Impact of Plant–Water Interactions on Long-Term Simulations in Deep-Rooted Plantations Using Noah Land Surface Model with Multiparameterization Options (Noah-MP)

Abstract

In water-limited regions, plant–water interactions significantly affect the hydrological cycle and vegetation dynamics, particularly in deep-rooted plantations where deep water uptake mitigates water stress during seasonal and interannual droughts. In this study, we improved the University of Arizona version of the Noah-MP model by incorporating actual soil thickness, along with new subsurface and water table schemes, to evaluate the long-term influence of plant–water interactions in Robinia pseudoacacia L. plantations. We found that soil water content was sensitive to both soil stratification and vertical root distribution, with Nash–Sutcliffe efficiency increasing from less than 0.20 to 0.63 in sensitivity experiments. Plant–water interactions resulted in persistent low soil water content within the root zone, whereas the static vegetation experiment overestimated soil moisture because of unrealistic infiltration. Transpiration and water uptake remained in dynamic equilibrium, and vegetation growth was not limited by water availability. Deep water uptake (>2 m) contributed 0.3–20.5% of transpiration during the growing season, with higher contributions observed in drier years. Compared to precipitation, evapotranspiration was more sensitive to soil water storage in the upper 0–2 m of soil. Our results emphasize the critical role of plant–water interactions in regulating water availability for deep-rooted plantations on the Loess Plateau under changing environmental conditions.

1. Introduction

Vegetation plays a vital role in regulating the terrestrial water cycle because of intrinsic connections between water and vegetation across various spatial and temporal scales [1,2,3]. Water availability controls the distribution and productivity of plant communities in both water-limited and humid environments [4,5]. Conversely, changes in vegetation driven by climate change and land use change influence water availability by regulating evapotranspiration, infiltration, and runoff patterns [6,7,8]. Incorporating realistic plant–water feedbacks into hydrological simulations effectively improves model performance [9,10]. Thus, understanding plant–water interactions is crucial for assessing ecosystem sustainability under changing environmental conditions.

Traditional hydrological models primarily represent vegetation effects through canopy interception and transpiration processes [11,12]. However, these models often employ oversimplified vegetation parameterizations and make excessive assumptions, leading to unrealistic estimations of vegetation impacts—particularly in ecosystems with complex surface characteristics [13]. Moreover, they struggle to capture the temporal dynamics of hydrological responses to vegetation and climate change, as they lack explicit linkages between water availability and plant physiological processes [1,14,15,16]. In contrast, recent developments in dynamic vegetation models and their coupling to land surface models allow for more comprehensive representation of phenology, photosynthesis, respiration, and carbon partitioning [17,18]. These coupled models have demonstrated improved capability in simulating long-term hydrological responses to climate and vegetation changes [19,20,21]. For example, the Noah-MP model effectively captures seasonal variations in evapotranspiration, runoff, and terrestrial water storage across the continental United States [22]. However, many land surface models still exhibit deficiencies in representing drought responses, often underestimating ecosystem resilience because of unrealistically low vegetation productivity and greenness [23]. For instance, CLM4.0 simulated a more rapid decline in the leaf area index than observed by MODIS during drought events [24], and Noah-MP underestimated both evapotranspiration and gross primary productivity in water-limited regions, particularly during the 2011 Texas drought and the 2012 Central United States drought [22,25,26]. A key limitation is static rooting depths and vertical root distributions in simulations, which fail to capture the dynamic interactions between water availability and plant carbon allocation [26].

Root–soil interactions are critical for accurately representing plant–water interactions in land surface models [27,28]. Root systems directly influence soil water dynamics and distribution, which in turn shape root growth and architecture. Both lateral and vertical root distributions exhibit spatial heterogeneity due to hydrotropism [29] and hydropatterning [30], where roots preferentially grow toward wetter soil layers. Plants in various ecosystems, including tropical forests, temperate woodlands, and arid or semi-arid regions, often develop deep root systems or higher root-to-shoot ratios to cope with water stress [31,32,33]. These plants also display flexible water uptake strategies: during dry periods, deep-rooted species increasingly rely on deep soil moisture [31,34,35], whereas under wet conditions, they preferentially utilize shallower water sources [36]. However, static root distributions in land surface models fail to capture such adaptive strategies, leading to unrealistic representations of water fluxes and vegetation growth under drought conditions. To address this limitation, recent advances in the Noah-MP model have incorporated a dynamic root model that enhances the representation of plant–water interactions, particularly in water-limited regions [26,37]

The Chinese Loess Plateau is characterized by a semi-arid to arid climate, with precipitation ranging from 200 mm in the northwest to 800 mm in the southeast [38]. Since 1999, larger-scale afforestation efforts have significantly increased the vegetation cover from 32% in 1999 to 60% in 2013 [39], with the leaf area index rising by 0.01 m² m⁻² and gross primary productivity increasing by 17.8 g C m⁻² from 2000 to 2016 [40]. Ecological plantations in this region predominantly develop deep fine root systems (diameter < 2 mm), enabling deep water uptake to meet transpiration [41,42]. Previous studies have reported that deep soil water serves as a critical buffer against seasonal and interannual droughts [43,44]. Robinia pseudoacacia L. (R. pseudoacacia), a dominant afforestation species in the Loess Plateau, exhibits rapid root growth, extending fine roots at a rate of 1.0 m yr⁻¹ and reaching depths more than 20 m within 15 years [45]. However, its rapid growth has led to deep soil water depletion [46,47]. A significant relationship was found between fine root length and soil water depletion within the 2–10 m soil layers in areas receiving more than 550 mm of annual precipitation [48]. Given the critical role of deep water uptake in sustaining plantation growth, a thorough investigation of its impact on plant–water interactions is essential.

