BMI routines are commonly organized into three phases (initialize, update, finalize) as described in Fig. 2 of Hughes et al. (2022). Figure 1 shows the coupling framework of STEMMUS-SCOPE-MODFLOW 6.

The initialization phase in a BMI setup is separate from the initialization routines associated with each of the individual models (explained in Sect. 2.4.2). During the BMI initialization routine, “hooks” are established between the coupled models to facilitate the exchange of calculated values during a forward run. During the initialize phase, initialized head (i.e., GWL) and temperature (i.e., GT) of the water table are defined in STEMMUS-SCOPE. Whereas, initialized groundwater net recharge (Rn), heat flux above the water table (Qheat), and net recharge temperature (RTn) are defined in MODFLOW 6.

The update phase is at the core of the coupling process (Fig. 1b). During this phase, the two models exchange variable values at the phreatic surface (the shared interface between the unsaturated zone and the saturated zone, marked by a bold blue line in the diagram of Fig. 1a). At this shared interface, STEMMUS-SCOPE requires the following bottom boundary conditions: (a) specified head (i.e. GWL) for the soil water module; and (b) specified temperature (i.e. GT) for the soil energy module; together to simulate groundwater gross recharge (Rg), capillary rise flux (Cap), root groundwater uptake (RWUg), Qheat, and RTn. In turn, MODFLOW 6 requires the following upper boundary conditions: (a) specified water flux (i.e. Rn, calculated as the difference between Rg, Cap, and RWUg) for the GWF module, and (b) specified heat flux (i.e., Qheat) and/or specified temperature (i.e., RTn) for the GWE module; together to simulate GWL and GT. From herein, the simulated variables (Rn, Qheat, and RTn) by STEMMUS-SCOPE are referred to as “STEMMUS-SCOPE variables” (marked by red colour in the flowchart of Fig. 1b), while the simulated variables by MODFLOW 6 (GWL and GT) are referred to as “MODFLOW 6 variables” (marked by blue colour in the flowchart of Fig. 1b).

The update phase runs in a looped manner (update loop), with number of updates equal to number of models' time steps. In principle, the BMI can work with different temporal discretization (number of time steps and the length of each time step) for each of the two coupled models. However, for ease of use, the same temporal discretization was used for both models in this study. For each time step: (1) MODFLOW 6 runs and updates the MODFLOW 6 variables; (2) the updated MODFLOW 6 variables are passed to STEMMUS-SCOPE (illustrated by the blue bold arrow in Fig. 1b), which are used as bottom boundary conditions by STEMMUS-SCOPE; (3) STEMMUS-SCOPE runs and updates the STEMMUS-SCOPE variables; (4) the updated STEMMUS-SCOPE variables are passed back to MODFLOW 6 (illustrated by the red bold arrow in Fig. 1b) to be used as upper boundary conditions for the next time step; and (5) updates continue until the last time step.

In this phase, both models finalize their simulation, and the model outputs are exported for further analysis. The model outputs include the STEMMUS-SCOPE variables, the MODFLOW 6 variables, and other variables of interest (marked by black colour in Fig. 1) as follows: θ, Ts, root soil water uptake (RWUs), subsurface evaporation (Ess), ET, which is the summation of Ess, RWUs and RWUg, GPP, NEE, and SIF.

MODFLOW 6 was not originally written to support BMI; however, Hughes et al. (2022) documented the restructuring of the source code to align MODFLOW 6 to support the BMI conventions. Similarly, Schilperoort (2025) and Zeng et al. (2025) documented the steps followed to make STEMMUS-SCOPE BMI-compatible. In this study, a groundwater module is also added into the source code of STEMMUS-SCOPE to: (a) recognize the effect of the saturated zone dynamics on the unsaturated zone through the MODFLOW 6 variables; and (b) prepare the necessary STEMMUS-SCOPE variables that are fed into MODFLOW 6. The groundwater module includes three main parts: (1) unsaturated zone bottom boundary; (2) recharge and capillary rise flux; and (3) root groundwater uptake.

