The Responses of Vegetation Production and Evapotranspiration to Inter-Annual Summer Drought in Northeast Asia Dryland Regions (NADRs)

Abstract

The impacts of drought on Gross Primary Productivity (GPP) and Evapotranspiration (ET) play an important role in understanding the carbon–water process of dryland ecosystems. However, just via correlation analysis, the response mechanism of vegetation production and ET to droughts is not well understood. Based on a modified Vegetation Photosynthesis Model (VPM) and a revised Penman–Monteith (PM) model, GPP and ET were simulated to examine their sensitivity to drought and quantitative dynamics among biomes with the drought index in NADRs. The diverse response of GPP and ET to drought depending on biomes, grassland, barren/sparse vegetation and shrub showed a positive response to summer drought, while cropland and forest showed a negative response to summer drought. From the normal summers to extreme drought summers, GPP and ET reduced by 0.36 g C m⁻² day⁻¹ and 0.18 mm day⁻¹, nearly 10.54% and 12.77%, respectively. Some compensation mechanisms (i.e., physiological changes of vegetation species to resistant drought) or drought timescale weaken the drought impacts in insignificant correlated regions (GPP or ET and SPEI) with lower reduction rates. Compared with persistent or multiple droughts, the impacts of abrupt wet–dry shifts on GPP and ET were weak with lower rates (4.44% for GPP, 0.92% for ET). Notably, the wet winter and warm spring weakens the summer drought impacts on GPP in some parts of grasslands. These observations would be useful to understand the ecosystem process and to account for the dynamics of ecosystem water use efficiency during drought disturbance in depth.

1. Introduction

The increasing frequency and intensification of droughts under global warming [1,2,3,4] and expansion of drylands in the coming decades [5,6] will alter the coupled carbon–water cycle by affecting the plant community composition [7,8], structure [9,10] and function of the ecosystem [11,12,13]. Understanding the response mechanism of the carbon–water process to drought could give us insights to cope with future droughts.

As the vital indicators of the carbon–water process, Gross Primary Productivity (GPP), Evapotranspiration (ET) and Water Use Efficiency (WUE) are widely used to assess the impacts of drought to ecosystems. Most of research shows that vegetation productivity [14,15,16,17,18] and ET [19,20,21] have been limited by drought at global or regional scale, whereas ecosystem WUE under droughts presents dissimilar variations with drought timing and duration [22,23,24] or intensity [13]. Nevertheless, not enough evidence comes from the drylands. Additionally, dryland ecosystems exhibit different sensitivity to droughts, such as semi-arid regions having larger sensitivity than arid ecosystem [25]. Thus, analyzing the inherent ecological responses to droughts is critical in order to carry out measures to reduce drylands degradation under climate change.

Currently, the methods, mechanisms and processes of ecological responses to droughts are still lacking [26]. Most of the previous research is based on the relationship between drought and indicators [15,27,28,29], or the anomalies of ecological indicators (i.e., GPP or WUE) during drought [18,22] or the difference between dry years and non-dry years [14,30,31,32]. Notably, ignored insignificant related regions between drought and GPP (or ET) might indicate that some compensation mechanisms resist the drought impacts; meanwhile, few studies focus on the impacts of abrupt changes in dry–wet events on carbon–water processes.

Northeast Asia Dryland Regions (NADRs), located in the east of the largest arid zone in the world (Afro–Asian arid zone) [33], are a typical vulnerable eco-environmental area to climate change (i.e., drought) [34]. Due to the vast barren area and grasslands, the responses of the dryland ecosystem to droughts present obvious regional and biomes differences [35], and are still unclear.

Here, our objective is to explain how the carbon–water process responds to droughts in NADRs via quantitative evaluation of the drought impacts on GPP and ET. Given the different responses of the carbon–water process to seasonal drought [24,29], the characteristics of the growing season in NADRs [36], the complexity and inconsistency of alternative WUE parameters [37], only summer GPP and ET were chosen as carbon–water indicators. First, we obtained summer GPP, ET via the Vegetation Photosynthesis Model (VPM) and the revised Penman–Monteith (PM) model (PM) in the whole NADR, respectively. Second, we evaluated the dynamics of GPP, ET and drought conditions. Third, we estimated the sensitivity of GPP and ET to droughts with correlation analysis. Last, we quantified the drought impacts on GPP and ET among different biome types.

