This paper used the Socio-Hydrological Water Resource (SHOWER) model setup based on with a lumped modelling simulation for soil moisture, three options for a groundwater-outflow representation, a surface water reservoir and water demand components for both anthropogenic and environmental water demand (Fig. ). A detailed model description can be found in Sect. S2 in the Supplement. Key modifications to SHOWER compared to are detailed in Sect. S3 in the Supplement and include (1) minimizing Hortonian runoff as this is less relevant in English catchments , (2) improving the representation of reservoirs so they can be modelled at both upstream and downstream locations and including release flows linked to the ecological minimum flow , and (3) linking the interchangeable groundwater modules to primary aquifers in England, removing the idealised setting in .

As shown in Fig. , SHOWER is driven by daily climate data, i.e. precipitation (P) and potential evaporation (PET). Non-evaporated precipitation (ETa) fills the soil moisture balance and soil characteristics determine how much of this water is runoff (Qr), stored or passed on to the groundwater component as recharge (QRch). SHOWER has three groundwater modules representing karstic, porous and fracture flow which are modelled using modified lumped approaches . Using one of the three groundwater-outflow modules, stored groundwater (Gs) can be released as baseflow (Qb). In the karstic module, baseflow release follows a power law. The porous module accounts for slow baseflow release with a small (7 %–12 %) leakage factor and baseflow in fractured aquifers is modelled using two parallel linear buckets . Discharge (Qs) is taken as the sum of baseflow, runoff and release flow (Qrel) if there is an upstream reservoir. The upstream reservoir catchment is defined by calculating the reservoir contributing area and is modelled as a percentage in the (semi-)lumped catchment approach. The percentage divides the driving data into contributing (reservoir sub-catchment in Fig. ) and non-contributing precipitation and potential evaporation . Consequently, two soil moisture balances are calculated which lead to different discharge and groundwater outputs. Water demand is calculated as a proportion of the long-term recharge and runoff (all water that enters the black box in Fig. ). Anthropogenic water demand can be divided between either surface water (Dsw) or groundwater (Dgw) and is abstracted from the surface water reservoir (Asw) and groundwater storage (Agw).

Water management impact is modelled using four separate drought management scenarios that are compared to a baseline scenario, which influence all fluxes in the black box in Fig. . In the baseline scenario there are no water management interventions and surface and groundwater water demand are simply abstracted from the reservoir and groundwater storage. The other four drought management scenarios were defined in and represent common drought management practices in the UK. The first is to increase groundwater supply, using more of (underused or old) licenced groundwater boreholes and the natural storage buffer that aquifers provide. The second is to reduce water demand, which often starts early with a media campaign to stimulate lesser or limited water use by the public. Severe measures can, however, restrict water use for commercial or non-essential public water use. These threshold-based scenarios (following drought triggers) depend on the severity of a meteorological and/or hydrological drought. Measures are often introduced gradually and their severity increases depending on thresholds for precipitation, discharge, reservoir or groundwater storage levels that are related to historical drought events for details see. The next two scenarios apply regardless of a defined drought and are integrating surface water and groundwater use. The third scenario manages surface water and groundwater in conjunction depending on which resource has a higher availability at a time. For example, in areas with large groundwater storage, more groundwater is used compared to surface water and vice versa for areas with low groundwater storage. In practice, this requires high flexibility in management operations. The last scenario aims to preserve ecological minimum flows in rivers by reducing surface water abstractions. Water is taken from groundwater instead.

We have modelled the threshold-based scenarios using average thresholds for precipitation, discharge, reservoir and/or groundwater levels, following . For the first scenario, we increased groundwater use whereas in the second scenario both surface water and groundwater are reduced equally. The third scenario integrates water storage and takes water from the highest store (either groundwater or reservoir storage) to meet water demand. This represents a non-restrictive application of conjunctive use practices. The last scenario maintains a threshold (Qb_eco) for the ecological minimum flow from baseflow (plus the release flow from an upstream reservoir, if present) and groundwater demand is ceased when this threshold is reached. In the case of water demand exceeding reservoir and groundwater storage, water can be complemented by imported water as either a fixed share or conditionally on (reservoir or groundwater) water levels (Qimp and GWimp).