The bottom boundary of the unsaturated zone for the soil water module in STEMMUS-SCOPE can be defined using one of three options (Zeng and Su, 2013): (a) specified matric head (Dirichlet) boundary; (b) specified moisture flux; and (c) zero matric head (i.e. gravity drainage). In this study, option (a) is used. This bottom boundary (specified matric head) is adaptively positioned at the phreatic surface (Fig. 1a), corresponding to the GWL at each time step (Zeng et al., 2019a). The GWL value is received by STEMMUS-SCOPE from MODFLOW 6 through the BMI (Fig. 1b). Then, at the phreatic surface elevation (GWL), the specified matric head is set to zero (matric head is negative above GWL and positive below GWL). At each time step, the GWL may rise or fall, causing the phreatic surface to shift upward or downward. As a result, the thickness of the unsaturated zone decreases or increases, and the position of the bottom boundary is updated to reflect this moving boundary condition.

Similarly, for the soil energy module, a specified temperature boundary condition is used as the bottom boundary. The value of the specified temperature is the GT (Fig. 1b), received by STEMMUS-SCOPE from MODFLOW 6 through the BMI and updated at each time step.

STEMMUS-SCOPE simulates two-phase mass and heat flow in the unsaturated zone using the modified Richards equation (Milly, 1982). The water flow is driven by both gravity and capillary forces (Eq. 1). 1δδtρLθL+ρvθv=-δδzqLh+qLT+qvh+qvT-S=ρLδδzKδψδz+1+DTaδTδz+δδzDvhδψδz+DvTδTδz-S where ρL and ρv (kg m⁻³) are the density of liquid water and water vapour, respectively; θL and θv (m³ m⁻³) are the soil liquid and vapour volumetric water content, respectively; qLh and qLT (kg m⁻² s⁻¹) are the soil liquid water fluxes driven by the soil matric potential gradient δψδz and temperature gradient δTδz, respectively; qvh and qvT (kg m⁻² s⁻¹) are the soil water vapour fluxes driven by the δψδz gradient and the δTδz gradient, respectively; K is the unsaturated hydraulic conductivity (m s⁻¹); DTa (m² s⁻¹ K⁻¹) is the transport coefficient of the adsorbed liquid flow due to temperature gradient; Dvh (kg m⁻² s⁻¹) is the isothermal vapour conductivity; DvT (kg m⁻¹ s⁻¹ K⁻¹) is the thermal vapour diffusion coefficient, and S is the sink term (m s⁻¹). More details of the equation are in Zeng et al. (2011a, b) and Zeng and Su (2013).

As explained in Sect. 2.2.1 and 2.2.2, the GWL is adopted as the bottom boundary condition of the unsaturated zone in STEMMUS-SCOPE. Consequently, a capillary fringe zone is formed above the phreatic surface (Fig. 1a), where water is pulled up from the saturated zone by capillary forces. Moreover, when precipitation infiltrates into the unsaturated zone and percolates downward, it has to go through the capillary fringe zone and continue downward as Rg to the saturated zone. Hence, the capillary fringe zone includes both the Rg and the Cap. Both Rg and Cap, the latter contributes to groundwater evaporation (Balugani et al., 2017), are calculated from the summation of qLh, qLT, qvh, and qvT fluxes (Eq. 2) at the capillary fringe zone. The result of that summation determines whether the calculated flux (Q, Eq. 2a) represents Rg or Cap. If the summation is negative (Eq. 2b), it means the flow direction is downward, and the Rg term is dominant. In contrast, if the summation is positive (Eq. 2c), it means the flow direction is upward, and the Cap term is dominant. In STEMMUS-SCOPE, upward fluxes (capillary rise) are positive, while downward fluxes (gravity drainage) are negative. However, MODFLOW 6 uses the opposite sign convention. Therefore, the Rg/Cap sign needs to be inverted, when transferred to MODFLOW 6 through BMI. 2aQ=qLh+qLT+qvh+qvT2bRg=-QifQ<00ifQ≥02cCap=QifQ≥00ifQ<0

The root water uptake (RWU) is calculated in STEMMUS-SCOPE based on a resistance scheme that takes into account the hydraulic gradient between the soil water potential (ψs) and the leaf water potential (ψl) (Eq. 3). The root water uptake module, developed by Wang et al. (2021), takes into account the θ as the only source of water for the roots (root soil water uptake, RWUs), meaning that RWU=RWUs. The RWUs is calculated for all the soil layers that are above the maximum rooting depth (dmax⁡). RWUs=∑i=1nψs,i-ψlrs,i+rr,i+rx,i and 3RWU=RWUs where ψs,i is the soil water potential of layer i (m), ψl is leaf water potential (m), rs,i is the soil hydraulic resistance (s m⁻¹), rr,i is the root resistance to water flow radially across the roots (s m⁻¹), and rx,i is the plant's axial resistance to flow from the root xylem to the leaves (s m⁻¹).

