Simulation of river flow in Britain under climate change: Baseline performance and future seasonal changes

Climate change is likely to manifest in river flow changes across the globe, which could have wide‐ranging consequences for society and the natural environment. A number of previous studies used the UK Climate Projections 2009 (UKCP09) to investigate the potential impacts on river flows in Britain, but these projections were recently updated by the release of UKCP18, thus there is a need to update flow studies. Here, the UKCP18 Regional (12 km) projections are applied using a national‐scale grid‐based hydrological model, to investigate potential future changes in seasonal mean river flows across Great Britain. Analysis of hydrological model performance using baseline climate model data (1980–2010) shows relatively good agreement with use of observation‐based data, particularly after application of a monthly precipitation bias‐correction. Analysis of seasonal mean flow changes for two future time‐slices (2020–2050 and 2050–2080) suggests large decreases in summer flows across the country (median −45% by 2050–2080), but possible increases in winter flows (median 9% by 2050–2080), especially in the north and west. Information on the potential range of flow changes using the latest projections is necessary to develop appropriate adaptation strategies, and comparisons with previous projections can help update existing plans, although such comparisons are often not straightforward.


| INTRODUCTION
Climate change will affect the hydrological cycle, and likely manifest in changes to flow regimes in rivers across the globe (Jiménez Cisneros et al., 2014). Such flow changes can have important consequences, relating to changes in seasonal flow patterns as well as changes in the frequency or magnitude of extreme flows. Flow regime changes can affect water quality, ecology, and energy production for example.
A range of studies have looked at the potential impacts of climate change on river flows in Britain (Watts et al., 2015). Studies often use the UK Climate Projections 2009 (UKCP09; Murphy et al., 2009), which provided a range of alternative products including Probabilistic Projections, an 11-member perturbed parameter ensemble (PPE) of a 25 km Regional Climate Model (RCM), and a weather generator (e.g., Kay & Jones, 2012). Table 1 summarizes three studies which used UKCP09 to investigate the potential impacts of climate change on seasonal river flows in the UK (Christierson et al., 2012;Prudhomme et al., 2012;Sanderson et al., 2012). In general, studies suggest decreases in summer flows and possible increases in winter flows.
The UK Climate Projections 2018 (UKCP18; Lowe et al., 2018) provide an update to UKCP09, and there is a need to update corresponding simulations of potential future changes in river flows.
One such update was provided by Kay et al. (2020), who applied the UKCP18 Probabilistic Projections with catchment-based models for 10 catchments in England, to look at the range of potential changes in measures of mean, median, high and low flow. The central estimates of change showed reductions in median and low flow in all catchments, with reductions in mean flow in eight catchments, and reductions in high flow in two to three catchments (depending on the emissions scenario). In all 10 catchments for all four flow measures, the central estimate of change from UKCP18 was similar to that from UKCP09 (A1B emissions), but the uncertainty range from UKCP18 was greater than from UKCP09.
Here, time-series data from the UKCP18 Regional (12 km) projections are used, with a fully distributed grid-based hydrological model, to look at seasonal mean flow changes across GB, thus providing an update to the national results of Prudhomme et al. (2012). The methods are described (Section 2), with results (Section 3), discussion (Section 4) and conclusions (Section 5).

| Hydrological model and observation-based driving data
The Grid-to-Grid (G2G) is a national-scale grid-based hydrological model for Great Britain that usually operates on a 1 km grid at a 15-minute time-step, and is parameterized using digital datasets (e.g., spatial soil grids) rather than through catchment calibration (Bell et al., 2009). The optional snow module (Bell et al., 2016) is applied here. G2G simulations of river flow perform well for a wide range of catchments (Bell et al., 2009(Bell et al., , 2016Formetta et al., 2018;Rudd et al., 2017), including those with a high proportion of baseflow (see Figure 5 in Bell et al., 2009), and particularly where the flow regime is relatively natural. Artificial influences such as abstractions and discharges are not generally included, so the model essentially simulates natural, rather than gauged, flows.
Input gridded time-series of precipitation and potential evaporation (PE) are required, as well as temperature for the snow module. An observation-based simulation was performed for December 1980-November 2010 (hereafter 'SIMOBS') using: • Daily 1 km grids of precipitation (CEH-GEAR; Tanguy et al., 2016), divided equally over each model time-step within a day; • Monthly 40 km grids of PE for short grass (MORECS; Hough & Jones, 1997), divided equally over each model time-step within a month and copied down to the 1 km grid; • Daily 1 km grids of min and max temperature (Met Office, 2019), interpolated through the day using a sine curve (Kay & Crooks, 2014).
The simulation was initialized using a states file saved at the end of a prior observation-based simulation (January 1970-November 1980). While the model produces 'river flow' for every 1 km 'land' box, only data from non-tidal grid boxes with a catchment area of at least 50km 2 are analysed (hereafter 'river pixels'). Model outputs include gridded time-series of monthly mean river flows, and timeseries of daily mean river flows for selected 1 km pixels corresponding to gauged catchments with flow data in the National River Flow Archive (www.ceh.ac.uk/data/nrfa/).

