The Vesdre catchment is characterized by a complex geological history, spanning from the Cambrian to the Holocene, marked by two orogenic cycles (Caledonian and Variscan) (Fig. 2). This evolution has resulted in significant lithological and pedological diversity. The nature of the soils in the studied catchments is shaped by two main contributions.

A large portion of the soils is developed on loess deposited during various glaciation phases, particularly the Würm. These loess deposits, rich in silt, have covered the entire region, especially on plateaus and areas with moderate relief. On plateaus and gentle slopes, loess is often preserved, forming a homogeneous cover that promotes good permeability and, consequently, effective drainage. These silts, relatively uniform and free of coarse fragments, are still found today in catchment 1 (C1), where they generally reach a thickness of 1 to 2 m. In contrast, in catchment 2 (C2), these deposits persist only in isolated areas and are generally associated with weathering residues from the underlying bedrock.

On steeper slopes, erosion tends to displace these deposits, reducing their thickness or even removing them completely in certain locations. In valley bottoms, colluvium and alluvium represent the parent material for most soils.

In parallel, the degradation and weathering of the underlying rocks contribute to soil formation. The resulting fragments mix with loess deposits and often reflect the lithological nature of the bedrock. The intensity of this process varies depending on topography and rock resistance. In areas where loess is preserved, such as plateaus and moderate slopes, the contribution of rock fragments tends to be less significant, resulting in soils that are less stony. Conversely, on steep slopes and in areas where loess has been largely eroded, the influence of the underlying rock becomes more pronounced, leading to soils with a higher coarse fraction and smaller depths of loose material.

Soil natural drainage primarily depends on several factors. The permeability of the substrate and the nature of the materials play a crucial role. Soils derived from recent deposits, such as colluvium, alluvium and loess, are generally permeable, facilitating infiltration. However, soils formed from more altered or clay-rich units, such as the Cretaceous Herve smectites (locally present in C1), exhibit reduced permeability, leading to slower drainage.

The presence of impermeable layers also influences drainage. In certain areas, particularly upstream of C2, alteration clays or compacted layers act as barriers, restricting percolation and leading to the formation of perched water tables. Under these conditions, the decomposition of fresh organic litter slows significantly, resulting in a peat-forming soil. If this process continues, it can eventually lead to the formation of sphagnum peat, as observed in the extreme eastern part of C2. Conversely, when soils lie directly on a more permeable substrate, drainage is improved. This is notably the case in the lower part of C2, where impermeable layers tend to be thinner and soils directly overlying schists benefit from the rock's schistosity, which enhances infiltration. Similarly, in the lower part of C1, limestone formations may also contribute to efficient drainage, despite the fact that the loess overlying these limestones tends to be more clay-rich.

Topography further influences drainage conditions. In valley bottoms, the water table is often permanent and reaches the surface, limiting infiltration in these soils. Locally, peat soils can be found in the valley bottoms of C2. In contrast, on plateaus and gentle slopes covered with loess, such as in C1, drainage is generally good, except in certain areas of the upper part of C2, where clay accumulations or the presence of impermeable layers may restrict infiltration.

The model selection was based on its capability to represent long time series of multiple hydrological processes (channel flow, overland flow, unsaturated and saturated flow, evapotranspiration, etc.), accounting for interactions between the saturated zone, vadose and surface, at the catchment scale using a fully distributed 3D lattice. Therefore, the coupled hydrologic/hydraulic MIKE SHE/MIKE 1D model was chosen. MIKE SHE/MIKE 1D solves partial differential physically based equations describing mass and energy transfers. The actual evapotranspiration is derived from the reference evapotranspiration using the Kristensen and Jensen method. The 2D overland flow is calculated using the diffusive wave approximation of the Saint-Venant equations. The 1D channel flow is derived from the fully dynamic higher-order approximation of the Saint-Venant equations. The 1D unsaturated flow is derived from the Richards equation. Saturated flow is described by the 3D Darcy's equation, numerically solved using the finite difference method (DHI, 2024). All input data are summarized in Table 1.

