Spatial Pattern of Drought-Induced Mortality Risk and Influencing Factors for Robinia pseudoacacia L. Plantations on the Chinese Loess Plateau

Abstract

During the large-scale vegetation restoration on the Loess Plateau, the introduction of exotic species with high water consumption, such as Robinia pseudoacacia L., led to widespread soil desiccation, and resulted in severe drought stress and increasing risk of forest degradation and mortality. Accurate assessment of drought-induced mortality risk in plantation forests is essential for evaluating and enhancing the sustainability of ecological restoration, yet quantitative research at the regional scale on the Loess Plateau is lacking. With a focus on Robinia pseudoacacia L. plantations, we utilized a coupled model of the Biome BioGeochemical Cycles model and plant supply–demand hydraulic model (BBGC-SPERRY model) to simulate the dynamics of the annual average percentage loss of whole-plant hydraulic conductance (APLK) at 124 meteorological stations over an extended period (1961–2020) to examine changes in plant hydraulic safety in Robinia pseudoacacia L. plantations. Based on the probability distribution of APLK at each site, the drought-induced mortality risk probability (DMRP) in Robinia pseudoacacia L. was determined. The results indicate the BBGC-SPERRY model could effectively simulate the spatiotemporal variations in transpiration and evapotranspiration in Robinia pseudoacacia L. stands on the Loess Plateau. The mean APLK and DMRP exhibited increasing trends from southeast to northwest along a precipitation gradient, with their spatial patterns on the Loess Plateau mainly driven by mean annual precipitation and also significantly influenced by other climatic and soil factors. The low-risk (DMRP < 2%), moderate-risk (2% ≤ DMRP ≤ 5%), and high-risk (DMRP > 5%) zones for drought-induced mortality in Robinia pseudoacacia L. accounted for 60.0%, 30.7%, and 9.3% of the study area, respectively. These quantitative findings can provide an important basis for rational forestation and sustainable vegetation management on the Loess Plateau.

1. Introduction

Forests play a key role in global water, carbon, and nitrogen cycles, and offer a wide array of ecosystem services such as soil and water conservation, air purification, maintenance of biodiversity, etc. [1,2,3]. Due to global climate change, widespread and intense alteration of regional precipitation patterns are causing more frequent and severe soil droughts, which are often accompanied by atmospheric drought induced by high air temperature and vapor pressure deficit (VPD), thus posing a substantial threat of drought-induced tree mortality in many forest biomes around the world [4,5,6]. Accurate assessment of the drought-induced mortality risk for trees is the basis of disaster preparedness and mitigation efforts, and is of vital significance for sustaining the ecological benefits of forest ecosystems.

The drought-induced mortality of trees is a complex eco-physiological process that occurs across different temporal scales [7,8]. Previous studies suggested that merely relying on environmental indicators such as meteorological parameters and soil moisture often proves inadequate for reliable assessment [7,9,10,11]. More accurate quantification of drought-induced mortality risk of trees requires the utilization of process-based eco-hydrological models that explicitly describe the physiological mechanisms involved [12,13]. Accordingly, mechanisms driving tree mortality under drought primarily include hydraulic failure, carbon starvation, and pest and pathogen infestation [14,15]. According to the hydraulic failure hypothesis, xylem embolism progressively occurs due to soil and atmospheric drought, which endangers plant hydraulic safety by impeding long-distance water transport in trees and further leads to dehydration and death of tissues [16]. In terms of carbon starvation, the simultaneous reduction in stomatal aperture and stomatal closure under drought stress diminishes carbon assimilation by photosynthesis, which usually couples with hindered carbohydrate transport in the phloem and increasing carbohydrate consumption to repair drought-damaged tissues, resulting in exacerbated carbon imbalance in plants and ultimately leading to tree mortality [17,18]. Additionally, drought diminishes the resistance of trees to pests and pathogens and also alters the population dynamics of insects and pathogens, including growth rates, mortality rates, and geographical distributions. Consequently, trees become more susceptible to pest and pathogen attacks, which also contributes to increasing mortality [19]. Among these mechanisms, plant hydraulic safety strongly influences stomatal regulation behavior during drought and further impacts carbon dynamics, thereby occupying a central role in determining drought-induced tree mortality [7,20]. Plant hydraulic safety can be better quantitatively simulated through current models than other mechanisms, making it one of the optimal approaches for predicting drought-induced tree mortality [14,16,21].

The Loess Plateau is one of the regions in the world with the most severe soil erosion. To mitigate soil erosion and enhance environmental quality, a series of vegetation restoration strategies have been implemented since the 1950s, including the “Three-North Shelterbelt Development Program”, the “Soil and Water Conservation Program”, the “Grain-for-Green Program”, etc. Robinia pseudoacacia L. is often employed as a pioneer species for vegetation restoration on the Loess Plateau due to its rapid growth and tolerance of poor soil fertility [22,23], and accounts for more than 80% of planted deciduous broadleaf forests in this area [24,25]. However, soil moisture rapidly decreased in most areas due to the high water consumption of the Robinia pseudoacacia L. plantations [26]. The consequent soil desiccation led to the formation of dry deep soil layers, which diminishes the capability of the soil reservoir to supply water to trees; this leaves trees entirely reliant on precipitation for their water use [27,28]. Due to the strong spatial and temporal variability of precipitation on the Loess Plateau, Robinia pseudoacacia L. plantations are susceptible to frequent and severe drought stress [29], and numerous field investigations reported their widespread degradation and mortality [22,30,31,32]. However, quantitative assessment of drought-induced mortality risk at the regional scale for Robinia pseudoacacia L. plantations on the Loess Plateau is still lacking. The sustainability of Robinia pseudoacacia L. plantations and the effectiveness of soil and water conservation efforts face serious uncertainties.