| Climate change projections and their application
UKCP18 provides information on potential changes in a range of climate variables over the 21st century, via a number of different products (Murphy et al., 2018). These include the UKCP18 Regional T A B L E 1 Studies using UKCP09 to investigate the impacts of climate change on seasonal river flows across Britain  (Riahi et al., 2011). Ensemble member 01 uses the standard parameterization. The data are available re-projected from the native climate model grid to a 12 km grid aligned with the GB national grid. The reprojected daily precipitation and daily min and max temperature are used here.
There are (generally positive) biases in monthly RCM precipitation (see Figure 4.4 in Murphy et al., 2018) so the data are bias-corrected as in Guillod et al. (2018); grids of monthly correction factors are derived by comparing baseline mean monthly precipitation totals (from each PPE member separately) against those from CEH-GEAR averaged up to the 12 km RCM grid, then the factors are smoothed using weights in a 3 × 3 neighbourhood ( Figure S1). There are a number of alternative bias-correction methods of varying complexity (Fung, 2018), and many issues and assumptions inherent in biascorrection (e.g., Ehret et al., 2012) which can potentially introduce artefacts into the 'corrected' data (Maraun et al., 2017). So the approach here is deliberately simple, aiming to correct seasonal mean biases while not adversely affecting higher-order moments. The biascorrected precipitation are downscaled to the 1 km grid using a spatial weighting derived from 1 km standard average annual rainfall patterns (Bell et al., 2007), and temporally downscaled as for observed rainfall (Section 2.1).
No bias-correction is applied to RCM temperature, as the PPE range encompasses monthly observations relatively well (see PE for short grass is not available directly, so is estimated from other (re-projected) daily climate variables using a formulation which replicates MORECS as closely as possible; essentially Penman-Monteith PE (Monteith, 1965) with some minor modifications, including an interception correction   in Murphy et al., 2018). PE between spring and autumn typically increases by 10-20%, with increases of 30% or more in south/east England for some ensemble members. While winter PE also generally increases, by 5% on average and up to 35% in some locations for some ensemble members, there are decreases of as much as −15% in places, but these are not likely to be important as winter PE is fairly low.
Each RCM-based G2G simulation (hereafter 'SIMRCM') was initialized in Dec 1980 using the same states file as SIMOBS (Section 2.1), and run through to Nov 2080.

| Baseline performance assessment
As development of weather features in the RCM PPE will not follow the observed weather over the baseline period, the performance assessment uses measures derived from flow duration curves and seasonal mean flows, thus comparing statistical characteristics rather than day-to-day equivalence.
The flow duration curve assessment uses simulated daily mean flows for 1 km pixels corresponding to a set of 96 gauged GB catchments; those within the UK benchmark network (Harrigan et al., 2018) which have a catchment area of at least 50km 2 and less than 20% missing data in the baseline period ( Figure S2a). The To provide a broader performance assessment across GB, grids of baseline seasonal mean flows from the SIMRCM ensemble are compared to corresponding grids from the SIMOBS run. For each river pixel and each season, a value p is assigned as performance of individual ensemble members varies around that of the pooled performance, particularly without bias-correction when some members show majority positive flow biases (e.g., 11) or more negative biases (e.g., 10, for low flows in particular). The biascorrection reduces the variation in performance between ensemble members, unsurprisingly since monthly mean precipitation in each member is separately corrected to the observed precipitation.
Maps of the flow duration curve assessment for the SIMOBS run and the pooled SIMRCM ensemble ( Figure S3) show the variation in performance across the country. In particular, the low flow volume is more likely to be under-estimated in north/west England and overestimated in south/east England and western Scotland in each case.
Without bias-correction, many catchments show over-estimation of median and high flows in particular; this is reduced by bias-correction.
The seasonal mean flow assessment also shows that biascorrection improves performance of the SIMRCM ensemble ( Figure 2). Without bias-correction, the SIMRCM range excludes the SIMOBS flow for well over 40% of pixels in winter and spring, over 15% in autumn, but just less than 10% in summer. With bias-correction, this reduces by at least half in winter, spring and summer, but only reduces slightly in autumn. Furthermore, with bias-correction the performance measure p is relatively large everywhere in each season (except for some pixels in the Scottish highlands in summer), indicating that the SIMRCM ensemble gives a relatively robust indication of SIMOBS. Measure p is generally lowest in summer, when the SIMOBS flows will typically be lowest, making it harder for the SIMRCM range to be much smaller than the SIMOBS flow.  Table 1).