Hydrometeorological data: the climate data comprised a 5×5 km grid detailing hourly precipitation intensities and daily reference evapotranspiration. Reference evapotranspiration was estimated using the Penman–Monteith equation (Allan et al., 1998) applied to data collected by the observation network of the Belgian Royal Meteorological Institute (RMI), including temperature, humidity, wind speed and solar radiation (Royal Meteorological Institute of Belgium, 2010). Precipitations were determined by combining RMI's gridded daily data with hourly point measurements from the Wallonia Public Service's rain gauge network (https://hydrometrie.wallonie.be, last access: 24 March 2023), enabling the hourly distribution of RMI's daily rainfall data.

Hourly stream discharge data were obtained from 11 May 2011 at the outlet of C1 (gauging station 1), from 30 November 2015 at the outlet of C2 (gauging station 2) and from 1 January 2000 in an upstream branch of C2 (gauging station 3) with a drained area of 19.6 km² (https://hydrometrie.wallonie.be, last access: 6 October 2022).

Topographic data: the topographic data were a hydrologically corrected digital elevation model (DEM) at a 2 m resolution (Service public de Wallonie, 2023a). Resampling onto the model grid involves selecting the minimum elevation value to maintain the hydrological correction applied.

Land cover–land use data: land use/land cover (LULC) information was used to define spatially variable surface roughness and vegetation. The LULC map was a 5 m resolution layer of the land cover of 2018 (Service public de Wallonie, 2024b; Bassine et al., 2020) mapping several LULC classes. Resampling onto the model grid was performed by selecting the predominant LULC within each grid cell. Each LULC class was assigned an M value of Manning.

The surfaces were also categorized into seven vegetation classes based on the LULC map. Apart from permanent grasslands, which were considered open production surfaces, most croplands were cultivated with maize. Consequently, maize development dynamics were assigned to all croplands, and in the absence of information on cropping practices, no crop rotations were modelled. Each vegetation class was assigned annual dynamics (temporally varying) of the leaf area index (LAI), crop coefficient factor (Kc) and root depth (Rd) (Appendix Fig. A1). These parameters are used by the model to refine the reference evapotranspiration (Penman–Monteith) into the actual evapotranspiration following the approach of Kristensen and Jensen (DHI, 2024). In addition, detention storage, which is a parameter aimed at accounting for ponding in depressions at sub-cell scale, was fixed at 4 mm in both catchments.

Soil and subsoil data: the 1D Richards equation was used to model vertical water fluxes in the unsaturated zone, which required the determination of the soil hydrodynamic properties (soil water retention and soil hydraulic conductivity curves) using the Van Genuchten and Mualem functions (Van Genuchten, 1980) at each location within the unsaturated zone. Saturated flow was modelled with the 3D Darcy's equation requiring vertical and horizontal saturated conductivities to be specified at each location within the saturated zone.

The soil and subsoil were represented as a series of layers with homogeneous hydrodynamic properties within each layer but varying between layers in terms of thickness and lateral arrangement. Except for peat and degraded peat soils, the upper soil profile was discretized into two layers: topsoil, ranging from 0 to 0.4 m, and subsoil, extending from 0.4 m to a maximum of 2 m deep. These two layers could be truncated by the base at depths of 0.2, 0.3 or 0.6 m, considering the soil depth information from the Belgian soil map (Service public de Wallonie, 2023b).

