Divergent Drying Mechanisms in Humid and Non-Humid Regions Across China

Abstract

Understanding the drying mechanism is critical for formulating targeted mitigation strategies to combat drought impacts. This study aimed to reveal divergent drying mechanisms in humid and non-humid regions across China from the multidimensional perspectives of climate, vegetation, and energy balance. During the period 1982–2012, the Standardized Precipitation Evapotranspiration Index (SPEI) revealed non-significant drying trends across China. Simultaneously, temperature and precipitation indicated a warming and drying pattern in the humid regions, contrasted with a warming and moistening pattern in the non-humid areas. The coupling effects of declined precipitation, increased vegetation coverage, and elevated temperature exacerbated dryness in the humid regions, while pronounced warming dominantly caused dryness in the non-humid regions. The inverse correlations between the actual evapotranspiration (ET) with precipitation and potential ET (PET) highlighted the principal role of moisture availability in divergent drying mechanisms over humid and non-humid regions. Random Forest models recognized precipitation and PET as the primary factors influencing SPEI in the humid and non-humid regions, respectively. Ongoing warming from 2013 to 2022 mitigated dryness in the humid regions due to the increased latent heat at the expense of sensible heat. Conversely, warming, amplified by the heightened sensible heat, exacerbated drought in the non-humid regions. By identifying the contrasting responses of humid and non-humid regions to warming and moisture availability, this study provides crucial insights for policymakers to mitigate drought impacts and enhance resilience in vulnerable non-humid areas.

1. Introduction

Droughts, characterized by persistent precipitation deficits for weeks or even months, are often evaluated based on the departure from long-term climatological norms [1]. The ongoing climate warming has generally accelerated hydrological processes [2,3] by modifying the intensities and distributions of precipitation and evapotranspiration [4], resulting in the increased frequency and magnitude of drought events [5,6,7]. Longer-than-expected dry spells have been revealed by observation-constrained projections based on climate models [8]. Given its high exposure, drought, one of the costliest and least understood natural hazards, has exerted disastrous effects on vulnerable systems and drawn global attention [9,10].

Despite the scientific consensus that drought frequency and duration have increased in most regions of the world, uncertainty in drought analysis remains high due to its spatiotemporal dynamics and the complexities in determining the onset and cessation of its consequent effects [11]. Therefore, establishing a universally applicable definition for drought remains unresolved [12]. Prior research characterizing droughts largely relies on drought metrics given their easy application and interpretation, such as the Palmer Drought Severity Index (PDSI) [13], the Standardized Precipitation Index (SPI) [14], and the Standardized Precipitation Evapotranspiration Index (SPEI) [15]. Since precipitation (P) and potential evapotranspiration (PET) are fundamental hydro-climatic variables directly influencing droughts [16], indices based on P and PET are deemed reliable and have been widely applied in traditional measurements and future projections of drought severity [17,18,19,20]. Although historical analysis and future projections of aridity changes have been widely discussed [21,22,23,24], the accurate measurement of drought risk remains challenging [8,10]. Reversals between dry and wet conditions have been extensively reported due to the changing drying mechanism under climate change [25,26,27]. Given the ongoing uncertainty in drought analysis arising from varying drought indicators and complex spatiotemporal dynamics of drought under climate change [9,18,28], there is a critical need for a deeper scientific understanding of drying mechanisms to effectively address the evolving nature of droughts.