Zhang, et al. [33] incorporated the plant supply–demand hydraulic model proposed by Sperry, et al. [34] into the Biome BioGeochemical Cycles model (abbreviated as the BBGC-SPERRY model). The BBGC-SPERRY model enables the coupled simulation of plant hydraulic traits with water, carbon, and nitrogen cycles in ecosystems using basic stand information and meteorological data, providing an effective tool for assessing tree drought-induced mortality risk based on plant hydraulic safety [35]. Using the BBGC-SPERRY model, the objectives of this study were to (1) simulate the long-term dynamics of plant hydraulic traits in the Robinia pseudoacacia L. plantations on the Loess Plateau; (2) quantitatively assess the spatial pattern of drought-induced mortality risk in the Robinia pseudoacacia L. plantations based on plant hydraulic safety; and (3) analyze the influencing factors of plant hydraulic safety and drought-induced mortality risk in the Robinia pseudoacacia L. plantations. Building on previous field investigations and analyses using climate factors on the Loess Plateau, this study provided a physiological-based quantitative evaluation of drought-induced mortality risk for Robinia pseudoacacia L. plantations at the regional scale. These quantitative findings contribute to the deepening of our understanding of the ecogeographical patterns of drought-induced tree mortality risk, and offer an important basis for rational forestation and sustainable vegetation management in this area.

2. Materials and Methods

2.1. Study Area

The Loess Plateau (100.90° E~114.55° E, 33.72° N~41.27° N) is situated in the middle and upper reaches of the Yellow River, covering an area of 64 × 10⁴ km² (Figure 1). The climate transitions from semi-humid (mean annual precipitation of 250 to 500 mm) to semi-arid conditions (mean annual precipitation of 500 to 800 mm) [36,37], with mean annual precipitation (MAP) decreasing from 700 to 200 mm and mean annual temperature (MAT) decreasing from 14.3 to 3.6 °C from southeast to northwest. The annual potential evaporation ranges from 865 to 1274 mm. The region is characterized by loess soil that is 30 to 200 m thick [38], with soil textures also following a spatial gradient of decreasing clay content and increasing sand content from the southeast to the northwest [39].

According to the hydrogeomorphic types, climatic conditions, soil characteristics, and vegetation construction scheme of the “Three-North Shelterbelt Development Program” (http://www.forestry.gov.cn/c/sbj/gcgk.jhtml, accessed on 13 May 2023), Robinia pseudoacacia L. was widely planted in the loess hilly-gully region located in the southeast of the Loess Plateau. Therefore, this loess hilly-gully region was set as the simulation area for analyzing the drought-induced mortality risk of Robinia pseudoacacia L. plantations (Figure 1). The study area covers 33 × 10⁴ km², and includes diverse landforms including hills, gullies, ridges, and basins, as well as large flat areas with little or no erosion. The predominant soil types in the study area include silt–clay, silt–clay–loam, silt-loam, loam, sand–loam, and loam–sand. The methodology and data sources of the study are shown in Figure 2.

2.2. BBGC-SPERRY Model Description

In the BBGC-SPERRY model, the Biome BioGeochemical Cycles model (Biome-BGC model) was coupled with the plant supply–demand hydraulic model (Sperry model) to simultaneously simulate the dynamics in plant hydraulic traits and water, carbon (C), and nitrogen (N) cycles at daily timesteps (Figure 3). Eco-physiological parameters and values used in this BBGC-SPERRY model are summarized in Table 1. The values of all parameters were derived from either published data for each species in the study area or recommended data for the deciduous broadleaf forest provided by White, et al. [40].

Complete details of the BBGC-SPERRY model were reported in previous studies [33,34,41,42]. Here, we provide a brief summary of the main algorithms. C is added to the system by photosynthesis, and lost by autotrophic respiration, heterotrophic respiration, fire, harvest disturbance events, etc. Photosynthesis is simulated based on the enzymatic kinetics of Rubisco in relation to temperature, CO₂ availability, and the rate of Rubisco regeneration as described in the Farquhar photosynthesis model [43]. For autotrophic respiration, maintenance respiration is modelled by a Q₁₀ function using temperature and N content. Heterotrophic respiration is assumed to be a constant proportion of all new tissue growth. Soil mineral N is added to the system by mineralization from the slowest soil organic matter pool, N deposition from the atmosphere, and N fixation, and lost from the system through leaching and volatilization. Water is input to the system through precipitation as either rain or snow, and rainfall is intercepted by the canopy using a prescribed interception coefficient based on the leaf area index. Water is lost by processes including canopy evaporation of intercepted water, transpiration, and soil evaporation, in which canopy and soil evaporation are calculated based on Penman–Monteith functions with different types of conductance [44], while transpiration and associated water transport processes in the soil–plant–atmosphere continuum (SPAC) are simulated by the Sperry model (Figure 3).

Table 1. Major parameters of the BBGC-SPERRY model for Robinia pseudoacacia L. used in this study.

