Designation and trend analysis of the updated UK Benchmark Network of river ﬂ ow stations: the UKBN2 dataset

Observational trend analysis is fundamental for tracking emerging changes in river ﬂ ows and placing extreme events in their longer-term historical context, particularly as climate change is expected to intensify the hydrological cycle. However, human disturbance within catchments can introduce arti ﬁ cial changes and confound any underlying climate-driven signal. The UK Benchmark Network (UKBN), designated in the early 2000s, comprised a subset of National River Flow Archive (NRFA) stations that were considered near-natural and thus appropriate for identi ﬁ cation and interpretation of climate-driven hydrological trends. Here, the original network was reviewed and updated, resulting in the UKBN2 dataset consisting of 146 near-natural catchments. Additionally, the UKBN2 provides user guidance on the suitability of each station for the assessment of low, medium, and high ﬂ ows. A trend analysis was performed on the updated UKBN2 dataset and results show that while the strength and direction of changes are dependent on the period of record selected, previously detected patterns of river ﬂ ow change in the UK remain robust for longer periods ( > 50 years), despite the recent prevalence of extremes. Such a quality assured observational dataset will provide a foundation for future scienti ﬁ c efforts to better understand the changing nature of the hydrological cycle.


INTRODUCTION
It is expected that anthropogenic climate change will intensify the global hydrological cycle as the world continues to warm (IPCC ), thereby increasing the frequency and severity of extremes such as floods (Hirabayashi et al. ) and droughts (Prudhomme et al. ), although strong regional variability and uncertainties in projections exist (Arnell & Gosling ). The recent UK Climate Change Risk Assessment report (ASC ) identified both increased flooding and water scarcity among the UK's most important climate change risks. The notable hydrological volatility experienced in the early decades of the 21st century (Hannaford ; for a fuller description of these episodes see the National River Flow Archive (NRFA) website: http://nrfa.ceh.ac.uk/occasional-reports) has exposed the UK's vulnerability to hydrological extremes and thus there is a clear scientific and socio-economic need to understand the changing nature of these extremes.
There is a growing body of work using large ensemble modelling approaches suggesting extreme hydrological events can be attributable, in part, to a direct human influence on climate (e.g., Schaller et al. ). Given inherent uncertainties introduced through the climate-hydrology modelling chain, and the complex and still poorly understood role of catchments in modifying climate signals, observations remain the foundation for any scientific understanding on climate change impacts on river flows, particularly when justifying costly adaptation plans. However, there are also many challenges involved with detection of a robust longterm climate change signal within observed river flow timeseries (Hannaford ). The UKBN is of fundamental importance in this regard, given the high population density and long history of settlement and water exploitation in the UK compared to many other countries; human influences on river flow regimes are pervasive, and in many catchments changes in longterm runoff patterns bear little relation to climate variability (Hannaford & Marsh ). Benchmark catchments can be considered reasonably free from human disturbances such as urbanisation, river engineering, and water abstractions, and hence can be used for detection of climate-driven changes in river flow. The first iteration of the UKBN, henceforth UKBN1, was designated 15 years ago by Bradford & Marsh () and included 122 catchments that met four primary criteria: (i) relatively natural flow regimes, (ii) good and consistent hydrometric data quality, (iii) relatively long records (ideally > 25 years) and (iv) were representative of UK hydroclimatic conditions with good geographical coverage. The core aim of the UKBN is to strengthen national capability to identify and quantify long-term trends and variability in runoff patterns and hydrological extremes. As well as being valuable for the national and international research community, this information is potentially useful for a wide range of practical applications including strategic water resources planning, environmental regulation, flood risk and engineering design, and climate change adaptation planning.
Since the designation of the network, it has been used extensively in trend studies on changes in UK runoff, low flows and droughts, high flows and floods, and seasonal nrfa.ceh.ac.uk/peak-flow-data) which includes gauging station rating curves, improved knowledge of artificial influences (AIs) such as water abstractions and discharges, and increased NRFA spatial and statistical analysis capabilities (Dixon et al. ). Given the extent and diversity of both qualitative and quantitative information, as well as the need to exercise expert judgement in many places, a completely objective application of the benchmark criteria was not possible, as was the case for the original UKBN1 designation. Nevertheless, the decision-making process was supported by a systematic framework underpinned by key evidence sources (Table 1). In step one of the appraisal, stations contained within UKBN1 were allocated one of three initial categories: endorse, review, or omit based on the review process shown in Table 1(A-E). An additional 54 stations that potentially met benchmark criteria were also considered, as they now have a record length >25 years. Of the original 122 stations, 67 were endorsed, 48 required further review, and seven were omitted, resulting in a total of 176 (including the 54 'Candidate' stations) considered in the overall UKBN2 appraisal.
It was apparent in the early stages of the review that compromises were needed in particular regions to achieve an adequate density of benchmark catchments. This primarily reflects both the ubiquitous nature of AIs on flow regimes, and the inherent difficulties of hydrometric measurement in the extreme flow ranges at many UK gauging stations; very few gauging stations can be considered truly 'full range' (Marsh ). For example, at low flows, hydrometric uncertainty arises due to insensitivity of measuring structures, or wide scatter in spot flow measurements (gaugings) used to derive rating curves, e.g., due to summer weed growth. Low flows are also the most heavily impacted by substantial surface and/or groundwater abstractions within the catchment. For high flows, common issues include unmeasured bypass flow and non-modularity (drowning) at gauging structures (Herschy ), or simply an insufficiency of gaugings to accurately define the high flow rating curve.
Given these challenges, the original aspiration (Bradford & Marsh ) of full-range benchmark catchments was a major constraint on the network. Recognising this limitation, and the often different uses and user communities for low flow and high flow assessments (e.g., Hannaford et al. b), the UKBN2 model advocates a classification system that allows 'sub-networks' to be defined. To facilitate this, and help the user community assess the utility of individual benchmark station records in the presence of these hydrometric challenges, their suitability for analysis at low, medium, and high flow was evaluated.
Any evaluation of the ability of a station to effectively measure extreme flows requires local knowledge of site and catchment conditions.
Step two in the benchmark review process ( Bringing knowledge together from steps one and two, the final step (Table 1G) assigned each station a benchmark score based on suitability for analysis of low, medium, and high flows (2 ¼ suitable, 1 ¼ caution, and 0 ¼ not-suitable). Thus a station scoring a maximum of 6 means it is suitable for use across the full flow regime.
Where a station scores 1 or 0 for a category, a brief benchmark qualifier is provided to help end users understand why the time-series might not be suitable for analysis or requires caution, if for example, water abstractions, poor high flow performance/bypassing, or artificial regulation of flows from hydroelectric power schemes were particularly prevalent.