The response-based model evaluation consists of a Global Sensitivity Analysis to determine the sensitivity of the model outputs to variations of the model parameters. The goal is to evaluate the consistency of the model behaviour with our understanding of key simulated processes, by checking that the “right” parameter controls the “right” output at the “right” time. For such analysis, parameters are sampled from ranges that are meant to represent the variability of hydrological characteristics across Great Britain – called “national parameter ranges” from now on.

For the data-based model evaluation, we first selected representative regions in the UK that captured typical groundwater typologies, which were matched with the groundwater-outflow modules in SHOWER. One study region is set in the Chalk aquifer, which is represented using a large groundwater storage with dominant karstic, non-linear, flow characteristics . The second study region covers the Permo-Triassic Sandstone aquifer using the medium groundwater storage with throughflow in the porous aquifer . Lastly, quick and shallow groundwater storage is modelled using the smaller groundwater storage, reflecting the Dinantian Limestone aquifer with predominantly fracture flow releasing groundwater storage .

Approximately 40 % of all CAMELS catchments (671) are located on one of these productive aquifers, but groundwater level data from the Hydrology Data Explorer were only available for a third of those CAMELS catchments. From these 103 overlapping CAMELS catchments, we identified catchments that had one substantial type of abstraction (surface water or groundwater) and minimal wastewater discharges (<10 % of discharge) in order to evaluate the management components in SHOWER.

The selected three catchments are shown in Fig. . The first catchment (C1) is set in the Peak district in the Dinantian Limestone aquifer area (Derwent river catchment, 335 km2; CAMELS ID: 28043), which is reported to have 25 % of recharge abstracted via surface water . These abstractions are likely to come from the large upstream reservoir (Ladybower) that affects discharge time series downstream . The second catchment (C2) in the Trent river catchment (163 km2; CAMELS ID: 28052) is located in the more urban midlands on top of the Permo-Triassic Sandstone aquifer. Groundwater abstractions are reported to be equivalent to 26 % of long-term recharge (P-PET) . The last catchment (C3) is the Pang river catchment (121 km2; CAMELS ID: 39027) located in Southern England in the Chalk aquifer where groundwater abstractions are approximately 30 % of long-term recharge.

The data-based model evaluation examines modelling performance in matching historical discharge and groundwater level observations in the three case study catchments over a 10 year period (1994–2014). We used CAMELS-GB daily climate time series to run SHOWER in each of the three case study catchments . For each catchment, we run the model against 10 000

A small number of simulations were discarded because parameter combinations created did not respect the non-overlapping range of wilting point, critical moisture content and field capacity (all in mm).

The time-varying global sensitivity analysis shows consistent results for all parameters across the thirty years (reported in Figs. S4–S6 in the Supplement), of which we show a subset (2009–2014) with significant wet and dry periods in Fig. . In general, we find higher parameter influence in wetter periods compared to drier periods and higher sensitivity values for discharge compared to groundwater storage. Soil parameters, such as critical moisture content (CR) and field capacity (FC), control recharge and they are most notably influential during wet periods for both model outputs. The influence of these parameters on groundwater storage lasts for longer compared to discharge. Reservoir parameters are non-influential for the karstic and porous module, which is to be expected with a downstream reservoir setup. From the two groundwater parameters (sK1 and Karst), the parameter regulating non-linear flow (Karst) is very influential for discharge, particularly during high flows. The influence of the discharge-outflow parameter (sK1) is relatively minor, particularly for groundwater, but this is a feature of the karstic module only as the discharge-outflow parameter in the porous (sP1) is much more influential (see Fig. S7 in the Supplement). We also see more sensitive groundwater-outflow parameters in the Fractured module (sF1 and sF2) with the larger one (sF2) being the most sensitive (see Fig. ).