Most previous studies on plant–water interactions in the Loess Plateau have relied on statistical analyses using remote sensing or reanalysis data, often focusing on linear relationships between vegetation indicators (e.g., leaf area index, gross primary productivity, enhanced vegetation index) and hydroclimatic factors (e.g., precipitation, terrestrial water storage) [40,49,50]. Some studies have examined vegetation responses to drought indices [51,52]. However, statistical methods cannot fully quantify plant–water interactions. With advancements in remote sensing and numerical modeling, recent studies have explored the impacts of climate, vegetation, and land use changes on the hydrological cycle using models that incorporate vegetation dynamics [53,54,55,56]. For example, ecological restoration reduced terrestrial water storage by 16.6 ± 5.0 mm yr⁻¹ in the Mu Us regions [53], and vegetation changes increased annual evapotranspiration by 11.2% over the loess Plateau [54]. Yet, most of these models assume shallow soil layers, limiting their ability to represent deep-rooting systems. Furthermore, they often neglect root–soil interactions and deep water uptake, leading to unrealistic estimates of soil moisture and evapotranspiration. Incorporating deep root–soil interactions into model simulations is thus essential for accurately evaluating plant–water interactions, particularly in R. pseudoacacia plantations.

In this study, we improved the Noah-MP model, developed by the University of Arizona, by integrating actual soil thickness as well as new subsurface and water table schemes. This enhanced model was applied to simulate vegetation growth and the hydrological cycle at an observational site within an R. pseudoacacia plantation. The objectives of this study were to (1) assess the sensitivity of soil stratification and vertical root distribution to vegetation growth and soil water dynamics and (2) quantify the impact of plant–water interactions on long-term simulations.

2. Materials and Methods

2.1. Model Description and Improvement

The Noah-MP land surface model was initially developed from the Noah land surface model, incorporating multiple parameterizations to enhance land–atmosphere coupling [17]. The model has evolved to simulate energy, water, and carbon fluxes, as well as their feedback in response to climate change and evaluated CO₂ concentrations [57,58]. The latest version of Noah-MP, developed by the University of Arizona (UA), implements a dynamic root model that explicitly represents root–soil interactions. This enhancement improves simulations of ecosystem productivity and transpiration under water stress, particularly in water-limited regions. In this study, we used the UA version of Noah-MP to better capture plant–water interactions.

2.1.1. Dynamic Vegetation Option

The dynamic vegetation option in Noah-MP describes the terrestrial carbon budget in relation to canopy photosynthesis, respiration, and carbon allocation to roots, stems, leaves, and wood. The default version allocates assimilated carbon preferentially to leaves, stems, and wood, with the remainder allocated to roots. However, the UA version modifies this allocation scheme by prioritizing root systems to meet transpiration demands [26]:

$$f_{root}=0.3(1-\beta )$$

(1)

$$f_{leaf}=(1-f_{root})\times 10000\times {\displaystyle \frac{e^{-2\times LAI}}{1+10000\times e^{-2\times LAI}}}$$

(2)

$$f_{stem}=(1-f_{root}-f_{leaf})\times 0.2$$

(3)

$$f_{wood}=1-f_{root}-f_{leaf}-f_{stem}$$

(4)

where f_root, f_leaf, f_stem, and f_wood are the fraction of the carbon flux that goes into roots, leaves, stems, and woods, β is the water stress factor for canopy photosynthesis and stomatal conductance, and LAI is the leaf area index.

At each time step (∆t), net primary production (NPP) is determined as the difference between gross primary production (GPP) and plant respiration. The turnover rate of all plant tissues (Ovt) and the dying rate of seasonal leaves and stems (Death) are calculated using experience formulas [17]. The plant carbon mass (m) is updated at the end of each time step as follows:

$$m=m+(NPP-Ovt-Death)\times \mathrm{\Delta }t$$

(5)

Based on the updated carbon mass of leaves (m_leaf) and stems (m_stem), the LAI and stem area index (SAI) are updated:

$$LAI=m_{leaf}\times LAPM$$

(6)

$$SAI=m_{stem}\times SAPM$$

(7)

where LAPM and SAPM are the leaf and stem area per unit mass (m² g⁻¹).