However, the roots can uptake water from an additional source (root groundwater uptake, RWUg) if the roots have access to groundwater (GWL is above dmax⁡). As explained in Sect. 2.2.1 and 2.2.2, the GWL is adopted as the bottom boundary condition of the unsaturated zone for the soil water module in STEMMUS-SCOPE; hence, the RWU can also be influenced. The RWU calculations were updated by splitting it into two components (RWUs and RWUg, Eq. 4). The RWUs is for the soil layers that are above the capillary fringe and above dmax⁡ (Eq. 5), and the RWUg is for the soil layers that are below the capillary fringe zone and above dmax⁡ (Eq. 6). The RWUs and RWUg are calculated by replacing the ψs,i in Eq. (3) with the updated ψs,i (due to the effect of GWL) in Eq. (5), and with the groundwater potential (ψg,i) in Eq. (6), respectively. For soil layers below dmax⁡, RWU = 0. 4RWU=RWUs+RWUg5RWUs=∑i=1nψs,i-ψlrs,i+rr,i+rx,i for the soil layers above capillary fringe zone and above dmax⁡. 6RWUg=∑i=1nψg,i-ψlrs,i+rr,i+rx,i for the soil layers below capillary fringe zone and above dmax⁡.

Both ST-SC and ST-SC-MF6 models of the three sites used a three-month model initialization phase (i.e. “spin-up”) that extended from 1 January to 31 March 2016. After this, the simulation was run for a period of 8 years (1 April 2016–31 March 2024). The total simulation had 144 624 half-hourly time steps.

The unsaturated zone in STEMMUS-SCOPE was modelled with site-specific total thicknesses – each equals or exceeds the site's maximum observed water table depth (Table 2). The unsaturated zone was vertically discretized into 69, 50, and 33 layers for Loobos, Cabauw, and Veenkampen, respectively (Table 2). The thickness of the layers was variable (minimum of 1 cm and maximum of 20 cm), started with fine resolution near the surface (1 cm), gradually increased to 20 cm, then gradually decreased again to 5 cm near the water table fluctuation zone.

The MODFLOW 6 models were horizontally discretized into a single cell that is 1 m by 1 m, and were vertically discretized into 20, 15, and 18 layers for the Loobos, Cabauw and Veenkampen sites, respectively (Table 2). The layering information was retrieved from the BRO GeoTOP v1.6.1 geological model (TNO – GDN, 2025a) and the BRO REGIS II v2.2.3 hydrogeological model (TNO – GDN, 2025b). Since the models were 1D models in the vertical direction, lateral flow considerations were neglected. To minimize the effect of this neglect, the saturated zone simulation was limited to the upper-most permeable hydrogeological unit, which terminates at a depth of 20 m b.l.s. for Loobos and 9 m for both the Cabauw and Veenkampen sites. This permeable hydrogeological unit was then sub-divided into thinner layers, started with a thickness of 0.25 m, then smoothly increased to 0.5, 0.75, and 1 m. At the base of the bottom-most active layer, a specified flux boundary condition was applied to compensate for the vertical flow of the excluded deeper hydrogeological units.

Each model of the three sites was initialized by an initial θ and Ts for STEMMUS-SCOPE, and an initial GWL and GT for MODFLOW 6. The initial θ, Ts, GWL, and GT values were retrieved from the collected datasets (Table 3) of the three sites as follows: (a) 0.1, 5 °C, 21.5 m, and 10 °C for Loobos, (b) 0.6, 5 °C, -0.8 m and 10.5 °C for Cabauw, and (c) 0.75, 5 °C, 5.1 m, and 11 °C for Veenkampen.

The top boundary of the soil water module in STEMMUS-SCOPE was a specified flux (precipitation, input retrieved from the collected datasets in Table 3), and the bottom boundary was the GWL (received from MODFLOW 6). The top boundary of the soil energy module in STEMMUS-SCOPE was a specified temperature (air temperature, input retrieved from the collected datasets in Table 3), and the bottom boundary was the GT (received from MODFLOW 6). The top boundary of the GWF module of the saturated zone in MODFLOW 6 was the Rn (received from STEMMUS-SCOPE), and the bottom boundary was a specified flux (unknown value that was initially set to 0.0005 m d⁻¹ and calibrated later). The top boundary of the GWE module of the saturated zone in MODFLOW 6 was the Qheat and RTn (received from STEMMUS-SCOPE), and the bottom boundary was a specified heat flux (unknown value that was initially set to 0.05 W m⁻² and calibrated later).

