Precipitation exacerbates spatial heterogeneity in the propagation time of meteorological drought to soil drought with increasing soil depth

The propagation of meteorological droughts to soil droughts poses a substantial threat to water resources, agricultural production, and social systems. Understanding drought propagation process is crucial for early warning and mitigation, but mechanisms of the propagation from meteorological drought to soil drought, particularly at varying soil depths, remain insufficiently understood. Here, we employ the maximum correlation coefficient method and the random forest (RF) model to investigate the spatiotemporal patterns and drivers of propagation time (PT) from meteorological drought to soil drought at four different depths across China from 1980 to 2018. Our findings reveal consistently higher PT in northern China and lower PT in southern China across varying soil depths, with more pronounced spatial heterogeneity with increasing soil depth. Furthermore, we identify temperature and precipitation as determinants of spatial patterns of PT in surface and deeper soil layers, respectively. Additionally, precipitation emerges as the dominant factor influencing changes in PT between different soil layers. Our study highlights a discernible shift in PT drivers from temperature to precipitation as soil depth increases and the significant impact of precipitation on exacerbating spatial heterogeneity in PT. This study contributes to an enhanced comprehension of the propagation process from meteorological drought to soil drought at different depths, which can aid in establishing practical drought mitigation measures and early warning systems.


Introduction
Drought, as a prevalent and intricate natural phenomenon, poses great threats to water resources, agriculture and ecosystems (Mishra andSingh 2010, Jasechko andPerrone 2020).Categorized based on affected objects, drought manifests in four types: meteorological, soil, hydrological and socioeconomic (Wilhite andGlantz 1985, Hao andSingh 2015).Within frequent land-atmosphere interactions involving heat and moisture, the drought signal can shift from one type to another, a phenomenon referred to as drought propagation (Eltahir andYeh 1999, Wu et al 2018).Drought propagation can potentially prolong the duration and expand the extent of droughts, causing more severe damages to natural and social systems (Jehanzaib et al 2019, Zhou et al 2021).Acquiring knowledge on the process and mechanisms of drought propagation is crucial for governments in establishing effective drought monitoring and early warning systems, which can mitigate the adverse impacts of droughts towards human civilization (Barker et al 2016, Bachmair et al 2018, Zhang et al 2021).
The propagation from meteorological drought to soil drought is a frontier problem of drought (Zhang et al 2022b, Adeyeri et al 2023).On one hand, soil drought is the direct successor of meteorological drought, and soil moisture plays a pivotal role in water-carbon cycle (Li et al 2022a, Wang et al 2023).Moisture deficiencies can easily diffuse from atmosphere to soil, triggering soil moisture deficit and amplifying the adverse influence on vegetation growth and terrestrial carbon uptake (Zhou et al 2019, Gampe et al 2021).Meanwhile, the propagation from meteorological drought to soil drought is a complex process influenced by various factors that affect the balance of soil water (Huang et al 2015, Zhu et al 2021, Ding et al 2021a).Recent research has achieved phased results on the process and characteristics of propagation from meteorological drought to soil drought (Xu et al 2021a, Dai et al 2022, Li et al 2022a).However, due to limitations in site-based data availability and the intricate nature of drought propagation process, research on the mechanisms of drought propagation is comparatively limited (Chen et al 2020, Zhang et al 2021).Thus, we can draw on prior studies to enhance research on the mechanism and drivers of propagation from meteorological drought to soil drought.
Previous research has claimed the specific spatial heterogeneity in the characteristics of propagation from meteorological drought to soil drought, primarily focusing on drought propagation time (PT) on global and regional scales (Dai et al 2022, Cheng et al 2023, Zhang et al 2024).However, most studies only describe the dry-wet condition of soil in terms of root-zone soil moisture.Conversely, the sensitivity and response rate of soil moisture to abnormal meteorological changes diverge across different depths, potentially resulting in distinct drought propagation characteristics at varying soil depths (Wu et al 2002, Wang et al 2016, Zhang et al 2020).It still remains unclear how drought propagation characteristics vary spatially and temporally with different soil depths.Additionally, some studies suggest that climatic factors, such as precipitation and evapotranspiration, can influence the propagation process from meteorological drought to root-zone soil drought and contribute to the spatial heterogeneity of drought PT (Ding et al 2021a, Li et al 2022a).There are rare studies attempting to uncover the process and mechanisms of meteorological drought to soil drought at different soil depths.However, soil moisture at different depths affects distinct processes of the watercarbon cycle; for example, the surface, root-zone and deep soil moisture are closely associated with overland flow, vegetation growth, and groundwater recharge, respectively (Xu et al 2021b, Wang et al 2023).Consequently, the moisture and heat transfer process in different soil layers may also be constrained by distinct physical processes that affect the atmosphere-land water balance.For instance, the abnormal meteorological anomalies are more easily detective by surface soil moisture in comparisons with deep soil moisture (Han et al 2023, Li et al 2023).Currently, it is still unknown the specific driving mechanisms of drought propagation at varying soil depths, given clear differences in the rates of heat and moisture transfer between the atmosphere and soil, as well as potential variations in spatial heterogeneity in drought PT.Accordingly, another pivotal question arises regarding the critical factors driving these variations, if both the spatial heterogeneity of drought PT and propagation mechanisms vary with soil depth.Clarifying the above issues is crucial to reveal the process and mechanisms of propagation from meteorological drought to soil drought with increasing soil depths.
China, as one of the most populous and largest agricultural countries globally, is also prone to severe drought-induced damages (Yao et al 2018, Zhang et al 2022a, Zhao et al 2023).In this study, we incorporated the maximum correlation coefficient method and the RF model to investigate the spatiotemporal characteristics and main drivers of the PT from meteorological drought to soil drought at four different depths (0-10, 0-50, 0-100, 0-200 cm) across China from 1980 to 2018.Additionally, we preliminarily discussed the mechanisms of drought propagation at varying soil depths and the potential factors driving the differences in spatial heterogeneity of drought PT among different soil layers.The objectives of this study were as follows: (1) to explore the spatiotemporal patterns of drought PT between meteorological drought and soil drought at different soil depths; (2) to identify the dominant factors accounting for the spatiotemporal patterns of drought PT at different soil depths; (3) to reveal the critical factors driving the variations in drought PT between different soil layers.These findings will elucidate the mechanisms underlying the propagation of meteorological drought to soil drought at varying soil depths and offer valuable insights for decision-makers to develop the early warning system of soil drought.