2.1.2. Dynamic Root Option

The UA version of Noah-MP incorporates a layered root biomass model to simulate dynamic root water uptake at macroscopic scales [26]. The fine root biomass in the ith soil layer (m_R,i, g m⁻²) is given by:

$${\displaystyle \frac{\mathrm{\Delta }{m}_{R,i}}{\mathrm{\Delta }t}}=GP{P}_{root,i}-E{X}_{root,i}-r{s}_{root,i}-g{s}_{root,i}-Ov{t}_{root,i}$$

(8)

where GPP_root,i (g m⁻²·s⁻¹) represents carbon allocation to fine roots in the ith soil layer, rs_root,i and gs_root,i are the maintenance and growth respiration of fine roots in the ith soil layer (g m⁻²·s⁻¹), EX_root,i (g m⁻²·s⁻¹) is the root exudates in the ith soil layer, and Ovt_root,I (g m⁻²·s⁻¹) is the turnover rate of fines roots in the ith soil layer. The equation for GPP_root,i is:

$$GP{P}_{root,i}=f_{root}\times PSN\times {10}^{-6}\times {\displaystyle \frac{\mathrm{\Delta }{z}_i\times w_i}{{\displaystyle \sum _i^{nroot}(\mathrm{\Delta }{z}_i\times w_i)}}}$$

(9)

$$w_i=e^{-\mathrm{a}{z}_i}\times \sqrt{{\displaystyle \frac{{\theta }_i-{\theta }_{wilt}}{{\theta }_{ref}-{\theta }_{wilt}}}}$$

(10)

where PSN (μmol m⁻²·s⁻¹) is the total photosynthetic rates from sunlit and shaded leaves, w_i is the weighting factor in the ith soil layer, z_i is the node depth in the ith soil layer (m), θ_i, θ_ref, and θ_wilt (cm³ cm⁻³) are the actual soil moisture, field capacity, and wilting point in the ith soil layer, ∆z_i (m) is the layer thickness in the ith soil layer, and a is an empirical parameter.

Water uptake by fine roots in each soil layer is related to the predicted root surface areas (A_R, m²·m⁻²), which are converted from root biomass. The change in plant water storage (M_q, mm) is related to the total water uptake (Q_R, mm·s⁻¹) and plant transpiration (Q_T, mm·s⁻¹):

$${\displaystyle \frac{\mathrm{\Delta }{M}_q}{\mathrm{\Delta }t}}={\displaystyle \sum _1^{nroot}{Q}_{R,i}}-Q_T$$

(11)

$$Q_{R,i}=A_{R,i}\left({\displaystyle \frac{h_{s,i}-h_R}{{\Omega }_R+{\Omega }_{s,i}}}\right)$$

(12)

where nroot is the total number of soil layers containing fine roots for a specific plant functional type, h_R and h_s,i, (mm) are the water potential of fine roots and soil water near the root in the ith soil layer, and Ω_R (s) and Ω_s,i (s) are the resistance to water flows through fine roots and soil matrix near the root surface in the ith soil layer.

Another improvement is that β is specified as a function of plant water storage:

$$\beta ={\displaystyle \frac{M_q-M_{q,wilt}}{M_{q,\max }-M_{q,wilt}}}$$

(13)

where M_q,wilt (mm) is plant water storage at the wilting point and M_q,max (mm) is the maximum plant water storage. The value of β ranges from 0 to 1, suggesting a maximum fraction of 30% for carbon allocation to fine roots under severe water stress. As β approaches 0, M_q approaches M_q,wilt, resulting in stomatal closure to prevent severe desiccation.

2.1.3. Soil Moisture Loss Without Dynamic Vegetation

When the dynamic vegetation option is disabled, the Noah-MP model utilizes a prescribed monthly leaf area index. Soil moisture loss through plant transpiration (Q_soil,transp, mm) is calculated as:

$$Q_{soil,transp}(i)=Q_{can,transp}\times F_{soil,transp}(i)$$

(14)

where i is the number of soil layers containing fine roots, Q_can,transp is the total amount of transpiration (mm), and F_soil,transp is the proportion of Q_can,transp in the ith soil layer. F_soil,transp is calculated as:

$$F_{soil,transp}(i)={\displaystyle \frac{r_i\times av{i}_i}{{\displaystyle \sum _1^{nroot}{r}_i\times av{i}_i}}}$$

(15)

$$av{i}_i={\displaystyle \frac{{\theta }_i-{\theta }_{wilt}}{{\theta }_{ref}-{\theta }_{wilt}}}$$

(16)

where r_i is the proportion of fine roots in the ith soil layer and avi_i is the water availability for transpiration in the ith soil layer.

2.1.4. Model Improvement

The default soil profile in Noah-MP has a depth of 2 m, divided into four layers with thicknesses of 0.1, 0.3, 0.6, and 1.0 m from surface to bottom. This configuration restricts root depth and excludes deep soil water uptake. To better represent deep-rooted plantations, we extended the soil depth to approximately 200 m. The number of hydrologically active soil layers (N_bedrock) was restricted to the actual soil thickness (Figure 1).