The upper soil profile was delineated into homogeneous units based on the Belgian soil map (Fig. 1) (Service public de Wallonie, 2023b). This map is derived from approximately 6 million samplings conducted with a soil auger (max 1.25 m depth) between 1947 and 1991 of the Belgian cadastral plans, according to a square grid of 75 m per side (1 to 2.5 observations per hectare). This map delineates 6000 soil units defined by characteristics such as texture, natural drainage, diagnostic horizon and stone content (Legrain et al., 2011). For each homogeneous unit, except for peat, degraded peat and impermeable soils, retention and hydraulic conductivity curves (Van Genuchten and Mualem models, with m= 1–1/n and L= 0.5) for both soil layers were derived from the European hydraulic pedotransfer function (Weber et al., 2020) number 1 using depth and the averaged soil texture (“Textures et fractions granulométriques de référence des sols de Wallonie – Série”, Service public de Wallonie, 2023c) as predictors (Szabó et al., 2021). The predicted saturated water content, θs, was subsequently corrected to account for the loss of porosity due to stone content, s, as represented in the Belgian soil map. 1θs-pebble=θs(1-s) For the peat soil units, retention curves were obtained by measurements of water content on peat samples of local bogs using pressure plates at pF values of 1.00, 1.60, 1.85, 2.00, 2.30, 2.78, 2.95, 3.62 and 4.14. Van Genuchten functions were then adjusted to measure the retention curves. The saturated hydraulic conductivity of peat is spatially highly variable and challenging to measure in laboratory. Saturated hydraulic conductivity was initially determined based on values found in the literature (Wastiaux, 2008). As suggested by Szabó et al. (2024), it was subsequently refined through calibration using discharge measurements conducted between 2012 and 2015 at the outlet (gauging station 4) of a small peatland catchment of 14 ha, located 5 km away from C2. The thickness of the degraded peat soil unit was assumed to be 0.4 m, while the peat soil unit was assumed to be 2 m thick. Beneath peat or degraded peat, a low-permeability layer of 0.6 m thickness with 60 % clay, 35 % silt and 5 % sand was assumed, and its hydrodynamic parameters were retrieved using euptfv2.

Below the upper soil profile, multiple geological units extending to a depth of 18 m were delineated based on a 1×1 km grid. Each geological unit consisted of multiple layers, exhibiting variations in thickness and vertical hydraulic conductivity. These values were defined through a bibliographic study of the hydrogeological context of the Vesdre catchment (Sohier, 2011). Due to their imprecise nature and large spatial variability, vertical hydraulic conductivities were initially defined as ranges of plausible values and were subsequently refined during the model calibration process. Horizontal hydraulic conductivities were assumed to be 10 times greater than vertical hydraulic conductivities (DHI, 2024).

Model specification, calibration and validation: for both catchments, a model of the present situation (further referred to as BASELINE) was calibrated and validated. The total simulation period was from 1 January 2002 to 31 December 2021. At the start of the simulation process, the soil water content was set to field capacity. The first 2 years of simulation were considered a warming period and were not considered in the data analysis. Outputs were stored on an hourly basis. To ensure a fine representation of the catchments while maintaining reasonable computational times, the grid resolution of the C1 model was 20 m, whereas it was 40 m for the C2 model. Calibration was conducted in the period between 1 October 2017 and 1 October 2021, while validation encompassed the remaining available data.

The calibration was performed manually and limited to very few parameters, for which the available data were considered uncertain and had the most impact (the most sensitive parameters) on the modelled hydrographs. These parameters included the Manning's M roughness coefficients for overland and channel flows (which are mostly sensitive regarding the temporality of hydrographs and peak flows), as well as the saturated hydraulic conductivity of the geological layers (which are sensitive regarding baseflow, peak discharge values and groundwater head) (XEVI et al., 1997). Calibration of the Manning's M roughness and the saturated hydraulic conductivities of peat was performed using discharge measurements conducted between 2012 and 2015 at the outlet of the small peatland catchment of 14 hectares (gauging station 4). Calibration and validation were performed in relation to historical discharge measurements taken at the river gauging stations. The ability of the model to reproduce observed discharges was assessed against the Moriasi et al. (2007) model evaluation guidelines. This included the computation of the Nash–Sutcliffe efficiency (NSE), the ratio of the root mean square error to the standard deviation of measured data (RSR) and the percent bias (PBIAS). 2NSE=1-∑i(QObsi-QModi)2∑i(QObsi-∑iQObsin)2 NSE varies from -∞ to 1. The closer to 1, the better the simulation performance. QObs and QMod are the observed and simulated discharges, respectively. i is the time step. 3RSR=RMSESTDEVobs=∑i(QObsi-QModi)2∑i(QObsi-∑tQObsin)2 RSR varies from 0 to +∞. The closer to 0, the better the simulation performance. 4PBIAS=∑i(QObsi-QModi)⋅100∑tQObsi PBIAS varies from -∞ to +∞. The closer to 0, the better the simulation performance. Positive values indicate model underestimation bias, and negative values indicate model overestimation bias.