Drought management scenarios are influential during flow recession and low groundwater storage, and non-influential during wet periods. When it is dry, both groundwater abstractions (Dgw) and scenarios (Management) become more influential compared to other dominant soil moisture parameters (CR and FC) and mostly determine the model output. The fractured module has, with its upstream reservoir setup, a different pattern with discharge being influenced by different parameters at different times (Fig. ; groundwater in Fig. S8 in the Supplement). The shape parameter (b; in soil moisture module) and Qeco determining the ecological (and release) flow out of the reservoir are more influential during recession periods compared to the other groundwater modules. Interestingly, the reservoir capacity (Rcap) is only influential for specific days; when the peak flow is high and the maximum capacity is reached. The most sensitive parameters are associated with the drought management scenarios. Drought management scenarios are particularly sensitive during recessions and much less so when discharge peaks.

When analysing the overall sensitivity to the modelled drought management scenarios, we find that model outputs are significantly controlled by the chosen scenario, see Fig. . In this figure, we show the distributions of mean discharge (left) and mean groundwater storage (right) obtained by varying all model parameters within the national ranges, while maintaining a particular management scenario. The Figure shows different results for the two integrated water management scenarios (conjunctive use and maintaining ecological minimum flow). These two scenarios are managing water demand interchangeably between surface water or groundwater depending (1) on storage levels at a time (conjunctive use) or (2) baseflow relative to the ecological minimum flow (maintaining ecological minimum flow).

In the karstic and porous groundwater modules (A-B-C-D), the difference between the conjunctive scenario (in yellow) and the other scenarios follows a similar pattern. There are fewer zero flow conditions and higher flow occurrences when conjunctive use is applied compared to the baseline (in black) and other management scenarios (coloured). Overall groundwater storage is lower in the conjunctive use scenario compared to the other scenarios. This stark difference demonstrates the leverage of the conjunctive use scenario. Other scenarios show very little leverage as their influence relative to parameter uncertainty is small, with exception of the Ecological flow scenario that has a longer tail, indicating high flows more frequently occurring compared to the baseline.

In the fractured module (E-F), discharge and groundwater simulations show three distinct clusters. Drought management scenarios that are trigger-driven, such as increasing water supply or reducing water demand (in blue and red), move the discharge distribution towards the left along the x axis resulting in fewer low flow conditions and overall higher flows compared to the baseline (E). This means that mean discharge is generally higher. Groundwater storage is lower for these scenarios. Integrated water management scenarios (conjunctive use and ecological minimum flow in yellow and green), also result in fewer zero flow occurrences but the distribution of high flows is more similar to the baseline. Integrated scenarios also result in fewer low groundwater storage conditions (Fig. F) and a clear increase in larger values, whereas drought-trigger scenarios are very similar compared to the baseline scenario.

The leverage of the modelled drought management scenarios is also reflected in the groundwater drought characteristics with drought deficit, i.e. the intensity of drought events being most sensitive (Fig. ). This intensification of groundwater drought events is largely due to an overall lower groundwater level in the karstic and porous module (Fig. ) following from the conjunctive use scenario. Detailed results highlight the strong negative difference between the baseline and the integrated management scenario (Figs. S9 and S10 in the Supplement). The overall influence on drought duration is positive, meaning shorter droughts, for most scenarios in the porous and fractured module. In the karstic module (and to some extent in the porous module) a larger spread of drought durations is found, which also reflects the increased sensitivity to groundwater demand under drought conditions for this specific aquifer type (Fig. ). Again, the largest differences are found for the conjunctive use scenarios compared to the trigger-driven scenarios (Figs. S9–S11 in the Supplement).

The distinct difference in leverage in fractured module between the trigger-driven and integrated drought management scenarios is mostly reflected in the drought frequency, as drought intensity and duration follow the same pattern – but less strongly negative/positive compared to the other groundwater modules. The smaller differences compared to baseline might be due to the overall higher groundwater storage levels in the integrated scenarios (Fig. ). The increased water supply and reduced water demand scenarios in/decrease drought frequency respectively. Integrated scenarios result in a large range in drought frequency with maintaining hands off flow scenarios slightly reducing overall frequency (Fig. S11).

Figure  shows the distribution of model performance over the calibration period according to different metrics. In all three catchments, we find consistent results between calibrating to fit discharge observations (KGE-Qs or NSE_log), groundwater observations (KGE-Gs) or a “Best Overall” (combined ranked performance criterion). The range in performance between the calibration approaches varies for the catchments and we find a trade-off between calibrations on either discharge, groundwater or both model outputs.