R. pseudoacacia develops fine roots beyond 10 m, with some extending deeper than 20 m (Figure 2a) [34,35,42,59]. About 50% of fine roots (D₅₀) were distributed in the top 3 m, and the soil depth corresponding to 95% of fine roots (D₉₅) ranged from 9.4 m to 20.4 m across different sites. The mean D₉₅ value of 12.4 m was adopted as the effective rooting depth in the model. In the dynamic root option, vertical carbon allocation is controlled by parameter a in Equation (10) (Figure 2b), which determines the exponential distribution of fine roots. Smaller values of a lead to deeper root penetration, while larger values allocate more biomass to shallow layers.

When bedrock is introduced into the soil structure, the subsurface runoff options in the Noah-MP model may be unsuitable. We incorporated the subsurface runoff scheme and water table scheme from CLM 5.0 into Noah-MP to calculate lateral subsurface runoff and water table depth. When a saturated soil layer exists within the soil column, the model begins to calculate total lateral subsurface runoff:

$$q_{drain}={\Theta }_{\mathrm{ice}}{K}_{baseflow}\tan (\beta ){(z_{bedrock}-z_{\mathrm{\Delta }})}^{N_{baseflow}}$$

(17)

where K_baseflow is an adjustable parameter, β is the topographic slope, N_baseflow equals 1, Θ_ice is the ice impedance factor, and z_∆ is the water table depth, determined by the first soil layer above the bedrock, where the ratio of actual to saturated soil water content is more than 0.90.

2.2. Study Area and Data

2.2.1. Study Area

The study area is located in the southern Loess Plateau (34.5° N, 107.9° E), where afforestation efforts have resulted in over 20% of the area being planted (Figure 3). The dominant plantation species, R. pseudoacacia, was planted in 2000 [60]. From 1980 to 2020, the annual average precipitation and temperature at Shouyang station were about 577.3 mm and 11.4 °C, respectively, indicating a semi-humid climate. The loess thickness at this site is around 130 m [61], while the depth to the water table ranges from 30 to 80 m.

2.2.2. Data

We used the 3-h China Meteorological Forcing Data (CMFD), including air temperature, downward shortwave radiation, downward longwave radiation, surface pressure, specific humidity, precipitation, and wind speed [63,64]. The CMFD data were validated against in situ observations from the China Meteorological Administration. Between 1980 and 2020, the daily air temperature, surface pressure, specific humidity, and precipitation matched observations well, with coefficients of determination (R²) of 0.99, 0.96, 0.98, and 0.61, respectively. The annual average precipitation in the CMFD closely matched the observed value (579.6 mm vs. 577.3 mm).

In the model simulation, vegetation type was set to broadleaf forest, with default vegetation-type-dependent parameters applied. The soil texture within the upper 0–5 m was silt loam. Soil-type-dependent parameters were sourced from the China dataset of soil hydraulic parameters [65], including saturated soil moisture, field capacity, wilting point, residual soil moisture, saturated hydraulic conductivity, saturated matrix potential, and parameters for the soil water retention curve. Vegetation and soil parameters were fine-tuned to better align with the observed soil water content, vertical root fraction, and leaf area index from published work [60].

2.3. Model Experiments

We conducted two sets of model experiments to explore the sensitivity of soil stratification and root distribution (parameter a in Equation (10)) to soil hydrology and vegetation growth: (1) Soil stratification schemes. Two soil stratification schemes were tested (

Table 1. Soil stratification schemes in the sensitivity experiments.

Scheme 1			Scheme 2
Sequence	Thickness	Total Thickness	Sequence	Thickness	Total Thickness
1–124	0.1	12.4	1–2	0.05	0.10
125–147	0.2	17.0	3	0.06	0.16
148–154	0.4	19.8	4	0.10	0.26
155–156	0.6	21.0	5	0.14	0.40
157–166	0.8	29.0	6–28	0.20	5.0
167–184	1.0	47.0	29–68	0.40	21.0
185–194	1.2	59.0	69	0.60	21.6
195–204	1.4	73.0	70	0.80	22.4
205–214	1.6	89.0	71–81	1.00	33.4
215–234	1.8	125.0	82–91	1.20	45.4
235–239	2	135.0	92–95	1.40	51.0
-	-	-	96–100	1.60	59.0
-	-	-	101–105	1.80	68.0
-	-	-	106–138	2.00	134.0

). In Scheme 1, the root zone was divided into uniform 10 cm layers, with layer thickness increasing below the root zone. In Scheme 2, the soil layer thickness gradually increased from the surface to the bottom. This approach, adapted from previous studies [66,67], avoided the occurrence of wet–dry oscillations caused by larger soil layer thickness. In both schemes, the maximum thickness of any individual layer was less than 2 m. (2) Parameter a Schemes. We tested seven values for parameter a (0.1, 0.2, 0.3, 0.5, 0.7, 1.0, and 3.0 m), which controls the vertical distribution of fine roots. The physical parameterization options selected for this study are shown in

Table 2. Noah-MP parameterization options selected.