In the absence of piezometric data at the study sites, a graphical comparison (Fig. 4) was conducted between the simulated vertical dynamics of the saturated zone (from October 2017 to October 2020) at each grid cell with the corresponding drainage class from the Belgian soil map (Table 3). This ensured that the modelled saturated zone dynamics match those described by the soil map, which was derived from field observations. Drainage class is a concept used in different soil classifications to describe the frequency and duration of soil wetness under natural conditions (Ditzler, 1999). In the Belgian soil classification, drainage classes are determined by the depth at which mottling and reduction processes occur in the soil profile, often corresponding to the levels of saturated zone fluctuations (Table 3). This method facilitates semi-quantitative spatial validation across all areas of the catchments.

The impact of NbSs on catchment hydrology was evaluated through the comparison of responses between BASELINE and POST configurations, under consistent meteorological forcing. The POST landscape configuration represented a scenario where soil–vegetation NbSs were implemented at the scale of the entire watershed. These landscape planning POST scenarios were developed in accordance with the “Schéma Stratégique Vesdre” (Service public de Wallonie, 2024a), which outlines a long-term vision for sustainable territorial development and proposes a range of potential measures to mitigate flood and drought risks. The modelled scenarios (POST) incorporate as many feasible measures as possible within specific contexts (agriculture-dominated and forest-dominated) to assess their potential for flood and drought mitigation. The scenarios compare the current situation to a long-term projection (2050 horizon), assuming full implementation and maturity of the measures. The transition process between current and future states is not modelled, although it may significantly influence the actual effectiveness over time. It should be emphasized that accounting for the transition phase may be essential, as some measures require several years or even decades to achieve full efficiency, and the speed at which a measure becomes effective may serve as an important criterion for prioritizing NbSs.

The following NbS measures were implemented in the agricultural catchment (C1) (Fig. 3a):

Hedgerows: a dense network of hedgerows was systematically implemented around all agricultural parcels, representing around 700 km of barriers within the entire watershed (≈ 175 m ha⁻¹) (Table 4a).

In the forest catchment (C2), the following NbS measures were implemented (Fig. 3b):

Restoration of peatlands: this involved restoring peatlands and open wetlands on degraded peaty soils, currently occupied by conifer plantations and moors (4.6 km², or 6.5 %). The scenario simulated the plugging of the existing surface drainage network with 275 plugs distributed along concentrated flow paths (drainage area from 2 to 5 ha) and the creation of 32 ponds of 640 m³ each (Table 4f).

These NbS measures are incorporated into the model by adjusting some key parameter values (Table 4). It should be noted that although some literature exists on how NbSs may influence these parameters, significant uncertainties remain regarding their exact values, as they cannot be assumed to apply directly to this specific case. These parameter values should be considered as hypotheses, while remaining within plausible ranges.

A complete processing chain of results focusing on the quantification of the hydrologic impact of NbS measures and scenarios was conducted (Table 5). Hydrologic indicators were calculated or developed to capture the impact of various measures within the watershed, emphasizing the spatial variability of their effectiveness. These indicators are computed to assess how NbSs could influence hydrological extremes, particularly focusing on flooding events and agricultural droughts.

Peak flow and flood volume percent reduction: these indicators were integrative metrics permitting the evaluation of the overall effect of a given scenario on flood reduction without discriminating against the impact of the various measures that comprise the scenario. To this end, these two indicators were calculated for a set of rainfall events generating a detectable signal on the hydrograph. A focus was also made on the rainfall event between 13 and 18 July 2021:

Peak flow percent reduction – the evolution of peak discharges between the BASELINE and the POST scenarios. 5Peakflowpercentreduction=100⋅PeakdischargePOST-PeakdischargeBASELINEPeakdischargeBASELINE Flood volume percent reduction – The evolution of the total amount of water discharged at the outlet between the BASELINE and the POST scenarios. 6Floodvolumepercentreduction=100⋅TotalvolumePOST-TotalvolumeBASELINETotalvolumeBASELINE Spatialized indicators: these indicators aimed to assess the local hydrological impact of specific NbS measures where they were located within the watershed. This allowed for discrimination of measures based on their relative impact within the same scenario. These indicators presupposed that the hydrological effects observed at a given location are exclusively attributed to the measure implemented at that site, irrespective of the presence of other measures elsewhere within the watershed.