Parameter	Value
(a) Biome-BGC model
Transfer growth period fraction of growing season (%)	0.2
Litterfall fraction of growing season (%)	0.2
Annual leaf and fine root turnover fraction (yr−1)	1.00
Annual live wood turnover fraction (yr−1)	0.7
Allocation new fine root C:new leaf C	1.0
Allocation new stem C:new leaf C	2.20
Allocation new live wood C:new total wood C	0.209 [40]
Allocation new root C:new stem C	0.22
Allocation current growth proportion (%)	0.5
C:N of leaves	18.8 [45]
C:N of leaf litter	32.2 [46]
C:N of fine roots	26.7 [45]
C:N of live wood	50
C:N of dead wood	550
Leaf litter labile proportion	0.35
Leaf litter cellulose proportion	0.40
Leaf litter lignin proportion	0.26 [45]
Fine root labile proportion	0.34
Fine root cellulose proportion	0.44
Fine root lignin proportion	0.22
Dead wood cellulose proportion	0.68
Dead wood lignin proportion	0.32
Canopy water interception coefficient (LAI−1 d−1)	0.045
Canopy light extinction coefficient	0.54
All-sided to projected leaf area ratio	2.0
Specific leaf area (m2 kg−1 C)	27.9 [45]
Ratio of shaded SLA:sunlit SLA	1.3
Fraction of leaf N in Rubisco	0.14
Cuticular conductance (m s−1)	0.00006
Boundary layer conductance (m s−1)	0.01
(b) Sperry model
Weibull function b and c for root, stem, leaf	b = 2.6, c = 4.1 [33]
Maximum whole-plant hydraulic conductance per leaf area (mmol s−1 m−2 MPa−1)	4.2
Maximum diffusive conductance (mol s−1 m−2)	0.1
Average % resistance in rhizosphere (%)	20.5
Root depth coefficient	0.99 [47]
Number of root and soil layers	5

Note: C = carbon; N = nitrogen; LAI = leaf area index; and SLA = specific leaf area.

Table 1. Major parameters of the BBGC-SPERRY model for Robinia pseudoacacia L. used in this study.

Parameter	Value
(a) Biome-BGC model
Transfer growth period fraction of growing season (%)	0.2
Litterfall fraction of growing season (%)	0.2
Annual leaf and fine root turnover fraction (yr−1)	1.00
Annual live wood turnover fraction (yr−1)	0.7
Allocation new fine root C:new leaf C	1.0
Allocation new stem C:new leaf C	2.20
Allocation new live wood C:new total wood C	0.209 [40]
Allocation new root C:new stem C	0.22
Allocation current growth proportion (%)	0.5
C:N of leaves	18.8 [45]
C:N of leaf litter	32.2 [46]
C:N of fine roots	26.7 [45]
C:N of live wood	50
C:N of dead wood	550
Leaf litter labile proportion	0.35
Leaf litter cellulose proportion	0.40
Leaf litter lignin proportion	0.26 [45]
Fine root labile proportion	0.34
Fine root cellulose proportion	0.44
Fine root lignin proportion	0.22
Dead wood cellulose proportion	0.68
Dead wood lignin proportion	0.32
Canopy water interception coefficient (LAI−1 d−1)	0.045
Canopy light extinction coefficient	0.54
All-sided to projected leaf area ratio	2.0
Specific leaf area (m2 kg−1 C)	27.9 [45]
Ratio of shaded SLA:sunlit SLA	1.3
Fraction of leaf N in Rubisco	0.14
Cuticular conductance (m s−1)	0.00006
Boundary layer conductance (m s−1)	0.01
(b) Sperry model
Weibull function b and c for root, stem, leaf	b = 2.6, c = 4.1 [33]
Maximum whole-plant hydraulic conductance per leaf area (mmol s−1 m−2 MPa−1)	4.2
Maximum diffusive conductance (mol s−1 m−2)	0.1
Average % resistance in rhizosphere (%)	20.5
Root depth coefficient	0.99 [47]
Number of root and soil layers	5

Note: C = carbon; N = nitrogen; LAI = leaf area index; and SLA = specific leaf area.

The Sperry model divides the SPAC system into leaf, stem, root, and rhizosphere components in series, and adopts a supply function and a demand function to solve the water flux, whole-plant hydraulic conductance (k_plant), and distribution of water potential and hydraulic conductance in the SPAC system [34]. The supply function (E(Ψ_leaf)) describes the relationship between transpiration rate (E) and leaf water potential (Ψ_leaf) at a given soil water potential (Ψ_soil) profile. For each component, the vulnerability curve (k(Ψ)_i) quantifies the decline of hydraulic conductance with decreasing water potential (Ψ). The steady-state flow rate through each component (E_i) is then calculated by the integral transform of k(Ψ)_i from the downstream (Ψ_down) to the upstream (Ψ_up):

$$E_{\mathrm{i}}={\displaystyle {\int }_{{\Psi }_{\mathrm{down}}}^{{\Psi }_{\mathrm{up}}}k{(\Psi )}_{\mathrm{i}}d\Psi }$$

(1)

The demand function assumes that stomata regulate the water potential drop from soil to leaf (ΔΨ) based on the fractional drop in soil–plant hydraulic conductance from its maximum:

$$\Delta \Psi =\Delta \Psi \prime [(dE\prime /d{\Psi }_{\mathrm{leaf}})/(dE/d{\Psi }_{\max })]$$

(2)

where ΔΨ′ is the unregulated water potential drop derived from the supply function with unregulated transpiration rate: E′ = VPD × G_max, where G_max is the maximum diffusive conductance. This regulated ΔΨ yields the regulated values for E based on the supply function. G is calculated as E/VPD, and further coupled to the photosynthesis routine. The root and rhizosphere components are partitioned into five layers with equal roots based on the root distribution function. The five-layer root surface water potential and the root crown water potential at the downstream junction can be solved using the multidimensional Newton–Rhaphson method. Root water uptake rate at each layer is then calculated using Equation (1), and further coupled to the water sub-model modified by Huang, et al. [48] for simulating soil water movement based on the Richards equation.