|
Possibly large future reductions in summer flows could have a range of implications. For example, where rivers provide important contributions to drinking water or irrigation there may be a greater chance of supply limitations in future, and adaptation measures will be needed to manage this (Harris et al., 2013). Low flows can also adversely affect water quality by reducing dilution of substances (Charlton et al., 2018;Nilsson & Malm Renöfält, 2008), and affect riverine ecology through changes in the physical extent or conditions of aquatic habitat (Laize et al., 2014;Rolls et al., 2012).
Possible future increases in winter flows could also have a range of implications. For example, it may lead to an increased frequency and/or magnitude of floods (Charlton et al., 2006). As well as the direct effects of floods, events can cause pollution and sedimentation through the addition of large amounts of organic and inorganic matter to rivers (Nilsson & Malm Renöfält, 2008;Ponting et al., 2020). Higher flows can also adversely affect riverine ecology, as can changes in the seasonal patterns of river flows (Laize et al., 2014).
Only one hydrological model has been applied, although a number of studies have suggested that the climate model is typically a larger source of uncertainty (e.g., Krysanova et al., 2017). The correction of monthly mean precipitation biases makes the baseline performance of the pooled climate ensemble more similar to that using observed data to drive the hydrological model, giving some confidence in use of the ensemble for projecting future changes in flows. However, the results use data from only one GCM/RCM combination-the latest Hadley Centre GCM (HadGEM3) and its regional equivalent-albeit as a PPE.
Other CMIP5 climate models within the UKCP18 Global projections, and (to a lesser extent) the UKCP18 Probabilistic projections, tend to give smaller decreases (or increases) in summer precipitation and a wider range of changes in winter precipitation than the Regional projections applied here (see Figure 5. The projected changes in flow are likely to be predominantly driven by the changes in precipitation (drier summers and wetter winters; Section 2.2), with additional spatial variation due to catchment properties, but the increases in PE could also be important (e.g., Kay & Davies, 2008), and in some locations flows could be additionally affected by changes in temperature via its effect on snow (e.g., Kay, 2016). Projected future changes in evaporation often get less attention than precipitation, yet evaporation is a key part of the hydrological cycle. There is uncertainty in the PE inputs necessary for hydrological modelling, especially under climate change (Kay et al., 2013). While the future PE here includes an increase in stomatal resistance due to stomatal closure under higher CO 2 concentrations, it does not include a potential increase in leaf area (and therefore number of stomata) due to carbon fertilization (Rudd & Kay, 2016).
Potential future changes in land-cover and abstractions/discharges are also not included.

| CONCLUSIONS
Applying a national-scale grid-based hydrological model with ensemble climate data from the UKCP18 Regional projections has suggested future decreases in summer flows across the Great Britain (median −45% by 2050 to 2080), but possible increases in winter flows (median 9% by 2050 to 2080), especially in the north/west. Such changes in flows could have significant implications, both for the natural environment and for society.
Information on the potential range of changes in river flows from the latest climate projections is necessary to develop appropriate adaptation strategies for water management (e.g., HR Wallingford, 2020), and comparisons with results from previous climate projections can help in the updating of existing plans. However, comparisons are often not straightforward, due to differences in the setup of projections (e.g., UKCP09 typically used a 1961-1990 baseline but the UKCP18 RCM data only starts in Dec 1980), the way projections are applied (e.g., bias-correction methods etc.), and the models used. Not all of these differences are avoidable, due to the evolution of science and methods.
Further work will investigate potential future changes in high and low flow frequency, as well as soil moisture, which could have important consequences for agriculture (Samaniego et al., 2018) and subsidence hazard (Pritchard et al., 2015). In addition, the UKCP18 Local (2.2 km) projections provide data from a 12-member convectionpermitting model ensemble, which shows greater increases in winter mean precipitation than the Regional ensemble (Kendon et al., 2019), so could lead to differences in the simulated impacts on river flows (e.g., Kay et al., 2015).

ACKNOWLEDGEMENTS
This work was supported by the Natural Environment Research Council award number NE/R016429/1 as part of the UK-SCAPE programme delivering National Capability. Thanks to UKCEH colleagues Emma Robinson and Rhian Chapman, for work on the estimation of PE from climate model data.

DATA AVAILABILITY STATEMENT
Data will be made available via EIDC at some point during in 2021, but before then they are available from the authors upon reasonable request.