Drought indices
We utilized the standardized precipitation index (SPI) and standardized soil moisture index (SSI), commonly employed in drought propagation analysis, to characterize meteorological and soil droughts, respectively (Ma et al 2019, Li et al 2020, 2023, Zhang et al 2021, Dai et al 2022).As recommend by McKee et al (1993) and Hao and Singh (2015), we utilized the gamma distribution to fit both precipitation and soil moisture series, and transformed the cumulative frequencies into standardized values via the inverse of the standardized normal distribution function.Additionally, we calculated the SSI series at four different soil depths, i.e. 0-10, 0-50, 0-100, and 0-200 cm, denoting as SSI 10 , SSI 50 , SSI 100 , and SSI 200 , respectively, to depict the dry-wet variations across various soil layers.

Drought propagation
The maximum Pearson correlation coefficient (MPCC) was extensively employed in identifying drought propagation owing to its adeptness in discerning the correlation between distinct droughts (Barker et al 2016, Wu et al 2018, Yu et al 2020, Zhang et al 2022a, 2022b).Similar to early investigations (Ma et al 2019, Xu et al 2021a, Li et al 2022a, 2022b, Han et al 2023, Cheng et al 2023), we utilized MPCC integrating SSI at 1 month timescale and SPI with accumulation period spinning from 1 to 12 months to explore the drought PT.The timescales of SPI aligned with prior studies suggesting an average soil moisture renewal cycle of 1 year (Ding et al 2021b, Li et al 2022a).Specifically, for each soil layer, we initially calculated the correlation coefficients (CC) between SPI and SSI for each calendar month, and subsequently identified the SPI timescale corresponding to the maximum CC as the PT from meteorological drought to soil drought.

RF model
We employed the RF model to identify the primary drivers of PT in China on a monthly basis.The RF model, an adept machine learning algorithm based on classification trees, excelled in generating robust predictions even for missing and unbalanced data due to its insensitivity to multivariate covariance (Zhu et al 2015).In light of potential influencing factors of meteorological drought and soil drought in pertinent research (Huang et al 2015, Ding et al 2021a, Dai et al 2022, Li et al 2022a), we selected four categories of potential explanatory variables for input into the RF model: (i) climatic factors, encompassing monthly mean precipitation (mm), temperature ( • C), vapor pressure deficit (VPD, hPa), evapotranspiration (mm), and potential evapotranspiration (mm); (ii) vegetation index, specifically monthly mean leaf area index (LAI); (iii) land use types, (iv) soil properties, including sand content, silt content, clay content, wilting point, field capacity, saturated moisture content, saturated hydraulic conductivity.All explainable variables were normalized to the range of 0-1 before being applied to the RF model.To curtail the chances of overfitting and enhance prediction accuracy, we trained and evaluated the RF model using k-fold cross-validation (k = 10 in this study), and ultimately built the RF model with 500 regression trees and a leaf node size of 8. We assessed the relative importance of each variable on PT via the built RF model.