2.3. Model Evaluation

We evaluated the performance of the BBGC-SPERRY model with data of annual actual evapotranspiration during the growing season (ET_a) and daily stem sap flux density collected from previous studies. At an annual scale, ET_a was determined based on the water balance method:

ET_a = P − R + ΔSWS

(3)

where ΔSWS is the change in soil water storage from the start to the end of the growing season (mm); and P and R are precipitation and runoff during the growing season, respectively. Data for daily stem sap flux density at the Yan’an site in 2008 were obtained from Wang [49]. Due to the lack of information on the stand sapwood area, sap flux density cannot be converted to stand-level transpiration. However, variations in sap flux density could largely represent the dynamic of transpiration, and the comparison between daily sap flux density with simulated daily transpiration could still reflect model performance [45].

2.4. Assessment of Drought-Induced Mortality Risk

To assess the spatial pattern of drought-induced mortality risk of Robinia pseudoacacia L. plantations, we simulated the long-term dynamics of plant hydraulic traits at 124 meteorological stations in the study area. Daily meteorological data including temperature (°C), relative humidity (%), and precipitation (mm) for 1961–2020 were obtained from the China Meteorological Administration (Figure 1). Solar radiation, daylight average partial pressure of water vapor, and day length were calculated using the Mountain Climate Simulator (MT-CLIM) program (https://www.umt.edu/numerical-terradynamic-simulation-group/project/mt-clim.php, accessed on 3 March 2024) [41]. Soil data at each site, including clay, silt, and sand content, as well as bulk density, were obtained from the SoilGrid250m dataset (https://www.soilgrids.org/, accessed on 3 March 2024). Soil water retention curves were estimated using the Arya-Paris model [50], and then fitted with the van Genuchten model using the RETC program [51]. Soil water content at field capacity was determined at a soil water potential of −0.033 MPa. Saturated soil hydraulic conductivity was estimated using a pedo-transfer function developed by a machine learning method for the study region [52].

The initial conditions were obtained with the “Spin-up” method at each site. Daily percentage loss of whole-plant hydraulic conductance (PLK, %) was calculated by:

$$\mathrm{PLK}=(1-{\displaystyle \frac{k_{\mathrm{plant}}}{k_{\max }}})\times 100\%$$

(4)

where k_max is the maximum whole-plant hydraulic conductance in the absence of any cavitation (Table 1). Annual average PLK (APLK, %) was then determined to represent the inter-annual variation in plant hydraulic safety. According to previous studies, we set full hydraulic recovery at the start of the growing season due to the generation of root and/or stem pressure and the new growth of plant tissues in the spring [7,53,54]. During the growing season, we imposed the condition that xylem embolism could not be refilled in Robinia pseudoacacia L. as suggested by previous studies [55].

Previous field and simulation studies indicated trees would face a high mortality threat when the APLK was higher than 60%, with this value recommended as the plant hydraulic threshold for predicting drought-induced mortality of trees and widely used in previous studies [53,56,57]. We also adopted this threshold for assessing drought-induced mortality risk for Robinia pseudoacacia L. After the preliminarily tests, we chose beta function to fit the probability distribution of APLK at each site. The probability of an APLK ≥ 60% was then determined with the fitted probability distribution function and defined as drought-induced mortality risk probability (DMRP) for Robinia pseudoacacia L. plantations at each site (Figure 4).

2.5. Statistical Analyses

We calculated the minimum, maximum, and mean values of APLK during the simulation period to characterize the plant hydraulic safety of Robinia pseudoacacia L. at each site. Spatial distribution maps of the minimum, maximum, and mean APLK and DMRP were generated through Kriging interpolation using ArcGIS 10.6 (www.esri.com, accessed on 3 March 2024) to examine the spatial pattern of each indicator. The spatial relationships between these indicators with climatic, soil, and topographic factors were determined by Pearson correlation and stepwise regression analysis.