2. Materials and Methods

2.1. Study Area

With an aridity index (AI), the boundary of the NADR in this study was defined (Figure 1, AI < 0.65) [38], including most parts of Mongolia, east of Kazakhstan, and arid and semi-arid regions of China except Tibet regions. The NADR has a marked dry continental climate and continental monsoon climate. The annual mean temperature and total annual precipitation have large heterogeneity and vary considerably [39,40]. The high temperature and precipitation are concentrated in the summer months (e.g., June, July and August, JJA).

As the main land cover types, grassland, barren/sparse vegetation and croplands account for 51.3%, 29.2% and 10.5% of NADR total area, respectively (Figure 1). Based on different climate conditions and dominant species, grassland could be divided into 6 classes in China, such as alpine steppe (AS), alpine meadow (AM), temperate desert steppe (DS), typical steppe (TS), meadow steppe (MS) and temperate meadow (TM). Specifically, AS is controlled by cold-xerophytic grasses (e.g., Carex moorcroftii, Stipa purpurea) [41], while perennial grasses (e.g., K. humilis, Kobresia pygmaea and K. tibetica) are dominant species of AM. DS is dominated by xerophytes (e.g., Stipa gobica, Stipa klemezii, Stipa breviflora, Stipa glareosa and Allium polyrhizum) [42], The dominant species of TS are Stipa grandis, Leymus Chinensis, Caragana microphylla, Agropyron cristatum, Stipa Krylovii and Cleistogenes squarrosa. MS is dominated by L. Chinensis, S. grandis, Filofolium sibiricum and Stipa baicalensis [43].

2.2. Estimation of Gross Primary Productivity (GPP) and Evapotranspiration (ET)

The estimation methods of GPP and ET have been studied deeply. For instance, the VPM model performed better than other light use efficiency models (i.e., Carnegie–Ames-Stanford approach (CASA), Global Production Efficiency Model (GLO-PEM)) for GPP in on a large scale [44,45]. The revised PM model could obtain the accurate daily ET data with MODIS time-series products [46,47,48,49]. Given the applicability and accuracy of the model via satellite data in large-scale dryland, the modified Vegetation Photosynthesis Model (VPM) and revised Penman–Monteith (PM) model were applied to estimate GPP and ET in this study, respectively. The details can be referred to in the flow chart in Figure 2.

$$\mathit{GPP}={\epsilon }_g\times \mathit{APAR}\times \mathit{Rsd}\times 0.48$$

(1)

$${\epsilon }_g={\epsilon }_{\mathit{max}}\times \mathit{min}(F_t,F_w)$$

(2)

$$F_t={\displaystyle \frac{\left(T-T_{min}\right)\left(T-T_{max}\right)}{\left[\left(T-T_{min}\right)\left(T-T_{max}\right)\right]-{\left(T-T_{opt}\right)}^2}}$$

(3)

$$F_w=\mathit{WI}\times \mathit{VPDs}$$

(4)

where R_sd is the download shortwave radiation ([46], Figure 2). ɛ_max is the potential LUE (g cm⁻² MJ⁻¹) and ɛ_max for barren/sparse vegetation is 0.389 g cm⁻² MJ⁻¹ [50,51]. F_t represents the effects of temperature on ɛ_max. T_min, T_max and T_opt are the minimum, maximum and optimum temperature for photosynthetic activities, respectively. F_w is reflected by VPD_S and land water index (WI) that was calculated by satellite-derived land surface water index (LSWI) with consideration of the phenology effects on ɛ_max, [11,45].