Previous studies attributing changes in drought have primarily focused on basic meteorological elements like air temperature, solar radiation, precipitation, relative humidity, etc. [29,30,31]. Global warming as a consequence of rising greenhouse gas emissions has been reported to increase drought frequency, duration, and severity [7,8,22]. Moreover, vegetation dynamics can affect drought by modulating water and energy exchanges between the land and atmosphere [18], making vegetation dynamics and plant physiological forcing influential factors [32,33]. Both climate and vegetation changes are considered to understand vegetation–drought coupling under climate change [34,35]. Furthermore, increased drought risks are also associated with large-scale natural climate variability modes, such as the El Niño–Southern Oscillation (ENSO) [23,36]. Empirical evidence suggests that the potential adverse consequences of aridification may be especially severe in global drylands subject to permanent or seasonal water deficiency [37,38]. Meanwhile, humid regions are likely to be exposed to drought risks driven by anthropogenic disturbances [39], such as urbanization [40]. Previous analysis has predominantly centered on the individual driver of drought changes [23,40], with growing evidence linking global warming and vegetation dynamics to the drought frequency and severity [34,41]. A comprehensive framework that integrates climate, vegetation dynamics, and energy and water flux exchanges is in urgent need to enhance our understanding of drying mechanisms for effective drought mitigation.

In China, drought is one of the primary natural disasters, with certain regions experiencing a drought frequency as high as 70% in summer [42]. It has been reported that drought losses in China could double with a temperature increase from 1.5 °C to 2.0 °C [43]. The warming and wetting trend in arid Northwest China has attracted much attention from the scientific community, government, and the general public [26,44]. Conversely, comprehensively understanding regional drying mechanisms in the context of climate change is vital for informing policymakers in developing targeted drought mitigation strategies for vulnerable areas. Therefore, the objective of this study is to investigate the differing drying mechanisms in humid and non-humid regions across China, focusing on processes involved in climate, vegetation, and energy balance. Specifically, we utilize surface meteorological observations and satellite-based vegetation data to explore drying mechanisms from 1982 to 2012. Subsequently, we validate our assumptions by extending the analysis from 1982 to 2022. The validated divergent drying mechanisms are critical for developing reliable adaptation strategies to minimize regional drought impacts under climate change.

2. Materials and Methods

2.1. Material

2.1.1. Observed Climate Data

We employ a 0.25° × 0.25° gridded meteorological dataset to compute drought indices from 1982 to 2022. This dataset includes daily average air temperature (Ta, °C), relative humidity (RH, %), precipitation (P, mm), and wind speed (u₂, m/s). The data are obtained from the China Meteorological Data Service Center (http://data.cma.cn/site/index.html (accessed on 17 July 2023)), which aggregates information from 2416 meteorological stations across China operated by the China Meteorological Administration (CMA). This high-resolution dataset offers a finer perspective on regional climate changes, facilitated by the dense network of observation stations [45], and has been extensively utilized in prior analyses [30,46]. In addition to calculating drought indices, the observed meteorological data also serve as important driving factors in analyzing drought mechanisms.

2.1.2. Satellite NDVI Product

The third-generation Normalized Difference Vegetation Index (NDVI) data are collected from the Global Inventory Modeling and Mapping Studies (GIMMS) group. The GIMMS-NDVI3g data provide maximum NDVI values twice per month (15-day interval), with a spatial resolution of 1/12° (~8 km), offering the longest record (1981–2015) of global vegetation dynamics and their spatiotemporal variation. This dataset has been corrected for sensor degradation, inter-sensor differences, cloud cover, solar zenith angle, viewing angle effects, and volcanic aerosols [47,48]. As an effective proxy for plant photosynthesis, the biweekly GIMMS-NDVI3g series has been extensively employed for monitoring vegetation growth [34,48,49]. To match the GIMMS-NDVI3g data with the observed meteorological datasets (0.25° resolution, monthly), we calculated the monthly average NDVI data and resampled it to 0.25° for consistency from 1982 to 2012. To minimize the impact of sparse vegetation in hyper-arid regions on the general reliability of the results, we removed grids with annual average NDVI values below 0.1 [50] in the vegetation-related analysis (Figure A1).