Both STEMMUS-SCOPE and MODFLOW 6 models have several parameters – herein, only the parameters that had a high impact on the simulations (based on an earlier sensitivity analysis) are calibrated and listed in Table 4. The main parameters that govern the flow simulation in STEMMUS-SCOPE were: the vertical saturated hydraulic conductivity of the unsaturated zone (Ksat), residual water content (θr), saturated water content (θs), empirical coefficients of the Van Genuchten equation (α and n), and dmax⁡. In MODFLOW 6, the main aquifer parameters were: the horizontal (Kh) and vertical (Kv) hydraulic conductivity, specific storage (Ss), and specific yield (Sy). Both STEMMUS-SCOPE and MODFLOW 6 used the same thermal parameters to simulate the heat transfer in both the unsaturated and the saturated zones, including: thermal conductivity of solids (KT), specific heat capacity of solids (cs), and bulk density (ρs). As explained in Sect. 2.4.1, the soil profiles at each site were discretized into multiple vertical layers (Table 2) to account for soil heterogeneity. Initially, all layers were assigned uniform values for each parameter. During the calibration procedure, these parameters were updated individually for each layer to reflect the vertical variability of the θ and Ts measurements. The ranges of the calibrated values for each parameter across the layers are summarized in Table 4.

The 8-year simulation period was divided into a 4-year calibration period (1 April 2016–31 March 2020), followed by a 4-year validation period (1 April 2020–31 March 2024). The collected observation datasets (Table 3) were used as calibration/validation state variables for the calibration/validation periods, respectively. The models were calibrated by trial and error until satisfactory qualitative (graphical) and quantitative (statistical) matches between the simulated variables and their observed equivalents were achieved. To ensure an unbiased comparison, both ST-SC and ST-SC-MF6 setups across the three sites were initialized using identical baseline parameter values derived from each site's measurements and literature. The manual calibration for both setups followed the same iterative step-by-step procedure within identical, physically realistic parameter bounds (Table 4). The goodness of fit was assessed using two statistical metrics: the Kling–Gupta efficiency (KGE; Gupta et al., 2009) and the root mean square error (RMSE). The KGE value can range from negative infinity to 1 and was deemed acceptable, when it was larger than zero (Knoben et al., 2019). The RMSE value was deemed acceptable, when it was minimized to 0.1 m³ m⁻³, 2 °C, 0.5 m, 1 °C, 1 mm d⁻¹, 3 g C m⁻² d⁻¹, 5 g C m⁻² d⁻¹, and 0.25 mW m⁻² µm⁻¹ sr⁻¹ for θ, Ts, GWL, GT, ET, GPP, NEE, and SIF, respectively, as similar ranges were indicated in former studies (Lekula and Lubczynski, 2019; Pacheco-Labrador et al., 2019; Thum et al., 2019; van der Velde et al., 2023; Benz et al., 2024; Qian et al., 2024; Yang and Lei, 2024; Wang et al., 2026). Additionally, to evaluate whether the increased complexity of ST-SC-MF6 is justified relative to parameter addition, the Akaike Information Criterion (AIC; Akaike, 1973) was computed across all observed variables.

Figure 4 presents half-hourly observations and simulated outputs from the ST-SC and the ST-SC-MF6 models for θ at available depths at the three sites. The improvements in capturing the θ dynamics within ST-SC-MF6 simulations were evident at all soil depths, but were most pronounced during the wet periods at Cabauw and Veenkampen sites, characterized by shallow groundwater depths (0–1.25 m). At these two sites, the ST-SC-MF6 models successfully captured the θ peaks that occurred during winter periods, along different soil depths (0.6 m³ m⁻³ at Cabauw and 0.75 m³ m⁻³ at Veenkampen), representing near-saturation conditions. Those peaks reflect strong connectivity between soil and groundwater and the strong influence of groundwater on the soil profiles. In contrast, the ST-SC models underestimated these winter θ peaks, as the absence of groundwater limited the θ profile's ability to reach saturation conditions. At Loobos, groundwater depth was relatively moderate (3.0–5.0 m), and had limited influence on the θ dynamics. Thus, the improvement of the θ simulation of the ST-SC-MF6 over the ST-SC model at Loobos was less associated with direct groundwater feedback and more related to improved model structure and parameter calibration within the ST-SC-MF6 model.