Data
We used three sets of gridded and continuous meteorological data (including precipitation, temperature, pressure, etc) from three reanalysis datasets, namely the Global Land Data Assimilation System (GLDAS) NOAH (Rodell et al 2004), the China Meteorological Forcing Dataset (CMFD) (He et al 2020), the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis 5 dataset (ERA5) (Hersbach 2019), for the SPI calculation.Similarly, we retrieved three sets of continuous soil moisture series at various depths from GLDAS (upper to 200 cm), ERA5 (upper to 289 cm), and Global Land Evaporation Amsterdam Model (GLEAM, upper to 100 cm) V3.6a datasets (Martens et al 2017) for the calculation of SSI (table S1 in supplementary material).The accuracy of these soil moisture datasets in soil drought identification for China had been validated by prior studies (Luo et al 2023, Wang et al 2023).We linearly interpolated all soil moisture data of different depths into four depths: 0-10 cm, 0-50 cm, 0-100 cm, and 0-200 cm (for GLDAS and ERA5).Edaphic variables at different soil depths were obtained from the national gridded soil map of China (Liu et al 2020) and transformed into the same four depths as the soil moisture data.LAI data were sourced from the 8d composite 0.05 • × 0.05 • the Global Land Surface Satellite (GLASS) product.Land use data was obtained from MODIS MCD12C1 land use type product with the International Geosphere Biosphere Program (IGBP) classification scheme.All datasets were processed into the same temporal resolution of one month and spatial resolution of 0.25 • × 0.25 • , covering the period from January 1980 to December 2018.In the application of RF model, the multi-year average monthly precipitation, temperature and VPD were calculated as the average results of GLDAS, CMFD, and ERA5, whereas the multiyear average monthly evapotranspiration and potential evapotranspiration were derived from GLEAM.

Spatiotemporal patterns of PT at different soil depths
We retrieved three sets of SPI, SSI 10 , SSI 50 , SSI 100 , and two sets of SSI 200 , resulting in 9 combinations for PT 10 , PT 50 , and PT 100 calculations and 6 combinations for PT 200 calculation (table S2 in supplementary material).We calculated the multi-dataset average CC and PT to address discrepancies across different datasets and ensure a balanced approach, though the Specifically, the PT values in the southern China were consistently lower than those in the northern China for four soil depths.Such spatial differences in PT were observed more pronounced with increasing soil depth (refer to standard deviations and semivariogram parameters in tables S3 and S4 in supplementary material).Temporally, PT the lowest values from July to October for four soil depths, as well as the smallest spatial differences between southern and northern China among all calendar months.We also found the PT values exhibited an increasing trend with the increase in soil depths, with average values of PT 10 , PT 50 , PT 100 , and PT 200 ranging from 2.4 to 5.3, 3.0 to 6.5, 3.5 to 7.3, 4.0 to 8.1, respectively (figure 1(e)).Notably, we also observed that the changes in PT between different soil layers (∆PT) were not uniformly distributed in space and time, with spatially higher values in northern China and lower values in southern China, as well as temporally higher values in March to August (figures S3(a)-(e) in supplemental material).Overall, it was expected that the shallower 10 cm and 50 cm soil layers were more responsive to meteorological variations (Wu et al 2002, Zhu et al 2019), resulting in a relatively rapid response to meteorological drought.In contrast, with the increase in soil depth, deeper soil layers exhibited a longer transfer rate of moisture and heat from the atmosphere to the entire soil column (Mayocchi and Bristow 1995), leading to a prolonged response time to meteorological drought.