2.1.3. GLDAS Water and Energy Fluxes

The Global Land Data Assimilation System (GLDAS) originated from the expansion of the pre-existing North American land data assimilation system project, aiming to compile a comprehensive dataset of land surface water and energy fluxes [51]. It integrates model simulations and observations from the year 2000 onwards. Recognizing the inherent imperfections of model predictions and observations, data assimilation techniques such as Kalman filters, optimal interpolation, and hybrid insertion methods are utilized to enhance data resolution and accuracy [52]. To investigate drying mechanisms from the perspective of energy balance, we collect 0.25° × 0.25°GLDAS monthly energy flux data (https://ldas.gsfc.nasa.gov (accessed on 16 June 2024)), including net solar radiation (Rn, W/m²), latent heat flux (LE, W/m²), sensible heat flux (SH, W/m²) and ground heat flux (G, W/m²), from 1982 to 2022. Furthermore, to evaluate the impact of water flux on drought development, we also collect canopy evaporation (Ec, W/m²), soil evaporation (Es, W/m²), and vegetation transpiration (Tv, W/m²) under the combined influence of surface moisture and vegetation conditions. The three variables were summed up to obtain actual evapotranspiration (ET, W/m²).

2.2. Methods

2.2.1. Calculation of AI and SPEI

To identify humid and non-humid regions in China, we employed the average aridity index (AI), calculated as the ratio of precipitation to potential evapotranspiration, from 1982 to 2022. China was divided into five distinct regions stretching from the southern and eastern areas to the northwest based on AI values including humid (AI > 0.65), dry sub-humid (0.5 < AI ≤ 0.65), semi-arid (0.2 < AI ≤ 0.5), arid (0.05 < AI ≤ 0.2), and hyper-arid (AI ≤ 0.05) regions (Figure 1a). AI has been widely acknowledged in previous studies as a useful metric to reflect climatic dryness of different levels [6,53,54]. For this study, AI was calculated using observed meteorological data collected from the CMA. The “humid regions” refer to areas with an AI exceeding 0.65, while “non-humid regions”, also defined as drylands, where AI is less than 0.65 [55,56], encompass sub-humid, semi-arid, arid, and hyper-arid regions (Figure 1b).

To calculate AI, we initially employed the Penman–Monteith equation [57] to calculate PET (mm/day) (Equation (1)). This formula can achieve reliable PET calculation since it accounts for both energy balance and aerodynamics theory [58].

$$PET={\displaystyle \frac{0.408\Delta (R_n-G)+r\frac{900}{T+273}{u}_2(e_s-e_a)}{\Delta +r(1+0.34u_2)}}$$

(1)

where $\Delta$ is the slope of saturation vapor pressure–temperature curve (kPa/°C), R_n is the net radiation (MJ/m²/day), G is the soil heat flux (MJ/m²/day), r is the psychrometric constant (kPa/°C), T is the daily average temperature (°C), u₂ is the 2 m high wind speed (m/s), and e_s and e_a are the saturation and actual vapor pressure (kPa).

With available monthly PET and precipitation (P) data, SPEI is calculated by fitting the differences between monthly P and PET (D_i, Equation (2)) to the log-logistic distribution (Equation (3)), followed by transforming probabilities to the standard normal distribution (Equations (4) and (5)).

$$D_i=P_i-PE{T}_i$$

(2)

$$f(x)={\displaystyle \frac{\beta }{\alpha }}{({\displaystyle \frac{x-\gamma }{\alpha }})}^{\beta -1}{[(1+{({\displaystyle \frac{x-\gamma }{\alpha }})}^{\beta })]}^{-2}$$

(3)

where $i$ represents the month. $\alpha$, $\beta$, and $\gamma$ represent the scale, shape, and origin parameters for $D$ values ($D<\gamma <\infty$), respectively.

$$f(x)={[1+{({\displaystyle \frac{\alpha }{x-\gamma }})}^{\beta }]}^{-1}$$

(4)

$$SPEI=W-{\displaystyle \frac{c_0+c_{1}W+c_2{W}^2}{1+d_{1}W+d_2{W}^2+d_3{W}^3}}$$

(5)

where c₀ = 2.515517, c₁ = 0.802853, c₂ = 0.010328, d₁ = 1.432788, d₂ = 0.189269, and d₃ = 0.001308. We select SPEI to monitor drought or its superiority by considering the joint effects of precipitation and evapotranspiration comprehensively. Moreover, SPEI can monitor drought on multiple temporal scales when compared to PDSI [59]. A decrease in SPEI suggests an increase in dryness and an increase in SPEI indicates a decrease in dryness.