Figure 5 presents half-hourly observations and simulated outputs from the ST-SC and the ST-SC-MF6 models for Ts at available depths at the three sites. The ST-SC-MF6 model showed a closer agreement with the observations than the ST-SC model at the three sites, but this was more pronounced at Loobos. The GT observations were only available at Loobos, enabling a direct assessment of groundwater influence on Ts, unlike the case at Cabauw and Veenkampen. At Loobos, the ST-SC-MF6 simulation well captured the Ts dynamics (Fig. 5a and b). In contrast, the ST-SC simulation showed an overestimation of the θ amplitude (fluctuations), which reduced (dampened) the Ts amplitude, leading to lower Ts peaks and higher Ts troughs than the observations. Similarly, at Cabauw, the ST-SC-MF6 simulation captured well the Ts dynamics, while in contrast, the ST-SC simulation showed an overestimation of the Ts amplitude, leading to higher Ts peaks and lower Ts troughs than the observations (Fig. 5c and d). At Veenkampen, both models generally reproduced the seasonal Ts patterns and corresponded well with the observations, but exhibited a systematic tendency to slightly overestimate Ts amplitudes along the entire soil profiles (Fig. 5e and f).

The GT, simulated by MODFLOW 6, was used as a bottom boundary for the STEMMUS-SCOPE soil energy module. At Loobos, where GT observations were available, the ST-SC-MF6 produced a better simulation of the Ts profile as compared to the ST-SC (Fig. 5a and b). At this site, the groundwater depth is relatively moderate (3.0–5.0 m b.l.s.), resulting in a balance between atmospheric and groundwater heat processes' influence on Ts. While the shallower soil layers closely follow atmospheric conditions, deeper layers are more influenced by groundwater conditions (Parrales et al., 2026). Both observed and simulated GT fluctuated between 10.6 °C during winter periods and 9.8 °C during summer periods (Fig. 3b). In contrast, Ts exhibited a much larger seasonal variability (Fig. 5a and b), with both observed and simulated Ts ranging from 15.0–18.0 °C during summer period to 5.0–8.0 °C during winter period across the soil profile. With increasing soil depth, the Ts is damped with a seasonal phase lag until gradually approaching stable GT values. By incorporating the simulated GT as a bottom boundary condition for the unsaturated zone, the ST-SC-MF6 simulation successfully reproduced the same patterns as the observed Ts. During winter periods, the GT was higher than Ts, generating upward heat transfer by diffusion, i.e. temperature gradient, from the saturated to the unsaturated zone. Towards the spring period (April–June), GT gradually declined, while Ts increased, resulting in a temporary equilibrium state between the two thermal regimes. Later during summer periods, the opposite situation occurs, with Ts exceeding GT, reversing the direction of the heat transfer from the unsaturated to the saturated zone.

It is notable that matching the simulated Ts with the corresponding observations at the Loobos site was challenging. The Loobos site has low values of θ (maximum value = 0.16 m³ m⁻³) along the soil profile (Fig. 4a and b), corresponding with relatively low Ts values (maximum value = 18.0 °C). Such a situation created extra difficulty in modelling, since in STEMMUS-SCOPE, the θ and Ts are simulated together with many influencing factors as follows: (1) the low θ may shift the energy partitioning to more sensible heat and less latent heat, which leads to low values of evaporation and extra soil warming (overestimation of Ts profile); and (2) the thermal parameterization (KT and cs) in STEMMUS-SCOPE follows the de Vries formulation (Yu et al., 2018), which is influenced by θ. In this case, the low θ values lowered the values of thermal parameters, which may lead to rapid soil heating and further contribute to the overestimation of Ts. Hence, the overestimation of the θ amplitude by ST-SC dampened the amplitude of the Ts.