Drivers accounting for spatial patterns of PT at varying soil depths
The acceptable results of the constructed RF models confirmed the accuracy of regressions between PT and potential explanatory variables (tables S5 and S6 in supplementary material).According to the importance results, climatic factors clearly predominated over vegetations, land uses, and soil properties in the spatial patterns of PT across all four soil depths (figures 2(a)-(d)).Additionally, we found that temperature emerged as the dominant factor for PT in surface soil layers (0-10 and 0-50 cm, figures 2(a) and (b)).In particular for PT 10 , temperature showed the highest and second-highest importance in 8 and 4 calendar months, respectively.In contrast, we noted the crucial roles of precipitation in determining PT for deeper soil layers (0-100 and 0-200 cm), with precipitation presenting the highest importance in more than 9 calendar months (figures 2(c) and (d)).Overall, our results noted an evident dominant factor shift of PT from temperature to precipitation as soil depth increased.We conducted further investigation into the factors driving the spatial variations of ∆PT between 10-50, 10-100, and 10-200 cm (figures 2(e)-(g)).Notably, precipitation emerged as the primary factor influencing ∆PT from March to August, coinciding with the months showing the most evident spatial variations of ∆PT.Specifically, the results indicated that precipitation determined the increments of PT between the top 0-10 cm soil layer and any deeper soil layer, thereby accentuating the spatial heterogeneity of PT.Similar dominant effects of precipitation were also observed for ∆PT in the intermediate soil layers, i.e. 50-100 and 100-200 cm, particularly pronounced from April to July (figure S4 in supplementary material).
For the months from April to July that PT of all soil layers showed obvious spatial variability (table S4 in supplementary material), we noted significant negative correlations between precipitation and ∆PT between any two soil layers throughout China (figure 3).This implied the lower precipitation led to the greater ∆PT along with soil depth.Additionally, we observed that the coefficients of precipitation and ∆PT in the 10-100 and 10-200 cm soil layers were both lower than those in the 10-50 cm layer (figures 3(a1)-( a4), ( b1)-( b4) and ( c1)-( c4)).In terms of three intermediate soil layers, the coefficient of precipitation and ∆PT in the 10-50 cm soil layer (−0.486) was lower than that in the 50-100 cm (−0.168) and 100-200 cm (−0.27) layers (figures 3(a1), ( d1) and ( e1)), coinciding with the higher values of ∆PT in 10-50 cm soil layer than those in the 50-100 and 100-200 cm layers in April (figure S3(e) in supplementary material).Similarly, the differences in the coefficients between 10-50, 50-100, and 100-200 cm layers were comparatively minor in May to July, consistent with the narrow ranges of ∆PT values in these months.These findings clarified that higher ∆PT values in specific soil layers corresponded to stronger negative correlations with precipitation, which also reinforced the dominant role of precipitation in ∆PT.
Drought propagation was a complex and gradual physical process influenced by numerous factors.Compared to vegetations, soil properties, and land uses, we recognized the critical roles of background climate in shaping the spatial patterns of PT across varying soil depths, which aligned with prior studies focusing solely on the root-zone soil layer (Ding et al 2021a, Li et al 2022a, Wang et al 2023).The prominence of climatic conditions in affecting the land-atmosphere exchange of heat and moisture over an extended period likely underpinned this finding (Cai et al 2009, Qiu and Ben-Asher 2010, Wu et al 2013).In contrast to previous studies, we further unveiled a noteworthy shift in the primary determinant of PT from temperature to precipitation with increasing soil depth (figures 2(a)-(d)).To validate this finding, we also computed the partial CC between PT, precipitation and temperature across different soil layers (figure S5 in supplementary material).These results also underscored a diminishing negative correlation relationship between PT and temperature, alongside an amplified negative correlation between  in dominance corresponded with the changes in the mechanism of soil evaporation, transitioning from energy-constraint, i.e. temperature-dominated in the surface soil to moisture-constraint, i.e. precipitationdominated in the deeper soil (refer to schematic figure in figure S6 in supplemental material).Specifically, due to frequent moisture exchange and sufficient moisture condition, energy condition, rather than moisture condition, governed the surface moisture transfer process, e.g.soil evaporation and transpiration (Wu et al 2002).Elevated temperatures could swiftly intensify soil evaporation and transpiration, expediting moisture transfer and consequent rapid depletion of surface soil moisture (Zhou et al 2019).However, the influence of temperature on soil moisture followed a top-down process due to the thermal lag effect (Jobbagy and Jackson 2000).Consequently, in deep soil layers, moisture condition took precedence over energy conditions as the primary determinant of moisture transfer processes.Precipitation, as the primary replenishment source of deep soil water, regulated the reduction rate of deep soil moisture during meteorological droughts (Guan et al 2019, Ding et al 2021a).Thus, adequate precipitation, closely associated with ample soil moisture, contributed to increased evaporation and moisture uptake by vegetation roots, which strengthened the moisture feedback from deep soil to atmosphere and resulted in a rapid decline in soil moisture under meteorological droughts (Wu et al 2013, Zhou et al 2021).
Moreover, we identified the predominant influence of precipitation in driving spatial disparities of ∆PT with the increase in soil depth (figures 2(e), (f) and 3).This was anticipated, as precipitation facilitated the diffusion of moisture deficiencies throughout the entire soil column (Dirmeyer et al 2006, Li et al 2022a).The significant influence of precipitation on ∆PT also aligned with the shift of the primary driver of PT, which played a critical role in shaping the spatial patterns of PT across distinct soil layers throughout China.Hence, we elucidated the possible reasons for the spatial variability of PT across different soil depths in China.High temperature in southern China accelerated the response time of surface soil to meteorological droughts, while low temperature in northern China prolonged the response time (figure S7b in supplementary material), thereby shaping the initial spatial patterns of PT across China (figures 1(a) and (b)).Additionally, frozen soil in the northern areas possibly hindered the heat and moisture interactions between atmosphere and surface soil, leading to an extended response time (Li et al 2022a).However, because of frequent interactions of moisture and heat between the surface soil and atmosphere, the spatial disparities in PT were less pronounced for surface soil layers.As soil depth increased, the influence of temperature on PT diminished, while precipitation consistently exerted dominance over ∆PT across different soil layers.Consequently, abundant precipitation in southern China facilitated moisture interactions throughout the soil profile, leading to a modest increase in PT as soil depth increased (figure S7(a) in supplemental material) (Muñoz-Sabater et al 2021).In contrast, low precipitation in northern China was associated with a diminished moisture transfer frequency in the soil, thus yielding a more substantial increase in PT across varying soil layers.Additionally, precipitation in northern areas mainly manifested as snow during winter, potentially impeding landatmosphere heat and moisture exchange and thus prolonging the response time (Li andWilliams 2008, Kormos et al 2014).Moreover, the predominant land use types in southern China were forests and woody savannas (figure S8 in supplemental material).The well-developed roots of plants in southern China corresponded to a high ability for moisture uptake from soil, which could accelerate the dry-wet transition in soil (Huang et al 2015).These cumulative differences in ∆PT between northern and southern China exacerbated spatial differences in PT across soil depths, ultimately resulting in significant spatial disparities in PT in the 0-200 cm soil layer across China (figure 1(d)).
Among climatic factors, we also noted the important role of evapotranspiration in determining PT in July and August (figures 2(a7), (d7), (a8) and (d8)).Such findings likely stemmed from the minor spatial variations in temperature and precipitation across China during these months, emphasizing the importance of alternative climatic factors, such as evapotranspiration, in influencing PT.Interestingly, we also observed edaphic factors and LAI contributed more on driving ∆PT between different soil depths than PT for specific soil layers (figures 2(e9)-( e12), (f9)-( f11) and ( g8)-( g12)).This was expected because most edaphic factors, especially soil texture and saturated hydraulic conductivity, were closely linked to the transfer rate of moisture throughout the soil (English et al 2005).Likewise, vegetation could potentially affect soil evaporation and the vertical distribution of soil water, thereby indirectly impacting moisture transfer between soil layers (Qing et al 2023, Sun et al 2024).Regarding land use distribution, we noticed its limited contribution to the spatial patterns of both PT and ∆PT (figure 2).However, we still found an evident shift from temperature to precipitation with increasing soil depth and the predominant effects of precipitation on ∆PT across all land use types (figures S9 and S10 in supplementary material).The consistency in dominant factors further corroborated the findings in figure 2 and supported the notion that land uses had limited impacts on PT distributions.