2.2.2. Other Statistical Method

(1)

Pearson’s correlation analysis

Pearson’s correlation coefficients (r) are calculated (Equation (6)) to capture relationships among SPEI, P, PET, and ET. The correlation analysis provides an initial discussion of the drying mechanisms over humid and non-humid regions.

$$r={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n(x_i-\overline{x})(y_i-\overline{y})}}{\sqrt{{\displaystyle {\sum }_{i=1}^n{(x_i-\overline{x})}^2}}\sqrt{{\displaystyle {\sum }_{i=1}^n{(y_i-\overline{y})}^2}}}}$$

(6)

where $x_i$ and $y_i$ refer to the values of SPEI, P, PET, and ET; $\overline{x}$ and $\overline{y}$ are the corresponding averages of $x_i$ and $y_i$.

(2)

Sen’s slopes and Mann–Kendall statistical tests

The non-parametric Sen’s slope estimator [60] was applied to detect trends in drought, climate, and vegetation variables. Moreover, we employed the non-parametric Mann–Kendall statistical test [61] to assess the significance levels of Sen’s slopes (Equations (7)–(10)).

$$S={\sum }_{i=1}^{n-1}{\sum }_{j=i+1}^{n}sng\left(T_j-T_i\right)$$

(7)

where $n$ is the number of years, and $T_i$ and $T_j$ are the annual average values of each variable in time $i$ and $j$ ($j>i$), respectively. The function $sgn$$(T_j-T_i)$ is given in Equation (8).

$$sgn\left(T_j-T_i\right)=\left\{\begin{array}{c}1if{T}_j-T_i>0\\ 0if{T}_j-T_i=0\\ -1if{T}_j-T_i<0\end{array}\right.ordatamissing$$

(8)

The variance of S is computed as Equation (9).

$$Vars(S)={\displaystyle \frac{n(n-1)(2n+5)-{\displaystyle {\sum }_{i=1}^m{t}_i(t_i-1)(2t_i+5)}}{18}}$$

(9)

where $m$ is the number of tied groups and t_i denotes the number of ties of extent $i$. The standard normal test statistic $Z_S$ is calculated using Equation (10).

$$Z_S=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}}ifS>0\\ 0ifS=0\\ {\displaystyle \frac{S+1}{\sqrt{Var\left(S\right)}}}ifS<0\end{array}\right.$$

(10)

The absolute $Z_S$ above 1.96 and 2.58 denote the 95% (p < 0.05) and 99% (p < 0.01) significance levels, respectively.

(3)

Variable importance by Random Forest models

We developed Random Forest (RF) models to measure the importance of 13 variables encompassing climate, vegetation, energy, and water fluxes in predicting SPEI across five regions from humid to hyper-arid climates. RF models offer significant advantages in accurately predicting and exploring mechanistic relationships within large, intricate datasets characterized by non-normality, non-independence, and collinearity [62]. This approach has been widely used in previous analyses to address multicollinear issues [36,63,64]. Before running RF models, we compute monthly anomalies of these 13 predictor variables by subtracting the 31-year (1982–2012) monthly average value from the corresponding monthly values. These monthly anomalies from 1982 to 2012 were then randomly partitioned into sets for model training and evaluation. RF models were iteratively fitted 100 times to maximize out-of-bag scores, and the importance of each variable was evaluated using the average permuted predictor delta error.