Spatial variation of infiltration rates: as a spatialized indicator to assess the effectiveness of NbSs in mitigating flood, the cumulative infiltration, Icum, between 13 and 18 July 2021 was calculated for each model grid cell. 7Icum=∑i(Infiltrationi-Seepagei) Infiltration is the amount of water (in mm) flowing into the soil. Seepage is the amount of water (in mm) outflowing from the soil. i is the time step.

Spatial vegetation drought stress duration frequency: this is an indicator to assess the effect of NbSs on vegetation water stress was (newly) developed, utilizing a frequency analysis. Consecutive days where plants at specific locations underwent water stress beyond a designated threshold against a return period were mapped. Thresholds were determined based on the soil water potential within the root zone. An extreme stress threshold was defined when the water potential in the root zone drops below -150 m (≈-1.5 MPa), closely approaching the permanent wilting point, which is classically recognized by soil scientists as the soil water potential beyond which most plants cannot survive (Wiecheteck et al., 2020). Additionally, a moderate stress threshold for vegetation was established at a water potential of -30 m (≈-0.3 MPa), corresponding in our study sites to approximately 30 % of the relative extractable water content in soil (Gourlez de la Motte et al., 2020; Granier et al., 2007). The methodology to construct this indicator is briefly described as follows: for each cell of the model and for each year (2004–2021), the maximum consecutive duration during which the water potential in the soil profile explored by the roots (rhizosphere) was below the stress threshold was extracted. Then, for each cell, a Gumbel distribution was used to establish the relationship between the duration (y) and the probability of occurrence (return period: TM) of the stress. 8y=-cln⁡ln⁡TMTM-1-a c and a are adjustment parameters. TM corresponds to the return period and is calculated according as 9TM=n+1m. n is the rank in descending order, and m is the total number of years of observations. Once the parameters a and c are fitted for a given stress threshold, y can be retrieved using Eq. (8) (Maidment, 1992).

In the agricultural context (Fig. 6a), model results suggest that hedgerows can be highly effective in improving infiltration during an extreme rainfall event, as reviewed by Rajbanshi et al. (2023). This modelled increase in infiltration during extreme events could be attributed to two primary factors: the enhancement of the soil's hydrodynamic properties and the generally drier soil conditions under hedgerows, which increase the unfilled soil porosity and the hydraulic gradient responsible for infiltration. Due to the lack of farm traffic, the greater incorporation of organic matter or a root network promoting macropores, soils under hedgerows are often found to be less compacted and more permeable than those in adjacent arable or pasture fields (Holden et al., 2019; Wallace et al., 2021). Drier soil conditions under hedgerows could be attributed to increased actual evapotranspiration (Benhamou et al., 2013) and canopy interception. A counterintuitive observation from our simulations is that, when considering a broader temporal window, the cumulative annual modelled infiltration decreased under hedgerows due to this increased canopy interception; a larger portion of the rainfall was directly evaporated before reaching the soil for smaller and more common events (results not shown). Hedgerows have also been modelled to reduce agricultural droughts, as their root systems extend deeper than those of crops or meadows (Table 2). This allows them to access water in deeper and wetter soils layers even though soils in the surface were generally drier.

Soil pitting in maize crop also appeared to enhance modelled infiltration during extreme rainfall events (Fig. 6c), as observed by Clement et al. (2023). However, model outcomes showed a minimal effect of soil pitting on agricultural drought mitigation. This may suggest that extreme rainfall events, when a surplus of infiltration is observed, may be temporally disconnected from droughts events; hence, they do not follow close enough together in time for the excess infiltrated water to remain in the root zone until the agricultural drought begins. Another potential factor could be the implementation of hedgerows along field margins in the same scenario, drying soils at the edges and negatively affecting water availability for adjacent crops (Forman and Baudry, 1984).