Conclusion
In this study, we investigated the spatiotemporal patterns and drivers of PT from meteorological drought to soil drought across four different soil depths throughout China from 1980 to 2018.Our analysis utilized both the maximum correlation coefficient method and the RF model, incorporating multiple hydrometeorological datasets to mitigate uncertainties associated with individual datasets.Our findings revealed consistent spatiotemporal patterns in PT across different soil depths, with higher values observed in northern China and lower values in southern China, as well as the lowest values occurred between July and October.Moreover, there was a discernible trend of increasing PT with deeper soil depth, highlighting pronounced spatial heterogeneity between northern and southern China.Through the assessment of variable importance using the RF model, we observed a shift in the dominant factor influencing PT from temperature for surface soil layers (0-10 and 0-50 cm) to precipitation for deep soil layers (0-100 and 0-200 cm).Furthermore, we found that precipitation emerged as the primary driver of spatial disparities in PT increments (∆PT) between different soil layers and exhibited significant negative correlations with ∆PT.
The findings of this study carried important implications for early warning, decision-making, and mitigation of soil drought in China.Likewise, these insights were instrumental in unraveling the dynamics of propagation from meteorological drought to soil drought across various soil layers.In this way, under the projected changes in precipitation linked to global warming, it was also anticipated that the spatial heterogeneity in the PT would further attenuate both in China and globally.