To demonstrate the performance of RF models, three statistical metrics were applied to compare RF-estimated SPEI values with the calculated SPEI based on meteorological observations including goodness of fit (R², Equation (11)), mean absolute error (MAE, Equation (12)) and mean square error (MSE, Equation (13)).

$$R^2=1-{\displaystyle \frac{{{\displaystyle {\sum }_{i=1}^n(y_i-{\widehat{y}}_i)}}^2}{{{\displaystyle {\sum }_{i=1}^n(y_i-{\overline{y}}_i)}}^2}}$$

(11)

$$MSE={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n(y_i-{\widehat{y}}_i)}}{n}}$$

(12)

$$MAE={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n\left|y_i-{\widehat{y}}_i\right|}}{n}}$$

(13)

where $i$ refers to the month, and $n$ is the total number of months ($n$ = 372) from 1982 to 2012. $y_i$ is the calculated SPEI based on observed P and PET calculated from observations. ${\widehat{y}}_i$ is the RF-estimated SPEI. ${\overline{y}}_i$ is the average monthly calculated SPEI.

While constructing RF models, SPEI is the sole response variable with 13 explanatory variables under consideration. For climatic variables, we use average air temperature, precipitation, relative humidity, wind speed, and potential evapotranspiration calculated from meteorological observations. In terms of plant physiology, we adopt the GIMMS-NDVI3g data. For energy and water fluxes, we employ net solar radiation, latent heat flux, sensible heat flux, ground heat flux, canopy evaporation, vegetation transpiration, and soil evaporation from GLDAS.

3. Results and Discussion

3.1. Drying Trends and Driving Factors

From 1982 to 2012, the SPEI revealed a non-significant drying trend (−0.019/10a) on average in China, with a larger drying trend (−0.046/10a) in the humid regions compared to the non-humid regions (−0.010–−0.038/10a) (Figure 2). Among the five regions, the SPEI consistently declined in the humid, sub-humid, and semi-arid regions. Conversely, the arid and hyper-arid regions experienced shifts from wetting (increased SPEI) to drying (declined SPEI) trends in 1993. These reversals in SPEI-based trends were predominantly attributed to the joint effects of elevated air temperature, reduced relative humidity, and intensified wind speed [30,65]. Although the arid region of Northwest China has been reported to undergo a warming and wetting trend characterized by increased temperature and precipitation since the mid-1980s [26], the rise in precipitation cannot fundamentally alter the arid climatic conditions in Western China, nor the nature of arid and semi-arid climate regimes [66].

In addition to SPEI, we also examined temporal variations in precipitation, temperature, NDVI, and ET to comprehensively understand dryness changes in China (Figure 3). From 1982 to 2012, precipitation exhibited a slight declining trend (−0.16 mm/10a) on average in China (Figure 3a), largely contributed by the reduction in precipitation (−0.71 mm/10a) over the humid regions. Notably, the non-humid regions experienced increased precipitation (0.21–0.43 mm/10a). The divergent patterns of “wet areas becoming drier, and dry areas becoming wetter” were also observed, with the rainfall amount and rainy days increasing in the dry regions but decreasing in the humid regions of China [67]. The increase in precipitation over China’s drylands may be caused by the higher convective precipitation, which concentrates in the mountainous areas [46]. Compared to the non-significant variations in precipitation, significant (p < 0.01) warming trends were detected in both the humid (0.34 °C/10a) and non-humid (0.37–0.46 °C/10a) regions, with overall warming (0.39 °C/10a) across China (Figure 3b). According to precipitation and temperature variations, the humid regions experienced a warming and drying trend from 1982 to 2012, while the non-humid regions exhibited consistent warming and wetting trends, which could be beneficial for ecosystems in arid regions [68].

Simultaneously, NDVI indicated significant (p < 0.01) greening over the humid (0.007/10a) and sub-humid (0.007/10a) regions, with non-significant greening trends in the arid regions (Figure 3c). The regional discrepancy in NDVI variations contributed to a non-significant greening trend (0.002/10a) on average in China. Due to the significantly increased NDVI, ET increased significantly over the humid (1.46 W/m²/10a, p < 0.01) and sub-humid (0.64 W/m²/10a, p < 0.05) regions. Influenced by the non-significant increases in ET over the arid regions, China experienced a non-significant increase in ET (0.83 W/m²/10a) from 1982 to 2012 (Figure 3d). Multiple ET products detected larger increasing trends in ET over the humid southeast region compared to the dry western regions in China [69]. We further investigated variations in the canopy evaporation (Ec), vegetation transpiration (Tv), and soil evaporation (Es) (Figure A2), and found that the increased Ec and Tv resulted in an enhanced ET over the humid regions, while the increased Es largely contributed to the increased ET over the sparsely vegetated hyper-arid region.