At Cabauw and Veenkampen, although no GT observations were available to control the simulations, groundwater still exerted potential influence on Ts profile, driven by the GWL rather than the GT. The shallow groundwater depths (0–1.25 m b.l.s.) at these two sites enhanced soil thermal conductance and heat capacity, which facilitated heat transfer between the topsoil and the subsurface (Alkhaier et al., 2012). Under such conditions, the unsaturated zone becomes more thermally connected to the atmospheric forcing rather than the saturated zone. Previous studies have reported a strong correlation between GWL and Ts, particularly under shallow groundwater depths, i.e. < 5.0 m (Alkhaier et al., 2012; Malik et al., 2021; Mandal and Roy, 2024; Oishy et al., 2025). The de Vries heat transport formulation, implemented in the energy module of STEMMUS-SCOPE (Yu et al., 2018), allows for producing such GWL–Ts relationship.

At Cabauw, the ST-SC-MF6 model showed a better simulation of the Ts profile as compared to the ST-SC model (Fig. 5c and d). During winter periods, the GWL rose (Fig. 3c), which was associated with high θ values, i.e. 0.6 m³ m⁻³ (Fig. 4c and d), nearly approaching saturation, and relatively low Ts (Fig. 5c and d). Nevertheless, both θ and Ts of the ST-SC-MF6 remained higher than the ones, produced by the ST-SC model and closer to the observed ones. During summer periods, the GWL declined, leading to low θ and high Ts values, while Ts remained higher than the values produced by the ST-SC model. Collectively, during the hydrological year, the presence of the shallow GWL in the ST-SC-MF6 simulation increased the seasonal amplitude of θ, i.e. raising maximum θ during winter while maintaining nearly similar minimum θ in summer. This behaviour, in turn, reduced the seasonal amplitude of Ts, i.e. increasing minimum Ts during winter and reducing maximum Ts during summer. In contrast, the ST-SC model, due to the absence of GWL, underestimated the θ amplitude and consequently overestimated the Ts amplitude.

At Veenkampen, the GWL showed little impact on the Ts simulation (Fig. 5e and f), although the ST-SC-MF6 model improved the θ simulation by raising the θ level, particularly the winter peaks, relative to the ST-SC model, whose θ showed a similar amplitude but was systematically lower (Fig. 4e and f). This shift in θ did not noticeably affect the thermal properties of the soil, which remained similar in both simulations, leading to comparable Ts dynamics. Additionally, in both models, the amplitude of the simulated Ts was slightly overestimated. The likely reason is the resistance scheme in the soil energy module in STEMMUS-SCOPE, which follows the big leaf concept, assuming that the boundary layer resistance is the same for all canopy layers – a simplification that could be refined in the future.

Beyond the groundwater influence on Ts dynamics, groundwater can also affect the thermal regime of the land surface and vegetation canopy. These effects may propagate upward, influencing surface soil temperature (Tss) and canopy temperature (Tc), which in turn regulate vegetation physiological processes such as stomatal conductance and carbon assimilation.

At Loobos, the groundwater – soil heat exchanges, simulated by the ST-SC-MF6 model, facilitated proper simulation of temperature-dependent processes within the SPAC. In particular, GT played a crucial role in cooling Ts, especially in the deeper soil layers, close to the capillary fringe zone, thereby stabilizing the root-zone thermal conditions and supporting the RWUg – as observed through the increase of RWUg, simulated by ST-SC-MF6 (Sect. 4.3.3). Furthermore, the moderating influence of the GT on the Ts propagated upward to cool down Tss and Tc. The ST-SC-MF6 simulation produced consistently cooler Tss and Tc (i.e. lower peaks) during the spring periods compared to the ST-SC simulation, with reductions of approximately 1 °C in Tss (Fig. 9a and b) and up to 2 °C in Tc (Fig. 9c and d).

At Cabauw, the influence of groundwater on the surface thermal regime is primarily controlled by the shallow GWL rather than GT. The shallow GWL moderates near-surface thermal conditions through its control on θ and thermal properties, enhancing the buffering of temperature fluctuations in the soil profile. The ST-SC-MF6 simulation produced cooler Tss (i.e. lower peaks, Fig. 9e and f) and Tc (Fig. 9g and h) during the spring periods as compared to the ST-SC simulation, with average reductions of about 1 and 2 °C, respectively. Such moderation of Tc, particularly during the spring periods, when the vegetation activity is high, would maintain stomatal conductance and support carbon uptake. Under good nutrient and water availability conditions, slightly cooler Tss and Tc would further enhance photosynthesis and carbon assimilation (Oishy et al., 2025), as observed in the NEE, simulated by ST-SC-MF6 (Sect. 4.4).