For reduced tillage practices (Fig. 6b), we did not model large changes in infiltration during extreme events, as reviewed by Clement et al. (2024), nor did we model mitigation of agricultural drought. A well-known positive effect of conservation tillage is the concentration of organic matter into the uppermost soil layer (Haddaway et al., 2017), improving, in turn, aggregate formation (Chen et al., 2020; Obalum et al., 2019), pore size distribution and soil water retention characteristics (Chandrasekhar et al., 2019). In our model scenario, reduced tillage was virtually implemented by reducing the Van Genuchten alpha parameter of the topsoil by 20 %. This adjustment slightly shifted the soil porosity towards smaller pores, specifically from the range of 0 to -330 hPa to the range of -330 to -15 000 hPa, consequently increasing the soil's available water holding capacity. Despite expectations that this improvement in soil physical quality would buffer agricultural drought, our simulation results showed no significant positive effect on mitigating it. Wittwer et al. (2023) recently suggested that while conservation tillage practices could improve soil characteristics, they do not significantly affect root water uptake patterns under severe drought conditions. Moreover, they emphasized that plant ecophysiology and cropping system diversification in space (crop mixture) and time (crop rotation) (Sanford et al., 2021) have a greater impact on crop performance under drought than soil quality. This is a conclusion that our simulation results seem to support. NbSs that were modelled to solely improve the retention characteristics of surface soils without affecting root depth, such as reduced tillage practices or even forest practices aimed at limiting soil compaction, showed a minimal effect on reducing vegetation water stress. One possibility is that, during dry periods, soil desiccates from the surface downward over time. Therefore, if the plant root system extends deeper than the surface soil layer, enhancing the retention properties of this superficial layer, where desiccation occurs first, does not seem to be very impactful for increasing plant resilience to prolonged droughts.

In the forest context, model results indicate that forest diversification could be effective in improving infiltration during extreme rainfall events and in decreasing days of agricultural droughts (Fig. 6e). Mixed forests are often shown to outperform monocultures regarding agricultural drought resistance, especially in dry areas (Liu et al., 2022; Pardos et al., 2021). In our model scenario, forest diversification was implemented primarily by increasing rooting depth, reflecting the observation that mixed forests often develop deeper and more structurally developed root systems than monocultures, especially spruce even-aged plantations that are common in the C2. This results from the inherent differences in rooting patterns between species, leading to generally deeper root extensions or denser fine root systems in mixed stands compared to pure stands (Jose et al., 2006; Meinen et al., 2009), as well as from the competitive environment in mixed stands, which may encourage greater root investment (Reubens et al., 2007). However, the drought resilience of mixed forests may also be attributed to other mechanisms, such as resource partitioning (e.g. root stratification), facilitation (e.g. active hydraulic redistribution), and selection (e.g. a higher likelihood of including drought-tolerant species) (Pardos et al., 2021). However, these complex mechanisms were not explicitly represented in the model.

Practices aimed at limiting forest soil compaction (Fig. 6d) also appeared to be effective in improving infiltration but exhibited low efficiency in reducing agricultural drought, as rooting depth was not increased with this measure.

In accordance with the current literature (Wastiaux, 2008), our model showed almost no improvement of infiltration and agricultural drought mitigation after peatland restoration in moors or in drained spruce plantations on degraded peat soil (Fig. 6f).

Representing the hydrological functioning of a watershed at a fine spatial scale remains a challenge for modellers. Physically based distributed hydrological models offer great potential for capturing detailed processes, but they require a large number of parameters to be determined or calibrated. In many cases, the data required for model calibration and validation are unavailable or highly uncertain at the necessary scale. For instance, calibrating and validating models at the watershed outlet does not necessarily ensure accurate representation of the spatialized hydrological processes within the watershed. Multiple combinations of spatialized model input parameters can yield identical modelled hydrographs at the catchment outlet. To address these challenges, a semi-quantitative spatial validation was carried out by comparing the spatially distributed average depths of the modelled saturated zone with the drainage classes from the Belgian soil map. This exercise demonstrated that the modelling approach effectively captures the fine-scale interactions between soil and water processes. Although this method is specifically tailored to the Belgian context, as this map is specific to Belgium, it provides a valuable framework that could be adapted to other regions. This calls for future research on spatially distributed and cross-scale hydrological observations to further enhance model calibration and validation.