3.2. Divergent Drying Mechanisms

The SPEI revealed consistent drying trends in China, while precipitation and temperature suggested warming and drying trends over the humid regions but warming and wetting trends over the non-humid regions. To elaborate on the disparity, we discussed the Pearson correlation among SPEI, P, PET, and ET (Figure 4). Consistent significant positive/negative correlations exist between SPEI and P/PET. The positive dependence of SPEI on P decreased from the humid (r = 0.82) to hyper-arid (r = 0.48) regions, while the negative dependence of SPEI on PET increased from the humid (r = −0.70) to hyper-arid (r = −0.97) regions. Notably, P shows a non-significant negative correlation with ET over the humid (r = −0.25) regions but a significant positive correlation with ET over the non-humid regions (r = 0.56–0.88). Moreover, PET and ET are positively correlated over the humid regions (r = 0.75) but negatively correlated over the non-humid regions (r = −0.23–−0.44). In summary, the humid and non-humid regions largely differ in the correlation between ET and P, and between ET and PET. According to a prior analysis, P is the dominant factor for ET in non-humid (water-limited) regions, while vapor pressure deficit dominates ET changes in humid (energy-limited) regions [69].

Following the correlation analysis, we sketched the relationship between SPEI, P, PET, and ET (Figure 5a,b) and calculated trends in these four variables (Figure 5c–f) to deduce the dominating factor of SPEI-based dryness. The dryness in the humid regions is jointly influenced by the significantly increased PET and declined precipitation. The ET dominated by PET over the humid regions indicated a larger increasing trend than that in PET. We assumed that significant warming and greening along with declined precipitation together contributed to the larger increasing trend in ET. Despite the increased precipitation in the non-humid regions, the dryness caused by the significantly increased PET exceeded the wetness caused by increased precipitation, resulting in widely spread dryness. Moreover, stronger warming also contributed to the dryness in the non-humid regions (Figure 3b). Drought-driven increases in ET in the non-humid regions are concerning as they quickly deplete water resources, causing flash droughts and acute stress on ecosystems [70].

In addition to the climate change and vegetation dynamics, we further examined divergent drying mechanisms from the perspective of the energy balance (Figure 6). In humid regions, warming and greening significantly increase ET, leading to more net solar radiation converted to latent heat. With the abundant precipitation supply, there is sufficient moisture supporting the accelerated evapotranspiration process against warming, resulting in less sensible heat. The reduced sensible heat could alleviate warming, and further ameliorate the dryness caused by increased ET by warming (Figure 6a). Conversely, the lack of precipitation in the non-humid regions fails to support the evapotranspiration process influenced by stronger warming, leading to more net solar radiation converted to sensible heat. The enhanced sensible heat can further strengthen warming; thus, the elevated temperature would in turn intensify dryness (Figure 6b). Climate models predict that the average temperature and precipitation of China’s drylands will continue to increase, and the wetting trend will be more significant in some mountainous regions of China’s drylands under the SSP585 forcing scenario [46]. Based on our assumption, regional disparities in drought over the non-humid regions would be further exacerbated under future warming.

Based on the above analysis, we summarized the divergent drying mechanisms over the humid and non-humid regions (Figure 7). Generally, warming (elevated temperature), greening (increased NDVI), and declined precipitation contributed to the dryness over the humid regions, and the dryness was expected to be relieved with ongoing warming (Figure 7a). Instead, the warming (elevated temperature) outweighed the enhanced precipitation, causing dryness in the non-humid regions, which would be further intensified with ongoing warming (Figure 7b). Future projections have also suggested that the drylands of Northwest China would gradually shift toward a “warming and drying” trend with evident dryness under the high emission scenario (SSP585) [71].