The predicative accuracy of the BBGC-SPERRY model was evaluated using the coefficient of determination (R²) between simulated and observed values, as well as statistical indicators including mean absolute error (MAE), root mean square error (RMSE), relative root mean square error (rRMSE), and Nash–Sutcliffe efficiency (NSE) [58]. R² was determined by linear analysis, and other indicators were calculated as follows:

$$\mathrm{MAE}={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n|O_{\mathrm{i}}-P_{\mathrm{i}}|}$$

(5)

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{(O_{\mathrm{i}}-P_{\mathrm{i}})}^2}}$$

(6)

$$\mathrm{rRMSE}={\displaystyle \frac{\sqrt{\frac{1}{n}{\displaystyle \sum _{i=1}^n{(O_{\mathrm{i}}-P_{\mathrm{i}})}^2}}}{\frac{1}{n}{\displaystyle \sum _{i=1}^n{O}_{\mathrm{i}}}}}\times 100\%$$

(7)

$$\mathrm{NSE}=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n{(O_{\mathrm{i}}-P_{\mathrm{i}})}^2}}{{\displaystyle \sum _{i=1}^n{(O_{\mathrm{i}}-\overline{O_{\mathrm{i}}})}^2}}}$$

(8)

where O_i and P_i are the observed and simulated values for the ith observation, respectively; ${\overline{O}}_i$ is the average of the observed data; and n is the number of observations. Lower values of MAE, RMSE, and rRMSE indicate higher predicative accuracy. The NSE represents ‘goodness of fit’, with values closer to 1.0 indicating a more accurate model.

3. Results

3.1. Model Evaluation and Uncertainty Analysis

The BBGC-SPERRY model could effectively simulate the spatiotemporal dynamics of transpiration and evapotranspiration for Robinia pseudoacacia L. stands on the Loess Plateau. We collected 57 ET_a data points from 11 sites within the study area, and the observed years spanned from 1982 to 2020. The simulated ET_a was consistent with observed ET_a with an R² value of 0.67 (p < 0.01). The BBGC-SPERRY model was also evaluated in terms of MAE, RMSE, rRMSE, and NSE with respective values of 51.8 mm, 59.6 mm, 12.8%, and 0.62, indicating it exhibited good performance for simulating spatiotemporal variations in ET_a of Robinia pseudoacacia L. stands on the Loess Plateau (Figure 5). At the Yan’an site, a total of 85 data were collected spanning from Day 119 to Day 292 due to the power failure in the field. The data covered the entire growing season and could also be used for evaluating model performance. The simulated stand transpiration dynamics were in good agreement with variations in monitored daily sap flux density during the 2008 growing season (Figure S1), with an R² value of 0.70 (p < 0.01). Stand evapotranspiration and transpiration are the primary processes driving water transport in the SPAC system, as well as representative indicators reflecting responses of vegetation to drought [59]. Therefore, the BBGC-SPERRY model is capable of simulating water transport processes through the SPAC system for Robinia pseudoacacia L. stands.

The sources of simulation uncertainty mainly include model structure, parameters, initial and boundary conditions, etc. [60,61]. Previous studies conducted detailed sensitivity analyses of the parameters in both the Biome-BGC model [62] and the Sperry model [34]. In this study, we selected Yan’an (MAP = 534.5 mm) as a representative site and chose a normal water year (annual precipitation of 510.7 mm in 1984) to estimate the sensitivity of APLK in the BBGC-SPERRY model to a total of 35 parameters, including meteorological, soil, and eco-physiological parameters of Biome-BGC model and Sperry model parameters (Table S1). Readers can refer to Li, et al. [63] and Li, et al. [64] for details on specific steps. The results indicate that APLK is highly sensitive to all meteorological parameters, soil hydraulic parameters except saturated hydraulic conductivity, Sperry model parameters except for average % resistance in rhizosphere, and 7 Biome-BGC model parameters such as specific leaf area, allocation to current growth proportion, etc. In this study, meteorological data were from continuous observations at each simulation site and subjected to standard quality control tests. Soil–water retention curves were estimated using the Arya–Paris model to determine soil hydraulic parameters, which was demonstrated to have high simulation accuracy [65] and was widely applied in eco-hydrological simulations [66,67]. Eco-physiological parameters of Biome-BGC model and Sperry model were determined based on published data in the study area or recommended values for the deciduous broadleaf forest. Previous field survey and controlled experiments indicated that the variability in plant eco-physiological parameters are relatively limited for mature Robinia pseudoacacia L. forests on the Loess Plateau and unlikely to substantially affect the simulation reliability [31,68]. All of these efforts should efficiently contribute to reliable simulations.

3.2. Spatial Pattern of Plant Hydraulic Safety

Plant hydraulic safety in Robinia pseudoacacia L. exhibited strong spatial variability in the study area, as reflected by the spatial variation in APLK (Figure 6). The mean APLK ranged from 0.3% to 29.3% with a coefficient of variation of 55.6% and showed an increasing trend from southeast to northwest. The maximum APLK exhibited a similar spatial pattern to the mean APLK, and ranged from 1.1% to 80.2% with a coefficient of variation of 36.7%. The minimum APLK ranged from 0.1% to 1.9% with a coefficient of variation of 35.6% and exhibited relatively lower values in the west and middle than in the south and north of the study area.

The correlation analysis indicated the spatial distribution of APLK was significantly influenced by climatic, soil, and topographic factors (Figure 7). The minimum, maximum, and mean APLK showed significant negative correlations with MAP, clay content, and silt content and significant positive correlations with sand content. Notably, MAT was negatively correlated with the minimum APLK but positively correlated with the maximum and mean APLK. Due to the negative correlation between MAT and elevation (r = −0.89, p < 0.001), elevation showed a negative correlation with minimum APLK but a positive correlation with maximum and mean APLK. Among the environmental factors, MAP exhibited the highest correlation coefficients with minimum, maximum, and mean APLK, making it an important factor influencing the spatial variability of plant hydraulic safety in Robinia pseudoacacia L.