The first calibration strategy is using only discharge (based on KGE-Qs) and results in slim ranges for all catchments: Pang (0.75–0.81), Trent 0.76–0.84) and Derwent (0.77–0.82). These ranges are larger when fitting to NSE_log, Best Overall or groundwater-only calibration strategies, see first column in Fig. ). These discharge-only calibrations translate in considerably less reliable results for groundwater storage (KGE-Gs: 0.13–0.62). Low groundwater storage conditions are typically under-represented resulting in a lower KGE-Gs. In the NSE_log scores (middle panel in Fig. ) we see the same pattern with the best scores for calibration on discharge droughts (based on NSE_log) and KGE-Qs. Capturing the drought periods in discharge seems more appropriate when applying the NSE_log criteria that result in scores ranging between 0.61 and 0.71 for all catchments.

In groundwater-only calibrations, KGE-Gs for groundwater can be very high and narrowly confined with scores ranging between (0.62–0.95; see last column in Fig. ). As found before, discharge-only calibration approaches result in a larger range for KGE-Qs (0.27–0.73) for all catchments. NSE_log scores for discharge are largely negative, suggesting this method is less suitable to represent discharge droughts. The “Best Overall” approach that combines all the performance criterion results in larger performance ranges for both model outputs, although score ranges are largely acceptable for both discharge (0.55–0.79; first column) and groundwater (0.27–0.91; last column).

In Fig. , we show the time series of simulated discharge and normalised groundwater storage over the validation period 2004–2014 for the Pang catchment, for the four calibration approaches. Coloured lines are simulations from the top 50 calibration results based on NSE_log (pink), on KGE-Qs (purple), and on the Best Overall metric (green). The model calibrated to discharge observations (pink) has an overall good performance despite some exaggerated high flows within this period. The overall seasonality and recovery to normal flow conditions is particularly well-captured in 2007 and 2011 in this heavily-abstracted Chalk catchment. The model calibrated with “Best Overall” criterion produces less good simulations with frequent underestimations of low flows. This is however a feature of the Chalk only, as validation time series of the Trent show well-captured low flows (Fig. S12 in the Supplement). Observed low flows in the Derwent are consistently lower compared to simulations, which might be due to simulations flattening out on an ecological flow between 5 %–30 % of baseflow (Fig. S13 in the Supplement). Groundwater time series are remarkably similar between the KGE-Gs and “Best Overall” calibration in the Chalk (Fig. , others in Figs. S12 and S13). Both calibrations result in well-captured periods of declining and low groundwater storage in 2004–2006 and 2010–2012 with a slight overestimation in 2007–2008 compared to the mean observed groundwater storage (dotted).

Finally, we investigate the values of the top 50 performing parameter sets for the “Best Overall” calibration across the three catchments. In Fig.  each of these 50 parameter combinations is represented by a grey line. On top of these lines, box plots help visualise the spread of these “optimal” parameter values. Overall, we see that calibration highly constrains the three soil moisture parameters (WP, CR and FC) in all three catchments. In the Pang catchment, the non-linear flow (karst) and groundwater abstraction parameter (Dgw) are also well constrained; and so are discharge-outflow parameters sP1, sF1 and Dgw in the Trent and Derwent catchment. Interestingly, in the Derwent catchment even more parameters (i.e. Kfc, b, Rcap and Dsw) are constrained compared to the Pang and Trent, which is likely due to the upstream reservoir setup which heavily influences discharge in the Derwent.

Parameters that are hardly constrained by the calibration vary for the three catchments. With an upstream reservoir setup in the Derwent, there are just two almost unconstrained parameters (Qeco and sF2), i.e. where top 50 values are almost evenly spread over the national range used for sampling. In the Pang and Trent, soil parameters such as the unsaturated hydraulic conductivity and shape parameter (Kfc, b) are also unconstrained, as are the reservoir capacity, ecological minimum flow and surface water demand parameters (Rcap, Qeco, and Dsw respectively), which are all affecting the downstream reservoir.