3.3. Validation of Divergent Drying Mechanisms

We further measured the importance of each driving factor to SPEI using Random Forest (RF) models. The 13 variables generally achieved a desirable estimate of SPEI (R² = 0.93–0.96, MSE = 0.009–0.018, MAE = 0.071–0.095) (Figure 8a1–a5). The RF-based importance analysis suggested that P, RH, and PET were the top three influential factors, with P and PET being the most influential in the humid and non-humid regions, respectively. This is expected as SPEI is calculated from the difference between P and PET. Meanwhile, RH is significantly (p < 0.01) correlated with PET across China (Figure A3), making P, RH, and PET the three most influential factors for SPEI. Next to P, RH and PET, canopy evaporation (Ec) is more influential in the humid and sub-humid regions, whereas temperature is more influential in the arid regions. Therefore, it can be concluded that declined precipitation is most influential on the surface dryness in the humid regions, followed by greening and warming. Instead, the PET enhanced by elevated temperature (Figure A3) is most influential on the surface dryness in the non-humid regions.

Based on the concluded drying mechanisms, we further investigated the temporal variations in the latent heat (LE) and sensible heat (SH) from 1982 to 2022 (Figure 9). The LE (50.5 W/m²) far exceeds the SH (30.54 W/m²) in the humid regions with significantly increased LE (2.42 W/m²/10a) and declined SH (−1.74 W/m²/10a). Thus, the difference between the LE and SH has largely increased, particularly in recent decades. Conversely, the SH (43.7–54.9 W/m²) far exceeds the LE (4.9–35.0 W/m²) in the non-humid regions and the average difference between the SH and LE increases drastically from the sub-humid (8.51 W/m²) to hyper-arid (49.94 W/m²) regions. The remarkable difference between the surface LE and SH has been reported in arid Northwest China and humid southeast China [72]. Based on the concluded drying mechanisms, we assumed that the significant increase in the LE at the expense of the SH may relieve the drying trend in the humid regions, while the increased trends in the SH over the arid regions from the semi-arid to hyper-arid regions may further intensify the dryness in the non-humid regions.

We extended the analysis of precipitation, temperature, PET, and SPEI to 2022 to validate the assumed dryness development based on the divergent drying mechanisms (Figure 10). Precipitation and temperature suggested warming and wetting trends across China (Figure 10a,b), while PET indicated significantly larger increasing trends in the non-humid regions than humid ones (Figure 10c). More importantly, the humid and sub-humid regions experienced non-significant drying trends from 1982 to 2012 with significant (p < 0.01) drying trends over the arid regions (Figure 10d). Although drought occurs in both regions, the humid regions may experience faster onset and recovery from drought than the arid regions in China [73]. In the recent decade (2013–2022), SPEI indicated increased dryness from the humid to hyper-arid regions, with notable dryness as reported in arid Northern China and hyper-arid Northwest China [65,74]. Notably, further drought over the non-humid regions during the period 2013–2022 was largely energy-driven as the declined SPEI was predominantly caused by the elevated PET instead of the increased precipitation. This finding is consistent with our assumption that dryness would be relieved over humid regions but intensified over non-humid ones with continuous warming, which justifies the divergent drying mechanisms that we have concluded.

3.4. Discussion

Through exploring divergent drying mechanisms, we identify substantial drought risks in both humid and non-humid regions. For the densely vegetated, humid regions in southern China (Figure A1), rising temperatures coupled with vegetation greening increase evapotranspiration rates, which can offset the moisture gains from increased precipitation and lead to a decrease in moisture availability. Although vegetation greening cools arid and semi-arid regions, southern humid China shows no significant temperature changes [75]. This is consistent with our findings, which indicate that abundant soil moisture in humid areas prevents further air temperature increases by limiting sensible heat. This balance between elevated evapotranspiration and increased precipitation suggests that humid regions remain at risk of drying due to accelerated atmospheric water loss. Earlier research also highlights that in water-surplus regions, the traditionally low vegetation–drought coupling shows a tendency to increase [34]. Importantly, accounting for vegetation–CO₂ feedback can reduce biases in extreme drought frequency by approximately 2% [76]. These findings underscore that warming and greening are critical factors in drought development in humid regions, signaling the need for targeted planning and adaptive strategies to enhance drought resilience.