Stepwise regression analysis further revealed the driving factors of the spatial pattern of APLK (

Table 2. Stepwise regression for the main variables driving the spatial distribution of minimum, maximum, and mean annual average percentage loss of whole-plant hydraulic conductance (APLK) and drought-induced mortality risk probability (DMRP) for Robinia pseudoacacia L.

Variable	Regression Equation	R2	p
Minimum APLK	0.885 − 0.494ELV − 0.543MAP + 0.151SA	0.50	0.0001
Maximum APLK	1.314 − 0.648MAP − 0.177MAT − 0.399CC	0.66	0.0001
Mean APLK	0.696 − 0.794MAP + 0.209SA	0.77	0.0001
DMRP	0.687 − 0.687MAP − 0.300MAT + 0.243SA	0.68	0.0001

Note: all variables were scaled by a min–max normalization method. MAP, mean annual precipitation; MAT, mean annual temperature; CC, clay content; SA, sand content; and ELV, elevation.

). MAP and sand content were the main driving factors for mean APLK, explaining 77% of the spatial variation. In addition to MAP and sand content, the driving factors for minimum APLK also included elevation; 50% of the spatial variation in minimum APLK could be explained by these factors. The spatial variation in maximum APLK was driven by MAP, MAT, and clay content, with 66% of the spatial variability explained by these factors.

3.3. Spatial Pattern of Drought-Induced Mortality Risk

The probability distribution of APLK at each site could be well fitted using the beta function, with an average R² value of 0.94 and average NSE value of 0.93. Based on the probability distribution curve of APLK at each site, the probability corresponding to APLK ≥ 60% could be calculated to determine the DMRP. From the southeast to northwest of the Loess Plateau, the DMRP increased from 0% to 9.2%, with an average value of 1.8% (Figure 8). The spatial distribution of DMRP was positively correlated with elevation and sand content and negatively correlated with MAP, MAT, clay content, and silt content (Figure 7). Stepwise regression analysis indicated the driving factors of DMRP included MAP, MAT, and sand content, which explained 68% of the spatial variability (Table 2).

Based on the intervals DMRP < 2%, 2% ≤ DMRP ≤ 5%, and DMRP > 5%, we classified low-, moderate-, and high-risk zones in the study area, with corresponding recurrence periods of drought-induced mortality for Robinia pseudoacacia L. of >50 years, 20–50 years, and <20 years, respectively. The low-risk zone covered 60.0% of the study area, primarily in the south. A gradual transition to moderate-risk zones then occurred moving towards the north and west and covering 30.7% of the study area. The high-risk zone covered 9.3% of the study area and was located in the northern and western corners.

4. Discussion

Based on the intrinsic specific depictions of physical processes and eco-physiological mechanisms, process-based models can comprehensively quantify the effects of environmental factors on vegetation growth and health at large spatiotemporal scales [69], providing a useful basis for evaluating and enhancing the sustainability of plantation forests [48,70,71]. The BBGC-SPERRY model enables the coupled simulation of plant hydraulic dynamics with water, carbon, and nitrogen cycling processes in an ecosystem, making it an effective tool for investigating climate–soil–vegetation interactions [33,35]. With data for field-observed daily sap flux density and ET_a collected from 11 sites, we verified the reliability of the BBGC-SPERRY model in simulating eco-hydrological processes in Robinia pseudoacacia L. stands at the regional scale of the Loess Plateau (Figure S1 and Figure 5). The BBGC-SPERRY model simulates SPAC water transport based on plant hydraulics, which provides a more mechanistic approach and helps to enhance predictive accuracy compared to traditional scaffolds of empirical response functions [33,34,57]. In the original Biome-BGC model, the Penman–Monteith equation with simulated canopy conductance was coupled with a root water uptake model to describe the responses of plant water use to soil and atmospheric drought [72,73]. The impact of soil drought on canopy conductance and root water uptake was modeled using prescribed empirical stress functions, which lack information about the mechanisms and are difficult to parameterize [34,74]. Recent advances in plant hydraulics significantly deepened our understanding of plant–water relationships [75]. The Sperry model establishes direct hydraulic relationships between water transport in the SPAC with soil and atmospheric drought based on stomatal regulation of plant water supply–demand relationships [34,56]. The model enables the calculation of a series of unmeasurable plant hydraulic traits and the simulation of physiological processes, such as the repair of xylem embolism, hydraulic redistribution at the root–soil interface, etc., which contributes to an improved mechanistic understanding of plant responses to soil and atmospheric droughts [29].

In recent years, with the advancement of plant hydraulic theory, models for predicting drought-induced tree mortality based on hydraulic traits developed rapidly. Although the BBGC-SPERRY model employed in this study effectively describes key plant hydraulic processes and biogeochemical processes at the stand scale, it may still have potential limitations compared with other models. Firstly, plant hydraulic modules were integrated into a range of distributed hydrological models (e.g., ParFlow-TREES [57]) and land surface models (e.g., Noah-MP [12], CoLM [76]) in recent years. Compared with the BBGC-SPERRY model, these models incorporate a broader array of ecohydrological processes, enabling more detailed simulations from watershed to global scales. Secondly, researchers developed predictive methods based on a series of optimality theories related to plant adaptation to water availability [77,78]. These methods predict plant hydraulic, stomatal, and photosynthetic traits across various temporal and spatial scales, which might offer a better approach to account for the spatiotemporal variability of plant parameters. Additionally, vegetation management such as thinning, pruning, and slope engineering measures is also a crucial aspect for sustainable ecological restoration. Recent substantial improvements to the Biome-BGC model (such as the BBGCMuSo [72]) focused on describing a wider range of management practices, which urgently need to be integrated into the BBGC-SPERRY model to enhance analysis of the roles of different management measures in enhancing the sustainability of plantation forests. In future research, comprehensive analysis combining models with different structures and complexities will help better predict the spatiotemporal patterns of drought-induced tree mortality risk under changing climate. There are increasing studies revealing the intrinsic correlations between plant hydraulic traits and other physiological processes, such as biomass repartition [79], root distribution [80], defoliation and damage of living tissues [81], etc., supporting that plant hydraulics are at the heart of plant, crops, and ecosystem functions in the face of climate change [75]. Future research could integrate plant hydraulic traits with a broader range of physiological processes to enhance the simulation accuracy of plants’ response to drought stress.