The blue bars (or dots) in Fig.  indicate the expected ranges (or point-values) of the model parameters based on local information about the catchment characteristics (where available). In some cases (wilting point WP and unsaturated hydraulic conductivity Kfc in the Trent and Derwent catchment), these ranges are as large as the national ranges, derived from the CAMELS catchment attribute data and the European Soil database . For other parameters, such as groundwater storage data, calibrated values are similar to what we expect/know about the catchment. Data detailed for the Kennet Valley and Stafford basin is largely in agreement with the calibrated values. In the Pang, proportionate groundwater demand (Dgw) is calibrated consistently to expected value of 30 % (based on ). However, in the other catchments, calibration yields parameter values lower (Trent) and higher (Derwent) than expected based on local information. For the field capacity parameter (FC) we identified a disagreement between calibration and local information (from both CAMELS and the EU database) in all catchments. The calibrated Reservoir capacity (Rcap) in the Derwent catchment is also slightly larger than the capacity of the Ladybower reservoir from. For the Trent, calibrated ranges for Rcap are as large as the blue range, as this information is only available for the larger Trent catchment (CAMELS ID 28009; also shown in Fig. ).

Overall, SHOWER simulations perform well in terms of baseflow and groundwater storage simulations in the three diverse groundwater-rich catchments analysed here. Despite the difficulty of capturing droughts and low flows using a bucket model , we have shown that calibration focused on low flows (using NSE_log as performance criteria) lead to capturing low flows well also in the validation period.

The SHOWER performances reported in this paper are comparable or exceed the results of other hydrological models for the same catchments. The largest improvements (measured in KGE) are found for the surface water-dominated G2G model where SHOWER improves the negative (Pang) and average (Derwent) KGE values. Other rainfall–runoff models, such as GR4J and GR6J yield similar results for their selected best runs in the Pang and Derwent catchments , although the authors indicate that groundwater-dominated catchments are problematic to model well. The recent coupled DECIPHeR-GW model results are similar to those of SHOWER , showing that adding the elaborate groundwater representation in this new DECIPHeR version improves model results for these areas . Even for similar modelling performances, a key advantage of SHOWER is that it explicitly accounts for groundwater and surface water abstractions and reservoir influence, which introduces the possibility of testing the impact of management strategies, which is not possible using the previously mentioned models.

Similar to , SHOWER has the ability to simulate both discharge and groundwater output. Groundwater results are indicative of catchment's mean storage, as modelled groundwater storage is relative and spatially uniform (lumped). Despite this simplification, we show that groundwater time series can be well-captured using SHOWER (KGE: 0.62–0.95) for both combined and groundwater-only calibration criteria. We could however not compare these performances to those of other models, as published UK groundwater results are either not open-access and/or at the right scale . The dual model output does complicate a standard calibration process, as model users might want to identify priorities for their model use and could adjust the calibration accordingly to optimise performance or with the aim to find a single set of parameters. Presented calibration strategies do however not have substantial trade-off, as we have shown in the response-based evaluation that both surface water and groundwater can be well-captured in both wet and dry conditions . The range in performance shown in Fig.  illustrates how various calibration strategies differ within capturing the overall flow variability and low flow conditions. Peak flows are however less well-captured when looking at high flows in 2010 in Fig. , which is to be expected with groundwater modules focusing on base flow generation . It might also be a consequence of using KGE and logged NSE as calibration criteria, which do not specifically emphasise high flow conditions . Overall, benefits of the dual model output translate into a “Best overall” parameter set that can be used to produce both discharge and relative groundwater storage with reasonable confidence. This could overcome data availability issues in ungauged discharge catchments or unavailable (recent) groundwater level records.