At Cabauw and Veenkampen, the Ess values simulated by ST-SC-MF6 models were increased by 7.3 % and 12.6 %, respectively, as compared to the ST-SC models. These increases can be attributed to two mechanisms. First, the formation of the capillary fringe zone (Fig. 1) above the phreatic surface enabled the calculation of the Cap flux. The Cap flux further contributed to the water loss that moved upward to the land surface, where Ess takes place (Hird and Bolton, 2017; Balugani et al., 2021, 2023). Second, the improved simulation of ψs,i and the corresponding θ created strong upward hydraulic gradients that promoted Ess (Huo et al., 2020). Both experimental (Huo et al., 2020) and field studies (Yang et al., 2021b) have demonstrated that shallow groundwater, i.e. generally less than 2.5 m, can sustain evaporation by maintaining moisture supply to the soil layers.

The inclusion of GWL as a dynamic bottom boundary has improved the simulation of the ψs,i across the soil profile by allowing proper seasonal dynamics of ψs,i with respect to the GWL fluctuations. The improved representation of ψs,i resulted in a substantial improvement of the RWUs by better capturing the hydraulic connectivity between the saturated zone and the root zone (Wang et al., 2023). At Cabauw and Veenkampen, RWUs, which is based on ψs,i (Eq. 5), was increased by 29.6 % and 35.9 %, respectively. Such improvement highlights the sensitivity of the RWUs to the groundwater depth, particularly when the groundwater depth lies within the range of 0.5–2.5 m from the land surface, as reported by Sulis et al. (2017). They demonstrated using two different models, i.e. TerrSysMP (Shrestha et al., 2014) and ParFlow-WRF (Maxwell et al., 2011), applied at vegetated areas in the Rur catchment in Germany, that the GWL fluctuations significantly influence the energy fluxes, further affecting the RWUs, through their impacts on the θ profile. This mechanism has been described by Huang et al. (2017), who showed that shallow groundwater can: (a) prevent the ψl from declining to the critical ψl that triggers stomatal closure; and (b) maintain high values of ψs,i, thereby preventing the rapid depletion of the θ. Consequently, these processes led to increasing the plant water storage and supported high RWUs, as has been observed in the ST-SC-MF6 simulations at Cabauw and Veenkampen sites.

The ψg,i was explicitly defined for the soil layers below the capillary fringe zone, which was further used to calculate RWUg (Eq. 6). The RWUg is an additional component that contributed 28.0 %, 15.2 %, 13.0 % to the total ET at the three sites (Loobos, Cabauw, and Veenkampen), respectively.

At Loobos, the dmax⁡ of the evergreen forest was defined in the models as 4.0 m, allowing the plant roots to access the groundwater, located at depths between 3.0 and 5.0 m b.l.s. The increase in the RWU originated mainly from the RWUg, while RWUs showed nearly no change. This is because of the similar θ simulation by both ST-SC and ST-SC-MF6 models, resulting in comparable RWUs estimates. Furthermore, the groundwater contribution to ET was more significant during dry periods (Fig. 6b and c), when θ was depleted owing to lack of rain, while the GWL continued to support the water uptake by plant roots. Such a mechanism has been reported in many previous studies (Zhu et al., 2009; Wang et al., 2023; Du et al., 2025). From a modelling perspective, Bieri et al. (2025) also showed that the failure of the models to reproduce proper dry season transpiration is often related to the absence of deep RWU mechanisms in the simulations. Moreover, the impacts of GT (discussed in Sect. 4.2.1) on cooling the Ts for the deeper soil layers, allowed for better RWUg during spring and summer periods.

At Cabauw and Veenkampen, the dmax⁡ of the grassland was set to 0.5 m in the models, allowing the roots to access the groundwater, which fluctuated between 0 and 1.25 m b.l.s. However, the contribution of RWUg to the total ET at these two sites remained smaller than at Loobos. That is because the soil water availability, reflected in the improved simulation of θ by the ST-SC-MF6 models, was sufficient to sustain higher RWUs, thereby reducing the reliance on the RWUg. In such shallow water table conditions, land surface processes, including RWU, are subject to more energy control rather than water control (Wang et al., 2023).