To estimate total ET, the canopy ET (λE_veg; Equation (6)) and soil evaporation (λE_soil; Equation (7)) were calculated [46].

$$\lambda E=\lambda E_{veg}+\lambda E_{soil}$$

(5)

$$\lambda E_{veg}={\displaystyle \frac{\Delta R_n+\rho c_p(e_s-e_a)/r_a}{\Delta +\gamma \times (1+r_s/r_a)}}$$

(6)

$$\lambda E_{soil}={\displaystyle \frac{\Delta R_{n,soil}+\rho c_p(e_s-e_a)/r_a}{\Delta +\gamma \times (r_{tot}/r_a)}}\times ({\displaystyle \frac{RH}{100}}{)}^{(e_s-e_a)/100}$$

(7)

where Δ (Pa K⁻¹) is the slope of the curve relating saturated water vapor pressure (e_s, Pa) to temperature (K); γ (Pa K⁻¹) is psychrometric constant; R_n (W m⁻²) is available energy; ρ (kg m⁻³) is air density, c_p (J kg⁻¹ K⁻¹) is the specific heat capacity of air; e_a (Pa) is the actual water vapor pressure; r_a (s m⁻¹) is the aerodynamic resistance; and the surface resistance (r_s, s m⁻¹) is the effective resistance to transpiration from the plant canopy.

2.3. Data Sets

2.3.1. MODIS Datasets

Due to the long time series and accessibility on a large scale of the MODIS dataset, the MODIS land and atmosphere products from 2003 to 2015 were collected to estimate GPP and ET, including the MYD09A1 surface reflectance, MYD11A1 land surface temperature and emissivity, MCD12Q1 land cover classification, MYD15A2 LAI, MCD43B3 albedo and atmosphere products (MYD04L2 aerosol and MYD07L2 atmosphere profiles). NDVI and EVI data were obtained from Terra and Aqua MODIS (MOD13A2 and MYD13A2) products to calculate the vegetation cover fraction.

2.3.2. Drought Indicator: The Standardized Precipitation Evapotranspiration Index (SPEI)

Due to the integration of precipitation (P) and potential evapotranspiration (PET) for monitoring drought [52], SPEI achieves remarkable results in eco-hydrological applications [53]. Because of the lagged effect of water on vegetation, 3-month SPEI data covering the period from 2003 to 2015 were selected from the SPEIbase v.2.4 product (https://spei.csic.es/database.html (accessed on 27 January 2025), [54]). The average of 3-month SPEIs in June, July and August were defined as the summer SPEI. Different drought levels were divided based on SPEI (e.g., humid: SPEI ≥ 1.0, normal: −1.0 < SPEI < 1.0, moderate drought (MD): −1.5 < SPEI ≤ −1.0; severe drought (SD): −2.0 < SPEI ≤ −1.5; extreme drought (ED): SPEI ≤ −2.0).

2.3.3. Flux Tower Data

The eddy covariance (EC) flux tower data were used to validate the ET and GPP retrievals derived from the models. Here, 8 flux tower sites were selected in northeast Asia, which represent various major regional biomes (Table S1). For a detailed validation of GPP and ET, refer to Supplementary Materials.

2.4. Methodology

The Man–Kendall trend analysis was used to determine the trends of summer GPP, ET and SPEI over the period of 2003–2015. Significant tests were defined by z value (z ≥ 1.96, significant increase; z ≤ −1.96 significant decrease) at a 95% confidence level [55].

The relationship between summer GPP anomaly (or ET anomaly) and summer SPEI were examined by correlation analysis. Except for the correlation coefficients, the slope of the correlation analysis was also calculated at a 95% and 90% significant level (p < 0.05, p < 0.10).

The anomaly value of GPP and ET is based on Equation (8):

$$X_{anomaly}={\displaystyle \frac{X-X_{mean}}{X_{sd}}}$$

(8)

where X is the variables (GPP or ET in this study), X_mean, X_sd is the mean value and standard deviation of X.

To quantity the drought impacts on GPP and ET, the mean values of GPP and ET were compared among each drought level (humid, normal, MD, SD and ED) at stable biomes during 2003–2015. The differences between normal and dry summers (e.g., MD summer, SD summer and ED summer) were also identified. Additionally, the relative change rates for the abrupt wet–dry shifts cases were evaluated at each stable biome during 2003–2015. The abrupt wet–dry shifts were defined as a normal summer followed by a dry summer, and the difference of GPP or ET between a normal summer and its dry summer followed as shown with the abrupt wet–dry shift index.