On the basis of validating the reliability of the BBGC-SPERRY model, we simulated the temporal variation in APLK at each site over a 60-year period and analyzed the dynamics of plant hydraulic safety in Robinia pseudoacacia L. under long-term climatic fluctuations. According to the mechanism of hydraulic failure, prolonged high APLK can lead to a severe reduction in stomatal conductance, photosynthetic rates, and productivity, further affecting cell expansion, membrane permeability, and xylem transport and thereby increasing sensitivity to photothermal stress, diseases, and pests [7,75]. The use of APLK can reflect the long-term plant hydraulic safety status [53,56,57], serving as a crucial indicator for predicting drought-induced tree mortality. Based on the literature analysis by Sperry and Love [56] and Adams, et al. [82], a high threat of drought-induced mortality occurs when the APLK exceeds 60% for a wide variety of tree species. This study also adopted this value as the threshold for predicting the occurrence of drought-induced mortality in Robinia pseudoacacia L. Similar to the use of storm frequency for designing flood control facilities, we calculated the probability of APLK exceeding 60% based on the probability distribution curve of APLK to determine the DMRP of Robinia pseudoacacia L. at each site (Figure 4). The probabilistic analysis provides a holistic understanding of the plant hydraulic safety of Robinia pseudoacacia L., and offers a framework for economic decision making for forestation [83,84]. Additionally, the predicted spatial pattern of drought-induced mortality risk in Robinia pseudoacacia L. in this study is supported by previous field investigations. For example, the main distribution area of “small old trees” on the Loess Plateau identified in field surveys by Hou, et al. [85] is in close agreement with the moderate- and high-risk zones predicted in this study. According to field investigations conducted by Wang and Li [86] and Liu [31] at the Chunhua, Yichuan, Yijun, Fuxian, and Ansai sites, the growth of Robinia pseudoacacia L. gradually weakened with decreasing precipitation but did not lead to mortality; these sites are located in low-risk zones as defined in this study. However, extensive canopy dieback and mortality in Robinia pseudoacacia L. plantations were observed at the Wuqi site, which is located in the moderate-risk zone as defined in this study. At the Huanxian site, also located in the moderate-risk zone, large-scale drought-induced mortality events of Robinia pseudoacacia L. plantations occurred in recent years (personal communication). These field-based findings supported the reliable identification of risk zones in this study.

The spatial variations in plant hydraulic safety and drought-induced mortality risk in Robinia pseudoacacia L. on the Loess Plateau were strongly influenced by climatic, soil, and topographic factors (Figure 7; Table 2). Correlation analysis indicated MAP was one of the most important factors driving the spatial distribution of APLK and DMRP. The Loess Plateau is a typical water-limited region. The depth to groundwater generally ranges from 30 to 100 m, and precipitation is usually the sole water source for most vegetation [27,87]. The spatial distribution of MAP becomes one of the most crucial factors determining vegetation growth and distribution. MAT also exhibited significant correlations with APLK and DMRP, although the correlation coefficients were lower than those with MAP. Specifically, the minimum APLK showed a positive correlation with MAT, which was largely attributed to increases in atmospheric evaporative demand with rising temperature that generally led to higher APLK during wet years. Conversely, the spatial distribution of mean APLK, maximum APLK, and DMRP were predominantly governed by their negative correlations with MAP. Spatially, MAP was negatively correlated with MAT, and could result in the positive correlation of mean APLK, max APLK, and DMRP with MAT. Correlation and regression analyses suggested the spatial distribution of APLK and DMRP were also influenced by soil properties. As clay content decreases and texture tends coarser, soil water holding capacity diminishes and water potential and hydraulic conductivity drop more rapidly with decreasing soil moisture, which in turn makes plant hydraulic structures more susceptible to hydraulic failure and increases mortality risk under drought [39,88,89].