Our results show that integrated drought management scenarios have the most leverage on the SHOWER model outputs, and particularly the conjunctive use scenario. Discharge and groundwater results are significantly altered by management scenarios, particularly during drought periods. The impact of modelling a particular management scenario dominates discharge and groundwater level simulations, as this impact exceeds model parameter uncertainty. Even though these are theoretical drought management simulations, results indicate a substantial influence regardless of the parameters used within the national range. In practise, there will be a combination of drought management strategies applied in a catchment at the same time and drought strategies will be adapted to reflect specific operations. For example, we applied a theoretical application of conjunctive use in which water use is fully integrated and non-restrictive. This approach is in line with large conjunctive use schemes where a range of water sources is used and the increased flexibility within a water distribution system increase drought resilience . The modelled surface water imports represent the common cross-company water transfers that are an additional tool to overcome (short-term) water shortages. These are commonly used when approaching low reservoir storage but likely used prior to reaching the 25th threshold as water is frequently transferred between regions to increase resilience high water demand or droughts . These local and regional water transfers can also be used to maintain ecological minimum flows . How a combination of these modelled measures translate to better protection of ecosystems can be complex to observe on the ground. It will also require more site-specific modelling efforts, particularly investigating the water quantity aspect of water resources management interventions, as maintaining specific surface or groundwater levels via conjunctive or augmentation schemes does not directly guarantee a “good” ecological status .

The notable influence of these integrated water management strategies on drought characteristics (Fig. ) is encouraging, but further research regarding the effectiveness of strategies is needed. Results indicate shorter drought durations by applying integrated management strategies with a mix of both intensified and relieved drought intensity particularly in the karstic and porous groundwater module. Additionally, the impact of demand measures that are widely applied in England, are most effective in a fast-responding shallow (fractured) groundwater module but less so in other groundwater modules, which emphasises the wider need for research. These measures are rarely modelled at scale or investigated in a larger policy context . Alternative strategies to identify management impact use paired events, and show the positive impacts of drought and flood measures implemented at location of two UK case studies . How current and future drought management strategies hold against future UK heatwaves, increased water demand and increased effort to maintain low flows is receiving increasingly more attention. Not only because extensive modelling work is required to assess drought resilience at scale , but also because results indicate that without “further interventions to reduce water demand and provide additional water supplies, there is a high risk of water shortages, in particular in the South East of England.”

SHOWER is designed for simulations of discharge, primarily baseflow, and indicative groundwater storage variations over time. SHOWER can be used as a screening tool to evaluate the impact of groundwater abstraction strategies on hydrological droughts. The model runs quickly in R (1–4 s per run on a Core i7 Intel laptop) and this provides the benefits of allowing on-the-fly calculations and what-if scenario analysis considering multiple combinations of parameters and/or management settings. SHOWER sits herein, in terms of mathematical complexity, between other bucket model approaches and (semi-)distributed models producing discharge . A crucial difference is that SHOWER aims to represent groundwater-rich areas and is particularly well-placed to analyse droughts and drought management strategies in regions with significant groundwater contributions to streamflow. Groundwater storage is represented in relative terms and lumped for each catchment, meaning that results are simplified compared to distributed groundwater models, such as , and . However, this simplification creates the opportunity to explore results droughts and management impact in more detail prior to investing in expensive detailed groundwater models. Moreover, the leverage of integrated drought management strategies shows the potential for SHOWER in decision-making processes and the modular, open-source model structure allows for adjusting SHOWER to local/relevant characteristics for a particular water management scenario/setting.

Possible model improvements to the current version might be relevant to users that focus on particular areas with impermeable surfaces or complicated land use areas (large urban areas), as SHOWER does not account for any differentiation in the soil characteristics or for urban water strategies (such as large sewage treatment works, ). Other hydrological models also deploy more sophisticated soil modules and we would recommend to either be aware of the simple setup or use the modular model structure of SHOWER to synchronize in/outputs with alternative models. For small to medium-sized catchments, we have found that SHOWER is quick to setup and can simulate both discharge and groundwater storage. However, the absence of a routing module affects its performance for large catchments (approximately ≥1000 km2). Again, we would advise model users to either insert a routing module into the modular open-source structure of SHOWER to account for this, or use the modular model structure to adapt SHOWER in alternative ways. With the specific modelling aim to represent baseflow and storage in groundwater-rich areas, we acknowledge that SHOWER is a social construct to further hydrological drought management strategies in groundwater-rich environments . However, the (open-source) modular structure of SHOWER, quick running time, and link with CAMELS-GB make for a versatile socio-hydrological model that could be applied to regions across the globe, using other CAMELS datasets to support groundwater management decisions on the ground.