3. Results

3.1. Dynamics of Summer Mean GPP, ET and Drought Conditions in NADRs

The spatial patterns of summer mean GPP and ET from 2003 to 2015 showed high consistency with that of ET, which increased from barren to grassland and further to cropland and forest (Figure 3a,b). The variations of GPP and ET have little difference with each other. For instance, most forests and cropland showed a lower variation of GPP (i.e., CV less than 15%) than that of ET. While GPP and ET presented a higher variation in the barren/sparsely vegetated regions in southern Mongolia (Figure 3c,d). As for the trends, GPP and ET had consistent patterns, which increased in most regions in the NADR with 18.25% and 15.60%, significant increases in the regions in the northeast and south of the NADR (Figure 3e,f).

The summer drought in the NADR was still serious from 2003 to 2015. Figure 4 illustrated the summer drought conditions, including drought frequency, trends and the percentage area of droughts that occurred in the NADR from 2003 to 2015. Most parts of the NADR experienced frequent droughts, especially moderate and severe droughts (Figure 4a,b). Extreme drought almost occurred in Mongolia and Inner Mongolia (Figure 4c). Although SPEI did not present significant trends (Figure 4d), the affected area of droughts significantly increased from 2003 to 2010 (Figure 4e). Drought impacted over 30% of the regions of the NADR in 2006, 2007, 2009, 2010, 2014 and 2015, and nearly half of the NADR (45.44%) in 2009. The area affected by extreme drought also reached the maximum (13.31%) in 2009 with low SPEI. The relatively wet summer only occurred in 2011 and 2012.

3.2. The Spatial Patterns of the Relationship Between GPP (or ET) and SPEI

The correlation between summer GPP anomaly (or ET anomaly) and SPEI from 2003 to 2015 had large spatial heterogeneity (Figure 5a,b). Both linear correlation slope (Figure 5a,b) and coefficients (Figure 5c,d) showed that GPP and ET had a positive correlation with SPEI inside the shrub, grassland and some barren/sparse vegetation, while having a negative correlation with SPEI in the ambient forest and croplands. It means drought induces the decreased GPP and ET for grassland and increased GPP and ET for cropland and forest in NADRs. Furthermore, the most significant positive correlated area (p < 0.05, 0.05 < p < 0.10) occurred in the east grassland of Mongolia and middle Inner Mongolia, which was located in the semi-arid regions. The significant negative correlated areas (p < 0.05, 0.05 < p < 0.10) were few, almost distributed in north mixed forest and northeast cropland, which are cold regions. Additionally, GPP and ET showed a minor different response to drought. For example, the significant correlation between ET and SPEI is also presented in the arid desert ecosystem, like Alashan aeolian sandy land (positive) and north of Tarim Basin (negative), where an insignificant correlation between GPP and SPEI was present.

3.3. The Quantitative Changes GPP and ET with Summer Drought

In the whole NADR, with a dryness increase (e.g., from humid to ED summer), the mean value of summer GPP and ET decreased by 0.36 g C m⁻² day⁻¹ for GPP and 0.18 mm day⁻¹ for ET. Additionally, the sensitivity of GPP and ET to summer drought also depends on biomes. As Figure 6 shows, GPP and ET for grassland and barren/sparse vegetation also decreased with increased dryness, while cropland and mixed forest presented opposite trends. Figure 6b,c quantified how much the percentage of GPP and ET changed from normal summers to dry summers. In the whole NADR, the difference of GPP (or ET) between normal and moderate summer (Normal-MD, −3.72% for GPP; −5.95% for ET) was less than that between normal and severe, extreme drought summer (Normal-SD: −6.85% for GPP and −8.74% for ET, Normal-ED: −10.54% for GPP and −12.77% for ET). Barren areas and grassland presented similar trends with NADRs by different change rate values, and the larger change rates (Normal-ED: −14.70% for GPP and −17.58% for ET) occurred in grassland, whereas cropland, ENF and MF have opposite trends. GPP and ET of MF increased 11.89% and 11.34% from a normal to extreme drought summer, respectively. Notably, although GPP (or ET) presented a positive and negative relationship with SPEI for shrubland and cropland, respectively, they first increased then decreased at shrubland (i.e., Normal-MD: +4.13% for GPP, +3.59% for ET; Normal-SD: −18.61% for GPP, −13.24% for ET) and decreased their amplification (i.e., Normal-MD: +5.85% for GPP, +5.29% for ET; Normal-SD: +2.10% for GPP, +0.19% for ET).