Our results indicated that soil infiltration capacity may be a crucial factor in determining the most suitable locations for implementing NbSs. However, accurately representing the spatial and temporal (Pirlot et al., 2024) variability of soil hydrodynamic properties remains challenging with current methodologies. The maps depicting the soils' hydrodynamic properties are inherently limited in their representativeness due to uncertainties in the applied pedotransfer functions, which are primarily derived from agricultural soils, overlooking much of the soil diversity, such as stony soils. Additional limitations arise from uncertainties and the spatial resolution of the underlying soil property maps (e.g. soil depth, texture, organic carbon and bulk density), as well as the suitability of the continuous function used to describe soil hydrodynamic properties (Weber et al., 2024). Although characterizing the spatiotemporal variability of soil hydrodynamic properties offers valuable benefits in many fields such as hydrology, agriculture, irrigation, ecology or environmental sciences, this remains poorly explored and warrants further research.

Similarly, vegetation plays a crucial role in the soil hydrological functioning at fine scales, significantly influencing watershed dynamics. A striking example is the impact of planting hedgerows, which locally decreases soil moisture, enhancing infiltration rates, resulting in substantial reductions in peak flow at the watershed outlet. While these effects highlight the importance of vegetation, key aspects of its characteristics (also used as model input), particularly tree rooting systems, remain surprisingly underdeveloped. Despite their critical role, our understanding of roots, including their growth, development and interaction with complex physical, chemical and biological soil properties, is still limited (Centenaro et al., 2018).

Three major sources of uncertainty affect estimates of NbS effectiveness in our study: (i) the choice of the hydrological model and its ability to represent key processes, (ii) the model parameterization and modelling of hydrological functioning across the entire catchment in the BASELINE scenario, and (iii) the parameterization of NbSs (Brauman et al., 2022). The first source arises from the inherent limitations of any modelling approach, as different models emphasize different processes and may not fully capture key local dynamics. Unlike many other models, the proposed approach explicitly accounts for the spatial variability of local conditions, which may influence NbS effectiveness (Lalonde et al., 2024). However, certain key processes remain unrepresented, for instance, the recycling of atmospheric moisture (Keys et al., 2018; Theeuwen et al., 2023) and the temporal variability of soil hydraulic properties (Pirlot et al., 2024), which may introduce additional uncertainties. Furthermore, it is critical that the model's spatial resolution corresponds to the scale at which NbSs are implemented. For example, the simulation of hedgerows is limited by the model's 20 m resolution, which may not accurately capture their true configuration, introducing uncertainty. The second source is partially addressed through our model validation process, which integrates all available data, including data from gauging stations and drainage classes from the Belgian soil map, enhancing confidence in the model's representation of catchment-scale hydrological dynamics. The third source has not been explicitly quantified. Instead, we compare our results with reported NbS effectiveness in the literature to assess whether our findings align with expected trends. While this provides a useful reference, variations in study contexts may introduce further uncertainties. This calls for more field-based evaluation of these measures on various conditions. A sensitivity analysis of NbS parameterization, spanning a plausible range of parameter values, could also offer valuable insights into how parameter input uncertainties influence model output uncertainties. Improving the robustness in the estimation of NbS effectiveness remains a critical area for future research.

Currently, there is no globally accepted or standardized approach for monitoring NbSs (Kumar et al., 2021a). The fine-scale spatial variability of NbS effectiveness demonstrated in our study highlights the need to carefully consider the appropriate methods and scales for monitoring these interventions after implementation. These approaches should be determined based on the type, extent and expected outcomes of NbSs. Macro-scale indicators at the watershed level, such as reductions in peak flow, can be highly useful for assessing the overall, cumulative effects of a combination of NbSs implemented upstream. While these indicators are useful, they do not distinguish the individual impacts of specific interventions. Therefore, fine-scale indicators are also necessary to optimize the benefits of NbSs. Nevertheless, these assume that individual interventions act independently and do not interact at small scales, an assumption that may not be always true.

Monitoring approaches often rely on ground-based measurement stations, which are often limited in number and scattered sparsely. Encouraging, recent advances in airborne and satellite-based remote sensing technologies may enable the possibility to bridge this gap between scales, improving the capacity to systematically monitor NbS performance (Kumar et al., 2021a). For instance, monitoring the effectiveness of NbSs in mitigating agricultural drought can be achieved through the comparison of vegetation indices such as the normalized difference vegetation index (NDVI) or the vegetation health index (Patel et al., 2012). Additionally, flood and estimated damage costs might be evaluated by GIS-based monitoring (Haq et al., 2012).