Parameterization Options	Used in This Study
Dynamic vegetation	2: Leaf area predicted
Dynamic root	1: Dynamic root
Canopy stomatal resistance	1: Ball–Berry
Surface runoff	3: Original surface runoff
Supercooled liquid water	1: No interaction
Frozen soil permeability	1: Linear effect
Radiation transfer	1: Modified two-stream
Snow surface albedo	2: CLASS
Snow–rain partitioning	1: Jordan91
Soil water retention	1: Van Genutchen
Snow/soil temperature time scheme	1: Semi-implicit

. All experiments were conducted from 1 January 2014 to 31 December 2018, with 30 model spin-up iterations to ensure model equilibrium. Only the output from the final iteration was analyzed. Model performance was assessed by comparing the simulated soil water content, leaf area index, and root distribution against observations.

Following the sensitivity experiments, the optimal soil stratification scheme and parameter a were selected for long-term simulations. To investigate the importance of plant–water interactions, we conducted two additional experiments: one with dynamic vegetation disabled (“nodynamic”) and one with it enabled (“dynamic”). The differences in the results between the two simulations reflected the effect of plant–water interactions. These experiments were conducted from 1 January 1980 to 31 December 2020, driven by the CMFD forcing data and seven spin-up loops. Outputs from the seventh loop were used for analysis.

2.4. Model Evaluation

In the sensitivity experiments, we used three quantitative statistics to evaluate model performance, including the percent bias (pbias), root mean square error (RMSE), and Nash–Sutcliffe efficiency (NSE). The equations for these metrics are as follows:

$$pbias={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n(x_i-y_i)}}{{\displaystyle {\sum }_{i=1}^n{y}_i}}}\times 100\%$$

(18)

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

(19)

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

(20)

where x and y are the simulated and observed values at time step i, respectively; y is the mean of the observed values at time step i, respectively; and n is the total number of time steps. In this study, the model simulation could be judged as satisfactory if pbias ≤ ±25% and NSE ≥ 0.50 [68].

3. Results

3.1. Sensitivity Experiments

3.1.1. Soil Stratification Schemes

As shown in Figure 4a, the simulated soil water content was more sensitive to the soil stratification schemes. In Scheme 1, the simulated soil water content in the upper 5 m closely aligned with the observed values, yielding pbias, NSE, and RMSE values of −0.83%, 0.63, and 0.01 cm³-cm⁻³, respectively. In contrast, Scheme 2 overestimated the soil water content from June 2015 to December 2018, resulting in a substantially lower NSE of 0.15. The simulated leaf area index and root distribution demonstrated low sensitivity to the soil stratification schemes, showing minimal differences between Scheme 1 and Scheme 2. The simulated root distribution in both schemes closely aligned with the observations, yielding NSE values up to 0.86 and RMSE values less than 4.0% (Figure 4d). The NSE for the leaf area index ranged from 0.59 to 0.62, and pbias ranged from 1.4% to 6.8%. Given the relatively short observation period for the leaf area index, we further compared the simulated leaf area index with the MODIS-derived data from 2014 to 2018 (Figure 4c). Overall, the simulated, observed, and MODIS-derived leaf area indices followed similar seasonal patterns. However, the MODIS leaf area index was much higher during the wet seasons of 2014 and 2015. Previous studies also reported that the MODIS leaf area index tends to be higher than field observations during wet periods [69].

3.1.2. Parameter a Schemes

Parameter a had a great impact on the simulated soil water content (Figure 5a). Increasing a led to a higher soil water content, improving the match with the observed values. Across the tested range of a values (0.1–3.0), the model performance metrics for soil water content were as follows: NSE from −0.43 to 0.63, pbias from −0.83% to −5.4%, and RMSE from 0.01 to 0.02 cm³ cm⁻³. Root fraction was strongly affected by parameter a (Figure 5d). Smaller a values resulted in a more uniform root distribution with less allocation to shallow soil layers. As parameter a increased from 0.1 to 3.0, the cumulative root fraction within the upper 2 m of soil increased from 49.7% to 98.6%, approaching the observed value of 90%. The corresponding NSE values ranged from −0.08 to 0.86, and the RMSE values ranged from 10.1% to 3.6%. In contrast, this parameter had a minor effect on the simulated leaf area index (Figure 5b). The RMSE for the leaf area index ranged from 0.51 to 0.63 m² m⁻², NSE from 0.59 to 0.73, and pbias from 1.4% to 8.0%.

3.2. The Effect of Plant–Water Interactions on Long-Term Simulation

Based on the sensitivity experiments, we evaluated the long-term net effects of plant–water interactions on plant growth and hydrological processes. In the dynamic experiment, the simulated leaf area index fluctuated with interannual climatic variability, ranging from 2.1 to 2.9 m² m⁻² during the growing season (Figure 6a). In contrast, the nodynamic experiment maintained a constant leaf area index of 2.5 m² m⁻² in the growing season and 0.3 m² m⁻² in the non-growing season.