Moreover, the impacts of abrupt dry–wet shifts (e.g., normal summer followed by a dry summer) on GPP and ET were also examined among different biomes (Figure 6e,f), which illustrated the lower change rates (Normal-ED: −4.44% for GPP, −0.92% for ET) but similar trends with multiple drought impacts (Figure 6c,d) in NADRs. Similar results also occurred in almost stable biomes except grassland of GPP (Figure 6e). To account for the difference, we classified the grassland into six sub-classes in China based on vegetation species information and the vegetation map as we mentioned at Section 2.1. The statistics results suggested that the GPP of temperate meadow and alpine meadow increased when the ecosystem experienced a transition from a normal summer to an extreme drought summer (Normal-ED, Figure 6g), which may explain the not obvious changes of grassland GPP in Figure 6e. Notably, the reduction of GPP for MS, TS and DS reached the maximum for the case of transition from a normal summer to a severe drought summer (Normal-SD) and not for an extreme drought summer (Normal-ED).

4. Discussion

4.1. The Diverse Responses of Carbon–Water Process to Summer Drought

Different from most research results, our results suggest that the responses of GPP and ET to summer drought differ substantially among biome types (Figure 5), which agrees with Wu and Chen [56]. The diverse response might result from the different main limited factors for ecosystems. For instance, water plays a key role in the functions and activities of semi-arid grassland regions dominated by herbaceous plants [57,58,59]. During drought, plants reduce enzymatic activities as well as stomatal closure to prevent water loss [60], causing the reduction of the transpiration and photosynthesis rate in grassland (Figure 5), whereas energy supply (or temperature) drives the variation of ecosystem functions largely in high latitude [61] and alpine regions [62]. Less precipitation and cloud cover with more coming solar radiation accelerate the carbon–water process [63,64], which could account for the negative response of GPP and ET to summer drought in mixed forests in cold regions and alpine meadows in Qinghai (Figure 5). The negative effects of drought on productivity for boreal regions are also supported by Zhang et al. [61].

Notably, with significant correlation between GPP (or ET) and SPEI, semi-arid grassland present larger sensitivity to drought than arid regions (Figure 5), which also are found in North America [65] and southeastern Australia [25]. As Vicente-Serrano et al. [66] suggested, the drought timescale determines the sensitivity of biomes to drought, and largely, arid biomes respond to drought at short timescales, while semiarid and sub-humid biomes respond at long timescales due to different root systems. After comparing the relationship between GPP (or ET) and the 1-month scale SPEI value (Figure 7), the sensitivity value gets higher, but the significant area is almost similar with Figure 4. It means that other compensation mechanisms except for drought timescale weaken the impacts of drought in insignificant correlated regions, like different vegetation species physiological responses. For example, on the one hand, the arid desert ecosystems in central Asia dominated by some drought-tolerant species (i.e., Tamarix ramosissima) are more resistant to drought with lengthy root systems to tap deep soil water [67,68]. On the other hand, the ephemerals and annual plants with shallow roots just reserve seeds during drought periods, while flowers grow fast after one or several precipitation events [69], which is difficult to reveal via 1- or 3-month SPEI data [66]. Moreover, some drought-tolerant species with shallow root systems (i.e., Haloxylon ammodendron) maintain photosynthesis but reduce transpiration during drought via effective morphological adjustment (i.e., shrinking shoot size) [64], which explains the different response of GPP and ET to drought in Alashan. In barren areas with very sparse vegetation, like northern Tarim Basin, drought mainly induces the increased soil evaporation rather than vegetation dynamics, which suggest the negative response of ET to drought in these regions. Overall, the insignificant sensitivity to drought in arid ecosystems hinders some compensation mechanisms (i.e., physiological changes of vegetation species to resistant drought), which should be paid much more attention to.