The spatial pattern of drought-induced mortality risk in Robinia pseudoacacia L. evaluated in this study can indicate practical implications for future vegetation construction and sustainable land management. In low-risk zones, forestation with Robinia pseudoacacia L. should be based on the local soil–water carrying capacity for vegetation and the balanced relationship between ecological and socio-economic water use to ensure the sustainability of ecological restoration [27,90,91]. In moderate-risk zones, Robinia pseudoacacia L. plantations should be planted with caution, and enhanced monitoring and management of tree health in existing Robinia pseudoacacia L. plantations is essential to prevent the occurrence of severe drought-induced tree mortality events. Recent field observations and model simulations suggested that the utilization of deep soil water by plants plays a crucial role in ensuring water transport safety and reducing the risk of drought-induced mortality [92,93,94], highlighting the importance of deep soil water management. Implementing slope engineering measures such as infiltration holes [95], fish scale pits [96], and rainwater collection and infiltration systems [97], etc., was reported to efficiently promote deep soil water recharge, thereby reducing the sensitivity of plant hydraulic safety to inter-annual precipitation fluctuations [98]. In high-risk zones, the establishment of Robinia pseudoacacia L. plantations should be avoided, and forestation with other tree species with lower water consumption and stronger drought resistance should be considered a priority for vegetation restoration. For existing Robinia pseudoacacia L. plantations, vegetation type conversion could be considered, and management measures such as pruning and thinning should be duly undertaken to reduce stand water consumption and mitigate the risk of plant hydraulic failure [61,99,100].

This study also faced some methodological limitations that need to be considered in future research. This study conducted long-term continuous simulations using historical meteorological data, which was regarded as an important method for forest management risk assessment [39,83]. However, the results also carry potential spatial and temporal uncertainties. Spatially, the simulation results from 124 meteorological stations were upscaled to the regional level using the Kriging interpolation method, which is computationally efficient but may have limitations in spatial resolution and accuracy. High-resolution simulations using gridded meteorological data are needed to address these limitations. Temporally, under future climate change scenarios, rising temperatures and altered precipitation patterns could increase the drought-induced mortality risk for Robinia pseudoacacia L. plantations on the Loess Plateau [35], which could potentially cause the expansion of the area of medium- and high-risk zones. Model simulations using future climate data under different scenarios will help address these uncertainties. Similar to other eco-hydrological models, the BBGC-SPERRY model obtained steady-state conditions through “Spin-up” as the initial conditions for simulation [42], and did not consider the effects of temporal factors such as developmental stages, stand age, and vegetation succession on drought-induced mortality risk. The BBGC-SPERRY model simulates stand-scale eco-physiological processes, and thus was not able to capture the effects of different densities on drought-induced mortality risk as well as density regulation through self-thinning within stands. According to previous studies, it is suggested that Robinia pseudoacacia L. stands on the Loess Plateau with an age older than 20 years approximately to the “steady-state” condition required by the model [26,101]. The stands used for model evaluation in this study also fall within this stage, and the stand densities ranged from 1000 to 2300 trees ha⁻¹. Therefore, it can be cautiously inferred that, the risk assessment results of this study apply to Robinia pseudoacacia L. stands with ages older than 20 years and densities within 1000–2300 trees ha⁻¹. Currently, most Robinia pseudoacacia L. stands on the Loess Plateau fit this description [102], ensuring the practical applicability of the study’s findings. It can be qualitatively inferred that stands with younger ages or lower densities are likely to have lower drought-induced mortality risk due to better water conditions and vice versa. Future efforts should incorporate population dynamics simulations into the model to enable a more comprehensive assessment of drought-induced mortality risk for stands with varying ages and densities. In addition, root depth coefficient is an important parameter relating to root distribution in the soil profile, and we set it as a universal value (0.99) in this study. Based on our previous field investigations, the root depth coefficient of mature Robinia pseudoacacia L. stands ranged between 0.987 and 0.990 along the precipitation gradient of the Loess Plateau, with no significant differences between sampling sites (p > 0.05) [47]. We selected a value of 0.99 in this study, with a maximum root distribution depth of 5.3 m, to simulate an adequate soil depth profile. According to our sensitivity analysis, when root depth coefficient changed from 0.983 to 0.992 [47], APLK varied from 10.8% to 10.4%, indicating a relatively limited impact. Therefore, this should not substantially affect the reliability of the results. We recommend developing more accurate dynamic root distribution algorithms to reduce uncertainty in further studies.

5. Conclusions

This study utilized the BBGC-SPERRY model to simulate the long-term dynamics of APLK in Robinia pseudoacacia L. and quantitatively assess the spatial pattern of drought-induced mortality risk on the Loess Plateau based on plant hydraulic safety. The BBGC-SPERRY model could accurately simulate the spatiotemporal variation in transpiration and evapotranspiration in Robinia pseudoacacia L. stands on the Loess Plateau. Spatial patterns of plant hydraulic safety and drought-induced mortality risk in Robinia pseudoacacia L. were mainly driven by the precipitation gradient from southeast to northwest, with other climatic, soil, and topographic factors also playing important roles. This study conducted a preliminary assessment of drought-induced mortality risk for Robinia pseudoacacia L. on the Loess Plateau based on the probability distribution of APLK at each site, with low-, moderate-, and high-risk zones accounting for 60.0%, 30.7%, and 9.3% of the study area, respectively. These findings could serve as a basis for the rational forestation and sustainable management of Robinia pseudoacacia L. plantations on the Loess Plateau, and the risk assessment methods of this study can also provide reference for other regions with similar climatic conditions. We recommend that vegetation restoration planning should incorporate the use of multiple models with different structures and complexities to assess drought-induced mortality risks for various species under both historical and future climate conditions. This comprehensive evaluation will contribute to selecting appropriate species that are adaptable to local climate, soil, and topographical conditions for ensuring the sustainability of vegetation restoration efforts. Future efforts should focus on further improvements of the model structure to consider more spatial and temporal factors, thereby facilitating a more accurate assessment of drought-induced mortality risk for plantation forests.