During the growing season, the dynamic experiment produced an average monthly evapotranspiration of approximately 71.8 mm—about 3.9% lower than in the nodynamic experiment (Figure 7a). Despite the higher leaf area index in the dynamic experiment, the monthly vegetation transpiration was about 18.0% lower (38.2 mm), while the monthly soil evaporation increased by approximately 22.0% (26.4 mm). The contribution of vegetation transpiration to total evapotranspiration was 53.3% in the dynamic experiment compared to 60.5% in the nodynamic experiment. Meanwhile, the soil evaporation ratio was higher in the dynamic experiment (36.8%) than in the nodynamic experiment (27.6%). These results indicate that plant–water interactions altered evapotranspiration partitioning, reducing the contribution of vegetation transpiration and canopy evaporation while increasing the soil evaporation. In the non-growing season, the monthly evapotranspiration was lower in both experiments, averaging below 20 mm. The monthly soil evaporation in the dynamic experiment was about 17.1% higher than in the nodynamic experiment.

On average, the ratio of vegetation transpiration to precipitation during the growing season was 49.4% in the dynamic experiment and 58.3% in the nodynamic experiment. At the interannual timescale, this ratio decreased linearly with increasing precipitation, with R² values of 0.71 and 0.79 in the dynamic and nodynamic experiments, respectively. In dry years, the ratios were higher—66.2% (dynamic) and 85.8% (nodynamic)—whereas in wet years, they declined to 33.8% and 37.4%, respectively. These results suggest that omitting plant–water interactions led to the overestimation of vegetation water use during drier periods.

In the upper 5 m of soil, the dynamic experiment generally exhibited lower soil water storage, with the depth of the infiltration wetting front shallower than 5 m during most simulation periods (Figure 8a). Notably, when the monthly precipitation exceeded 100 mm, such as in 1981, 1983, and 1984, the wetting front in the dynamic experiment extended beyond 5 m. In contrast, the wetting front in the nodynamic experiment frequently reached depths greater than 5 m during the growing season (Figure 8b). Consequently, the dynamic experiment revealed a distinct zone of low soil water content in the root zone (Figure 8a). We found larger differences in the soil water storage between the two experiments in the root zone, especially in the 5–12.4 m of soil (Figure 8c). The fluctuations in the differences declined with soil depth.

4. Discussion

4.1. How to Influence the Soil Water Content in the Soil Profile

Root distribution is crucial for accurately modeling hydrological processes and vegetation growth, particularly for deep-rooted species [29,70]. The sensitivity experiments highlighted the importance of soil stratification and root distribution in accurately simulating the soil water content. Similarly, a previous study also found the importance of vertical discretization in the soil column for the soil water content [66]. In the former experiment, the depth of the infiltration wetting front was less than 5 m in both Scheme 1 and Scheme 2. However, Scheme 2 had higher infiltration and lower surface runoff, leading to a higher soil water content in the upper 5 m. Consequently, the annual transpiration under Scheme 2 (239.7–301.4 mm) was consistently higher than that of Scheme 1, with increases ranging from 3.8% to 16.3%. In the deeper 5–15 m, the soil water content was lower because of deep water uptake, with Scheme 1 showing relatively lower values because of greater water uptake. Our findings suggest that the improved model performed well in simulating the soil water content when the root zone was uniformly divided into 0.1 m soil layers.

In the latter experiment, the differences in the soil water content were primarily driven by variability in root water uptake (Figure 9). As parameter a increased, water uptake in the deeper layers (5–12.4 m) declined, while water uptake within the upper 5 m became a non-linear function of parameter a. Lower values of parameter a lead to greater water uptake in deeper soil layers. Previous studies have shown that R. pseudoacacia exhibits an exponential root distribution pattern [34,35,42,59]. In this study, 90% of the fine roots were concentrated within the upper 2 m of soil, indicating that a relatively large parameter value should be used in model simulations. The annual average transpiration increased from 260.2 mm to 268.6 mm as parameter a rose to 0.5. However, when a exceeded 0.5, the annual average transpiration began to decline, reaching a minimum of 243.9 mm (Figure 9e).

A previous study found that about 81% of the water lost through transpiration by R. pseudoacacia was sourced from the top 0–2 m of soil during the growing season [35]. Therefore, accurately simulating the soil water content in the upper 5 m is sufficient to capture the major influence of root water uptake. Although the improved Noah-MP model extended the hydrological active soil profile to ~130 m, the physical realism of water movement at such depths requires further discussion. Groundwater tables in the Loess Plateau are typically very deep—often exceeding 50 m—making direct groundwater uptake by plant roots highly unlikely. As such, our simulations focused on the unsaturated zone in the upper 20 m, which serves as a long-term water reservoir recharged by episodic infiltration events. Compared to the default drainage scheme, the improved model showed larger differences in soil moisture in the deep 100–130 m layers. Under the default scheme, excessive subsurface runoff leads to a persistently unsaturated profile, highlighting the absence of a groundwater table in the simulation. A previous study that incorporated actual soil thickness into CLM4.5 at the Changwu station found that only monthly precipitation events exceeding 230 mm could recharge soil water deeper than 75 m in deep-rooted apple trees [67]. In our simulations, only five months during the study period had precipitation exceeding 200 mm, suggesting that groundwater recharge is rare. Therefore, the contribution of groundwater via capillary rise is negligible in this region and was not considered in the model. Future work should integrate in situ observations of deep soil moisture across the entire profile and investigate the coupling between capillary fluxes, water table dynamics, and deep soil water availability to improve the physical representation of deep soil processes in land surface models.