4.2. The Quantitative Evaluation of Summer Drought Impacts on Carbon–Water Process

In NADRs, the difference of average GPP (or ET) between humid and extreme drought summer is near 0.36 g C m⁻² day⁻¹ (or 0.18 mm day⁻¹), and GPP (or ET) reduces by nearly 10% (Figure 6c,d) from a normal to extreme drought summer during 2003 to 2015, which is consistent with the global average value [70]. The quantitative impacts of drought were weakened by the insignificant correlations between GPP (or ET) and SPEI. As Figure 8 shows, the difference of GPP (or ET) between humid and extreme drought summer decreases (0.23 g C m⁻² day⁻¹ and 0.11 mm day⁻¹, p > 0.10), thus only analyzing the insignificant correlated regions of the NADR.

As for the different biomes, with the increased dryness, the change directions of GPP and ET at grass vegetation decreases and MF regions (Figure 6a–d) agree with the correlation analysis results (Figure 5). However, shrubs and cropland have few different responses with correlation results (Figure 6a–d). For shrubs, due to the root length and small leaf area, the plants with high water use efficiency could continue the photosynthesis and transpiration process to maintain growth during moderate drought [67,68]. Therefore, GPP (or ET) increases a little from normal to MD summers (Figure 6a–d). For cropland, radiation (or temperature) is the main limited factor, which improves the photosynthesis and transpiration process, but photosynthesis rate will not increase when temperature reaches a threshold [71].

Compared with the abrupt shifts during a continuous two years (i.e., a normal summer followed by a dry summer), the persistent or multiple droughts intensify the impacts on the carbon–water process in NADRs with higher change rates (Figure 6c–f). It should be noted that the impacts of abrupt shifts on grassland GPP are a little different due to the different response of steppe types (Figure 6f). However, the transition from a normal summer to an extreme drought summer did not reduce GPP seriously in meadow steppe, temperate steppe and desert steppe, which disagrees with correlation analysis and common knowledge. Fortunately, we noted that the extreme summers were not frequent (1 to 3 times, Figure 4c) and there is a wet winter and warm spring before extreme summer in the grassland of China (Figure 9). It means that the previous wet winter and warm spring weakens the summer drought impacts on vegetation production, which is also verified in the US [72], because the precipitation in winter could be stored in the soil to support plant growth in the spring [73] and warm spring increases ET and carbon uptake to compensate for the GPP reduction during extreme summer drought [72]. Overall, the wet winter and spring weakens the summer drought impacts on GPP in NADR grassland.

4.3. The Advantages and Disadvantages of This Study

Based on the above analysis, the responses of carbon–water processes to summer droughts have been revealed deeply. Not only were different biomes but also the abrupt wet–dry shifts and drought timescale were taken into consideration to explore the response mechanisms. Meanwhile, the lower sensitivity of GPP and ET to summer droughts was also discussed with some compensation mechanisms. However, some disadvantages should be paid attention to. For example, the accuracy of GPP and ET in barren areas or low-NDVI regions was limited by the satellite sensor. Next, the seasonal differences of carbon–water process to droughts should be analyzed in depth.

5. Conclusions

Through the impacts analysis of summer droughts to GPP and ET, we found that the impacts are diverse and depend on the biome types, the significant relationship between the carbon–water process and drought, and drought timescale. Generally, during summer drought, grassland, barren/sparse vegetation and shrub reduce the GPP (or ET), while cropland and forest regions improve GPP (or ET) due to different main limited factors. And the arid barren/sparse vegetation exhibit smaller sensitivity to droughts than semi-arid grassland due to some compensation mechanisms, which weakens the summer drought impacts on the carbon–water process. Whereas, compared with abrupt wet–dry shifts, persistent or multiple droughts intensify the impacts on the carbon–water process in NADRs with higher change rates. Notably, the wet winter and warm spring weaken the summer drought impacts on GPP in some parts of grasslands. In summary, the response mechanism of carbon–water processes to summer drought is complex but could help us to analyze and account for the dynamics of ecosystem water use efficiency during drought disturbance in further research. Also, this study is conducive to making measures or actionable recommendations for policymakers or land managers dealing with dryland ecosystems.