall percentage of agricultural area impacted. These results emphasise the need for regional drought adaptation strategies and climate-resilient agricultural planning.

Figure 10 presents global maps of the normalised DEI under two SSPs: SSP2-4.5 (panel a) and SSP5-8.5 (panel b), for the period 2025–2050. The DEI values, scaled to the range [0–1], enable consistent spatial comparison across regions and scenarios. These maps offer a detailed overview of projected drought exposure, with higher values reflecting greater potential for agricultural drought driven by soil moisture deficits.

The results reveal pronounced regional disparities, with especially high DEI values across major croplands in South Asia, Southern Africa, South America, southern Europe, and parts of the United States. Under the high-emission scenario (SSP5-8.5), both the extent and severity of exposure increase markedly, particularly in intensively farmed regions, highlighting the intensifying impact of climate change on soil moisture dynamics. The global mean DEI rises from 0.7128 (SSP2-4.5) to 0.7793 (SSP5-8.5), indicating a widespread increase in drought exposure under the more extreme pathway. These findings emphasise the critical need for targeted adaptation efforts in vulnerable agricultural zones, especially as emissions pathways diverge.

Pathways to a sustainable future

Addressing the increasing frequency, severity and duration of climate change-driven soil moisture droughts requires a comprehensive approach that integrates scientific advancements, policy interventions, and sustainable land management strategies^94,95. Effective mitigation and adaptation efforts in the context of SDGs must be guided by robust projections, such as those presented in this study, to inform decision-making at local, regional, and global scales^96,97.

• One crucial pathway is the “implementation of climate-resilient agricultural practices”^98,99. Sustainable farming techniques, including precision irrigation, conservation tillage, and the adoption of drought-resistant crop varieties, can enhance soil moisture retention and reduce vulnerability to prolonged droughts^100,101,102. Additionally, agroforestry and regenerative agriculture can improve soil structure and water-holding capacity, contributing to long-term sustainability^103,104.

• Another key strategy is “enhancing water resource management” through innovative technologies and policies^105,106. Integrated water resource management approaches, such as managed aquifer recharge, rainwater harvesting, and smart irrigation systems, can optimise water availability during drought periods¹⁰⁷. Furthermore, international cooperation in transboundary water governance is essential in mitigating the impacts of droughts, particularly in water-stressed regions^108,109.

• “Climate adaptation policies and global governance frameworks” must also play a pivotal role^110,111. Governments and international organisations should prioritise the inclusion of soil moisture droughts in climate adaptation plans, ensuring that drought risk reduction strategies are embedded within national policies¹¹². This includes improving drought monitoring networks, developing early warning systems, and investing in climate-informed decision-support tools¹¹³. Strengthening policy incentives for sustainable land-use practices and fostering climate finance mechanisms will further enhance resilience¹¹⁴.

• Moreover, “advancing climate modelling and data-driven decision-making” can support proactive planning^115,116. High-resolution reliable climate projections, improved remote sensing techniques, and artificial intelligence-based projecting models will provide more accurate drought risk assessments¹¹⁷. This will enable policymakers, stakeholders, farmers, and water managers to take timely actions and reduce the adverse impacts of extreme drought events¹¹⁸.

• Lastly, “community engagement and capacity-building” are fundamental for a sustainable future^119,120. Local communities must be empowered with knowledge, resources, and adaptive strategies to manage soil moisture deficits effectively¹²¹. Raising public awareness on water saving²³, education, stakeholder participation, and interdisciplinary collaborations will be critical in ensuring that scientific findings translate into practical solutions^122,123.

The results of this study reveal a clear intensification of soil moisture drought characteristics across many global agricultural regions under both SSP scenarios, with more pronounced drying trends under SSP5-8.5. The increase in drought duration suggests a shift toward longer and more persistent drought events. This behaviour has also been reported in previous global drought assessments^124,125 and indicates a transition from short episodic droughts toward prolonged soil moisture deficits that pose greater risks to agricultural systems.

The continental analysis highlights distinct regional sensitivities. South America, southern Europe, South Asia, and parts of North America show consistent drying signals, while Australia exhibits strong interannual variability, reflecting high climate sensitivity. When these projected drought characteristics are combined with cropland distribution through the DEI, the results clearly demonstrate that agricultural regions are expected to experience substantially higher exposure to soil moisture droughts by mid-century. This confirms the importance of moving beyond hazard-based drought analysis toward impact-driven assessments that explicitly link drought behaviour to agricultural vulnerability.