4.2. Relationship Between the Leaf Area Index and Vegetation Transpiration

Vegetation transpiration is directly related to canopy conductance, which is positively correlated with the leaf area index, indicating that the leaf area index directly affects transpiration [71,72]. To further understand the importance of plant–water interactions, we compared the relationship between the leaf area index and transpiration in both the dynamic and nodynamic experiments (Figure 10). We observed a significant positive correlation between the leaf area index and transpiration at monthly scales, with R² values greater than 0.80 (p < 0.05). Similar results have been reported in previous studies, although this relationship is constrained by a threshold in the leaf area index [72,73,74]. When the leaf area index exceeds this threshold, transpiration may either stabilize or decrease. Moreover, the threshold value varies across different vegetation types and regions, influenced by many factors such as soil moisture, nutrients, light, and climatic conditions [72,73,74]. In the nododynamic experiment, a wide range of transpiration values corresponded to the same leaf area index at the monthly scale. However, transpiration negatively correlated with the leaf area index in the nododynamic experiment at the annual scale. In contrast, the relationship between the leaf area index and transpiration in the dynamic experiment more closely reflected the actual conditions. Specifically, when the simulated leaf area index was below 4.0 at the monthly scale, transpiration increased alongside the leaf area index.

Plant–water interactions significantly influence transpiration at both timescales, as evidenced by the varying ratios of dynamic to nodynamic (Figure 10). A linear relationship between transpiration and the leaf area index was evident at both timescales, consistent with previous studies [74,75]. However, the deviation from linearity was more pronounced at the annual scale when the leaf area index was either greater or smaller in the dynamic experiment, indicating that transpiration was lower in the dynamic experiment. We also observed some deviation from a 1:1 relationship at the monthly scale, particularly the dynamic-to-nodynamic ratio of transpiration near 1.0 (Figure 10d). This deviation may be attributed to soil water limitations [75].

Previous studies have also revealed that vegetation dynamics significantly influence hydrological processes [55,76]. For instance, vegetation dynamics in the improved CLM4.0 produced higher leaf area index and evapotranspiration but lower runoff and root zone soil water in the Wuding river basin [55]. Similarly, CLM5.0 produced a high leaf area index in summer and autumn across most of the Three Rivers Source Region of the Yellow River [76]. The UA version of Noah-MP outperformed the multi-model ensemble mean in capturing interannual variations and long-term trends in the leaf area index in drylands [77]. Although several land surface models include vegetation dynamics, models such as ORCHIDEE and LPJ-GUESS often simulate unrealistic hydrological behavior in dryland ecosystems because of their use of simplified bucket-type hydrological schemes. When a discretized soil hydrology scheme was implemented in the ORCHIDEE model, the simulation of daily and monthly evapotranspiration in semi-arid regions improved; however, the transpiration-to-evapotranspiration ratio remained lower than observed values [78].

4.3. Relationship Between Root Water Uptake and Vegetation Transpiration

In water-limited regions, the balance between water supply and demand determines plant survival under water stress conditions. During wet seasons, transpiration is mainly sustained by current precipitation, whereas in dry seasons, it predominantly relies on precipitation from prior months or seasons [79,80,81]. The relationship between water uptake and transpiration reflects how plants regulate water use in response to drought stress. Our findings revealed a significant positive correlation between transpiration and water uptake at both monthly and annual scales, aligning with previous studies [82,83]. Over the long-term simulation period, the dynamic experiment demonstrated a stable balance between transpiration and root water uptake, suggesting that water availability did not limit vegetation growth from 1980 to 2020 (Figure 11a).

During the growing season, approximately 92% (78.5–101.1%) of the transpiration was derived from root water uptake within the upper 2 m of soil (Figure 11b). This finding is consistent with a previous study that observed that transpiration in apple trees and R. pseudoacacia mainly came from shallow soil (0–2 m), contributing about 68% and 81% of the total transpiration [35]. Observations from apple and peach orchards further indicated that approximately 95% and 97% of the transpired water originated from newly available water (i.e., recent precipitation), with a smaller portion coming from pre-existing soil moisture (water age more than planting years) [62]. Deep water uptake (>2 m) contributed 0.3–20.5% to the transpiration in the growing season, particularly higher in dry years (Figure 11b). This aligns with previous research indicating that deep-rooted plants extract more water from deeper soil layers to sustain transpiration during drought periods [84,85]. Additionally, the ratio of transpiration to total evapotranspiration was 53% in the dynamic experiment and 61% in the nodynamic experiment. Prior studies have reported transpiration-to-evapotranspiration ratios for broad-leaved forests ranging from 43% to 79% [72,86,87], further supporting the validity of our findings.