It is important to recognize that the proposed “Spatial variation of infiltration rates” indicator represents just one aspect of NbS effectiveness in reducing flood risks. While it provides valuable insights into how an NbS mitigates the hazard itself, a comprehensive assessment should also account for downstream exposure or vulnerability, which are crucial factors to consider when planning flood risk management (Klijn et al., 2015). For instance, interventions may reduce flood risks locally, but their impacts can also be felt much farther downstream (Lane, 2017). For a comprehensive assessment of the effectiveness of NbSs, it is crucial to adopt cross-scale approaches because the effects of NbSs are not necessarily confined to the immediate area of implementation. Therefore, in-depth evaluation of NbSs requires models capable of capturing complex interactions across spatial scales.

Furthermore, the effectiveness of NbSs may vary over time, depending on (i) meteorological variability (e.g. antecedent moisture conditions) and (ii) the time required for a measure to become fully effective. For instance, some measures, such as forest diversification, may take several years or even decades to reach full maturity. In the current study, while the effect of antecedent moisture conditions was effectively modelled, the transition phase between the BASELINE and POST scenarios was omitted. However, time-variable effects of NbSs may be important to consider when defining priorities in a specific local catchment context (Fennell et al., 2023b). These time-variable effects also raise the question of incorporating future climate scenarios into the modelling of NbS effectiveness (Gómez Martín et al., 2021).

In the present article, the potential synergies or trade-offs that can arise with NbSs for mitigating floods and agricultural droughts were evaluated. Beyond these findings, other co-benefits or trade-offs may emerge, which should be assessed according to the specific context and interests of the region in question. For example, afforestation is known to increase watershed evapotranspiration, reducing river flow and groundwater levels. While this trade-off may be limited in humid climates (Hou et al., 2023; Zhang et al., 2017), which are also less prone to hydrological droughts, such practices, like planting alien trees, could be problematic in arid or semi-arid regions that are more sensitive to long-term hydrological and socioeconomic droughts (Rebelo et al., 2022). Hence, the impacts of these actions extend beyond hydrological effects, encompassing important social, cultural, economic and environmental aspects (Sowińska-Świerkosz and García, 2021). NbSs are inherently integrative and multifunctional; focusing solely on a hydrological performance without considering the broader social, economic, political (governance) and environmental context can undermine the success and legitimacy of projects (Brauman et al., 2022). This underscores the value of multi-objective optimization procedures, which help identify optimal scenarios that balance competing objectives (Strauch et al., 2019; Castro, 2022; Yang et al., 2023).

As demonstrated in this study and many others, NbSs offer significant potential for mitigating floods and droughts. To translate this potential into real-world implementations, several challenges must be carefully addressed. Raška et al. (2022) identified four key aspects that influence the successful implementation of NbSs: (i) uncertainties regarding their effectiveness, (ii) decisions on optimal locations, (iii) institutional frameworks, and (iv) resource availability. The first two aspects primarily arise from the inherent variability of NbS effectiveness across spatial and temporal scales, as well as their multi-objective nature – issues that this study, somehow, aims to elucidate. By providing a quantitative assessment of NbS effectiveness, this research contributes to a more informed decision-making process for stakeholders and policymakers, creating a foundation for discussions on implementation strategies. Strong stakeholder engagement plays a key role in ensuring successful NbS deployment. NbS deployment may generate conflicting interests between different groups, creating both winners and losers (Bark et al., 2021). An interdisciplinary approach that integrates hydrology with social sciences such as sociology and economics can help promote transparency and fairness, particularly in designing effective compensation mechanisms that encourage participation. Moreover, aligning NbS implementation with governance structures is essential. While NbSs are most effective when they are planned at the watershed scale, decision-making often occurs within administrative boundaries that may not fully reflect hydrological realities. Strengthening coordination between authorities and clearly defining responsibilities will enhance the feasibility of NbS adoption.