Despite the advantages discussed, this study is not without its challenges and limitations. The following outlines the key challenges and limitations associated with this study:

(a) Limited availability of soil moisture data: Studies have demonstrated that the limited availability and low quality of soil moisture data can introduce significant uncertainty into spatio-temporal drought assessments^126,127. Inadequate gauge coverage can further compound this issue, especially when data gaps are filled using information from various sources based on statistical assumptions¹²⁸. As a result, combined errors in variable estimation and the poor quality of reference datasets can compromise the reliability of SSMI calculations^129,130.

To mitigate these uncertainties, high-quality data are essential for improving initial conditions, which form the basis for accurate drought projection¹³¹. Data assimilation techniques, such as the Ensemble Kalman Filter and Four-Dimensional Variational assimilation, are powerful tools for integrating diverse observational data (from remote sensing satellites and/or in situ measurements) into hydrological and climate models¹³². These methods enhance drought prediction by optimally combining model data with observations, thereby reducing uncertainties and improving the representation of land surface processes¹³³. However, satellite data can also be limited by spatial and temporal resolution, cloud cover interference, and the need for ground-based validation¹³⁴.

In recent years, several Land Data Assimilation Systems (LDAS) have been developed and widely used to improve drought assessments under climate change conditions^135,136. These systems provide refined parameterisations for simulating seasonal and diurnal variations in key variables critical to drought monitoring across different timescales¹³⁷. Moreover, some LDAS models supply soil moisture data at multiple depths, thereby increasing the accuracy of drought evaluation¹³⁸. However, land surface models continue to evolve, particularly in refining soil hydraulic property representations in both horizontal and vertical dimensions, including the incorporation of features such as sinkholes¹³⁵.

(b) Challenges and uncertainties in CMIP6 data: Drought modelling presents significant challenges and uncertainties due to the inherent complexity of drought events and their interactions with climate, hydrology, and socio-economic systems^139,140. A major source of uncertainty stems from the varying complexity of global climate models, as differences in how Earth system processes are represented can substantially influence model outputs. Analysing ensembles of CMIP6 models helps quantify these uncertainties by evaluating the internal spread across multiple simulations^141,142. In an effort to reduce model-related uncertainty, this study incorporated a wide range of CMIP6 models.

Uncertainty also arises from the initial conditions used in CMIP6 simulations. Incomplete observational data and errors in data assimilation can propagate through simulations, affecting how climate variability is represented¹⁴³. Employing diverse reference datasets and rigorous evaluation processes can help mitigate these effects.

Beyond model and ensemble selection, spatial resolution introduces further challenges. Coarse-resolution models may fail to capture small-scale processes, while higher resolution usually comes with significant computational costs¹⁴⁴. To improve the accuracy of our projections, we applied a bicubic interpolation method in combination with a bias correction approach. Future research could explore more downscaling techniques, such as machine learning or dynamical methods, to further refine drought projections^145,146.

Additional uncertainty stems from the use of SSPs, as socio-economic trends and climate policy choices can lead to divergent future scenarios. This study selected SSP2-4.5 and SSP5-8.5 to represent contrasting climate futures. Future studies could consider a broader set of SSPs to provide a more comprehensive assessment. Addressing these diverse sources of uncertainty is essential for improving the reliability of future drought assessments.

(c) Limitations of the bias correction method: The applied bias correction method is a simple and computationally efficient approach used to adjust systematic biases in climate model outputs¹⁴⁷. It operates on the assumption that the relationship between observed and modelled data can be corrected using a linear transformation, ensuring greater consistency with historical observations¹⁴⁸.

However, despite its simplicity, availability and effectiveness, this method has some limitations: firstly, it assumes a constant linear relationship between observed and modelled data, making it ineffective for correcting nonlinear changes¹⁴⁹. While it adjusts mean values, it does not account for variability, extremes, or changes in statistical distributions. In addition, it presumes that bias characteristics remain unchanged over time, which may not hold under evolving climate conditions¹⁵⁰. It also does not adequately correct seasonal and interannual variations, potentially leading to biases in projected climate trends¹⁵¹.

We used the linear bias correction method due to its simplicity, computational efficiency, and ability to preserve long-term trends. Given the large volume of climate model data in our study, more complex post-processing techniques would have significantly increased computational costs, making the linear approach the most practical and effective solution. Since our focus was primarily on adjusting systematic biases rather than capturing distributional changes or extreme events, the linear approach provided an effective and pragmatic solution.