The regulatory role of plant–water interactions on long-term water balance is likely to be broadly applicable, especially in areas characterized by deep-rooted vegetation, thick loess deposits, and semi-humid climatic conditions. However, the Loess Plateau exhibits considerable spatial heterogeneity in terms of climate, soil properties, and vegetation structure, all of which can influence plant water use strategies and hydrological responses. Hence, we recommend that future studies conduct long-term monitoring and simulations across diverse regions to further validate our findings and improve the regional applicability of the model.

4.4. Evapotranspiration, Soil Water Storage, and Precipitation

The sensitivity of evapotranspiration to soil moisture is a core parameter in the land–atmosphere system. Numerous studies have explored the response of evapotranspiration to soil moisture [88,89], the effect of precipitation on evapotranspiration and soil moisture [90], and the interaction between soil moisture and hydro-meteorological factors across different vegetation types [91]. In this study, we found a significant positive correlation between evapotranspiration and soil water storage in the 0–2 m layer during the growing season (R = 0.50, p < 0.05) and at the annual scale (R = 0.58, p < 0.05) (Figure 12). This finding aligns with previous research showing strong positive correlations between soil moisture and evapotranspiration in various vegetation types, with longer evapotranspiration decay times in dryland or forested ecosystems due to deep root systems [89,90]. In dryland ecosystems, soil moisture is the primary factor influencing evapotranspiration. However, during prolonged droughts, when shallow soil layers dry out, evapotranspiration may be affected by climatic variables or a combination of multiple factors [92]. For example, in regions with low interannual variability in evapotranspiration (averaging 363 mm·yr⁻¹), evapotranspiration is influenced by soil moisture, solar radiation, air temperature, and vapor pressure deficit [93]. However, evapotranspiration exhibited no significant correlation with precipitation in this study. This contrasts with findings from a previous study reporting strong correlations between evapotranspiration and precipitation in grasses and shrubs of arid regions [88]. This may be attributed to the influence of initial soil moisture conditions at the beginning of each year, which could lead to a decoupling between precipitation and evapotranspiration dynamics.

5. Conclusions

In this study, we used the latest version of Noah-MP to evaluate the impact of plant–water interactions on hydrological processes in the R. pseudoacacia plantation located in the southern Chinese Loess Plateau. By introducing actual soil thickness and new subsurface and water table schemes, we evaluated the sensitivity of the leaf area index, soil water content, vertical root fraction to soil stratification schemes, and vertical carbon allocation determined by parameter a. We then quantified plant–water interactions by comparing simulations with dynamic and static vegetation representations.

Our sensitivity experiments underscored the critical role of soil stratification schemes in regulating soil water dynamics. Dividing the root zone uniformly into 0.1 m soil layers yielded the most realistic soil water content because of lower infiltration and higher surface runoff. The soil water content and root fraction were particularly sensitive to vertical carbon allocation. As parameter a increased from 0.1 to 3.0, the root profiles transitioned from uniform to exponential distributions, leading to better agreement with observations. As a result, lower a values exhibited higher root water uptake from deeper soil layers (>2 m). Overall, the improved model enhanced the simulation accuracy of the soil water content and vertical root fraction.

The dynamic experiment revealed a more realistic relationship between the leaf area index and transpiration. Plant–water interactions led to lower soil moisture in the root zone as deep roots extracted soil water to meet transpiration. Shallow root water uptake (<2 m) contributed 92% of the total transpiration, while deep water uptake (>2 m) accounted for 0.3–20.5% of the transpiration during the growing season, with higher contributions in dry years. The simulated transpiration and water uptake remained in a dynamic equilibrium, fluctuating around the long-term average, indicating that water availability did not limit vegetation growth between 1980 and 2020. The simulated evapotranspiration was significantly influenced by soil water storage within the upper 2 m during the growing season and at the annual scale. However, an insignificant relationship was found between precipitation and evapotranspiration. Excluding plant–water interactions led to unrealistically deep infiltration in the soil profile. In this case, rainfall infiltration could replenish soil moisture below 5 m, leading to an elevated soil water content, 3.9% higher evapotranspiration, and 18.0% greater transpiration compared to the dynamic experiment. These findings highlight the crucial role of plant–water interactions in regulating water availability for extensive plantation areas with deep-rooted vegetation across the Loess Plateau, particularly under changing environmental conditions.

It is important to note that the conclusions drawn in this study are based on observations from a single site in the southern Loess Plateau, which may limit the spatial representativeness of the findings. Given the substantial heterogeneity in climate regimes, soil properties, and vegetation restoration practices across the Loess Plateau, further research is needed to assess the generalizability of the findings to broader areas. Comparative studies at multiple sites with varying environmental conditions would help to verify the robustness of the modeled plant–water interactions and support more region-wide ecological restoration strategies.