In contrast, non-humid regions are at an even higher drought risk. Here, the already elevated evapotranspiration rates are further amplified by increasing temperatures, intensifying the drying effects. Global drylands, for instance, have been observed to experience about 44% more warming than humid lands [53]. In these areas, increased precipitation often does not mitigate drought, as the primary driver of drought risk has shifted [65]. Whereas precipitation used to be the main factor influencing drought, hot and dry air masses have now become dominant [30,65]. This shift underscores the growing role of atmospheric aridity, which will continue to increase in arid zones, posing a significant threat to dryland ecosystems under global warming [77]. While increased warming may not directly cause droughts, it does accelerate drought onset and intensify its effects, making it more severe when it occurs [78]. With the ongoing climate change, “flash droughts”, characterized by their abrupt onset and rapid intensification over short periods, have become more prevalent, as documented in numerous studies [41,79,80,81]. This dynamic calls for close attention to both humid and non-humid regions as climate warming continues.

Our study does have several uncertainties and limitations. We employed SPEI to monitor drought conditions, a widely used index but one among others like the AI, PDSI, SPI, and SPEI. Different drought indices, each emphasizing different variables and processes, may yield varying perspectives on drought development, as noted in prior studies [20,82,83]. For instance, while SPI indicates a wetting trend based only on precipitation (P) [82], both AI and SPEI, which account for P and PET, reveal a drying trend across China on average from 1982 to 2012 (Figure A4a–c). SPEI has proven to be especially effective in assessing drought severity across appropriate temporal scales, making it suitable for our analysis [84,85]. Moreover, incorporating both PET and P is vital for distinguishing the divergent drying mechanisms between humid and non-humid regions. Consequently, despite the limitations of relying solely on SPEI, our findings on regional drying mechanisms, based on both P and PET, remain robust and reliable.

Another source of uncertainty lies in the GLDAS energy and water flux data, as we did not apply any data corrections. The GLDAS tends to overestimate downward solar radiation, leading to a larger Rn, LE, and SH compared to model simulations [53]. Nevertheless, the GLDAS provides extensive temporal and spatial coverage of water and energy flux data, meeting the requirements of our study. While individual flux values may lack precision, the regional differences and trends in LE and SH remain valid for supporting our conclusions. Future research should aim to address these uncertainties, with an emphasis on identifying critical thresholds within these divergent drying mechanisms using more precise, corrected data.

4. Conclusions

In this study, we uncovered divergent drying mechanisms in humid and non-humid regions across China considering climate, vegetation, and energy balance. The main findings are summarized below.

From 1982 to 2012, the SPEI indicated consistent drying trends in China, whereas temperature and precipitation suggested a warming and drying trend in the humid regions, contrasting with a warming and wetting trend in the non-humid regions. The observed decline in precipitation, coupled with greening and warming, contributed to the dryness in humid regions, whereas the elevated temperature played a dominant role in the dryness of non-humid regions by increasing PET. Based on the divergent drying mechanisms, ongoing warming is predicted to convert energy to latent heat in humid regions, reducing sensible heat and potentially alleviating drought. Conversely, more energy was likely to be converted into sensible heat, which could exacerbate warming and intensify dryness in non-humid regions. The Random Forest models confirmed that precipitation was the primary factor influencing SPEI in humid regions, while PET dominated in the non-humid regions. The observed increase in dryness from the humid to hyper-arid regions in the recent decade (2013–2022) verified the above divergent drying mechanisms.

Our results indicate that declining precipitation has initiated dryness in humid regions, a process further exacerbated by vegetation greening and rising temperatures. Moreover, with continued warming, drought risks are expected to intensify in non-humid regions due to the positive energy feedback.