(d) The impacts of temperature on soil moisture droughts: Understanding the full spectrum of temperature-driven land-atmosphere interactions remains a key challenge in climate science. While the effects of rising temperatures on snowpack and subsequent hydrological processes are relatively well understood, the sensitivity of evapotranspiration and associated moisture losses to temperature remains less clear^152,153. This uncertainty stems from the complex responses of vegetation to both atmospheric conditions and long-term climate change^154,155, with varying magnitudes of impact reported across models, experiments, and observations^156,157. Moreover, although temperature plays a critical role in shaping climate model projections¹⁵⁸, many models still fall short in accurately representing temperature-driven vegetation-climate feedbacks¹⁵⁹. Enhancing the reliability of future projections, therefore, requires improved alignment between modelled vegetation processes and observational records, complemented by insights from paleo-ecological and paleo-hydroclimate data^126,160.

(e) Non-stationary climate: selection of probability distribution for the chosen baseline period: Choosing the right probability distribution is crucial for developing reliable drought indices under climate change, particularly when accounting for stationary versus non-stationary patterns of climate variables¹²⁶. If selecting the distribution is not approached carefully, it can lead to significant uncertainties. In case of using parametric distributions, due to the non-stationary nature of climate variables under future scenarios, the parameters of soil moisture’s probability distribution will evolve over time. Consequently, it is crucial to account for this non-stationarity by adjusting the probability distribution parameters across various timescales to enhance drought assessments in the context of climate change^126,161.

The selection of the baseline period is also crucial when comparing future droughts under climate change to historical droughts as a reference¹⁵². Using a longer period as the baseline allows for better calibration of drought indices, enabling more accurate comparisons between future and historical drought events. However, an inappropriate choice of the baseline period for evaluating drought events can introduce bias in drought assessments under climate change¹²⁶.

(f) Limitations of individual drought characteristic analysis and the importance of joint modelling: Uncertainties in drought analysis arise due to the complex, interdependent nature of drought characteristics including duration, frequency, and severity/intensity¹⁶². Analysing these variables in isolation, particularly severity without considering its associated duration or frequency, can lead to misleading conclusions, as severe droughts are often long-lasting or recurrent events^163,164. To capture the full spectrum of drought behaviour and address these uncertainties, it is recommended to use joint probability approaches¹⁶⁵. Copula theory provides a robust statistical framework for this purpose, enabling the modelling of dependencies among multiple drought attributes without assuming linear relationships^166,167. By employing Copulas, researchers can more accurately assess compound risks and generate realistic return periods, thereby enhancing drought risk management and planning under uncertainties¹⁶⁸.

(g) Uncertainty due to future land cover changes: One of the key sources of uncertainty in this study arises from probable future changes in land cover, which can significantly influence soil moisture dynamics and, consequently, the projected characteristics of agricultural droughts¹⁶⁹. Land cover modifications, including deforestation, urban expansion, and shifts in agricultural practices, can alter evapotranspiration rates, soil infiltration capacity, and surface runoff, all of which affect soil moisture availability¹⁷⁰. However, CMIP6 models primarily rely on prescribed land use scenarios¹⁷¹ that may not fully capture regional-scale variations or abrupt land cover transitions driven by policy changes, socio-economic developments, or extreme climatic events. This limitation introduces uncertainties in the projected drought characteristics, as the static or simplified land cover representations in climate models may not reflect the complex interactions between land use change and hydrological processes. Addressing this uncertainty requires integrating dynamic land cover projections or coupling climate models with land surface models that account for human-induced land use changes.

(h) Determining the impacts of anthropogenic influences: Human activities, alongside natural climate variability, significantly influence drought initiation, progression, and societal impacts^172,173. Accordingly, drought risk management focuses on adapting to natural drought drivers and mitigating human-induced impacts. Identifying the anthropogenic causes of drought is key for improving water management policies. However, incorporating human factors into hydrological models is still in its early stages, making it difficult to fully characterise droughts under climate change^174,175. The challenge also lies in distinguishing between natural and human influences due to the complex interactions between climate, soil, and vegetation¹²⁶.

Detection and attribution techniques, which combine observations with climate models like CMIP6 in virtual forcing scenarios, attempt to estimate drought characteristics in the absence of human influence. Yet, human infrastructure, such as dams and reservoirs, can significantly affect soil moisture and the propagation of droughts¹⁷⁶. Drought indices need to account for such changes, as well as other human factors like irrigation efficiency and population growth¹⁷⁷. These dynamics, however, remain underdeveloped in large-scale hydrological models, and scientific advancements are needed to address these challenges^178,179,180.