The new UKBN2 dataset
The UKBN2 appraisal identified 146 of the 176 stations under review as qualifying for benchmark status ( Figure 1 and Table 2  implemented to ensure reproducibility of subsequent analyses through time and we envisage that on each major update, a routine trend analysis using the methodology outlined below will be undertaken.

TREND ANALYSIS METHODS
The second aim of this paper is to develop a standardised trend analysis procedure to apply routinely to the Benchmark Network, based on established methods within the hydroclimatic literature, with a first application on the newly designated UKBN2 dataset.
Various trend assessment methods have been applied to UKBN1 previously. Here, we set out the following as a rigorous, standardised approach focusing on three components aimed at understanding spatio-temporal changes in river flow: 1. Trend analysis using two fixed periods (short and long) to identify the spatial nature of changes in river flows.
2. Assessment of temporal variability of changes in light of the known influence of DCV.
3. Investigation of persistence of trends for the full available time-series.

Hydrological indicators and catchment selection
For each year, a set of 12 hydrological indicators used in

Trend analysis tests
Evidence for monotonic trends was assessed using the  Tables 3 and 4 counts both statistically significant results from the traditional MK test (for non-significant serially correlated series) and for significantly serially correlated series using BBS with L ¼ 4 and is also used for reporting statistically significant trends in the maps in Figures 3 and 4.
There is much debate in the field of hydroclimatology around trend significance testing such as the existence of long-term persistence, which could introduce a statistically significant trend when none is present (Cohn & Lins ;

Fixed period trends
In low, medium, and high flow indices for the 1985-2014 short period, positive trends are prominent (Table 3). Over There are several cases where the number of statistically significant increasing/decreasing trends was reduced when block-bootstrapping was applied to serially correlated series (Tables 3 and 4).

Temporal variability analysis and persistence of trends
While it is necessary to analyse trends using fixed periods for a relative comparison of direction, magnitude, and spatial patterns, these are just snapshots of the temporal evolution of changes over timeas demonstrated by the marked differences in trends between the two fixed periods (i.e., English lowlands are in the À10-30% range for several catchments (Figure 3(e)), especially for summer ( Figure   4(c)), and might be important for water management.
Wilby () showed the signal-to-noise ratio for basins in the UK is low, particularly in summer, and that robust statistically detectable trends are not expected for several decades yet. This is further highlighted in Figure 5(a) with strong evidence of DCV, and hence trends are sensitive to the period of record analysed ( Figure 5(b)). This is most prevalent for records beginning in the 1960s and 1970s, which is the case for the majority of UK trend studies as hydrometric network expansion coincided with a period of a particularly high degree of natural variability.
In addition to the previous limitations, few catchments in the densely populated region of southern and eastern England can be considered pristine in the strictest sense, so caution must be exercised in interpreting changes in low flows in this region. Nonetheless, the consistent temporal and spatial pattern across low, medium, and high flow indices ( Figure 5, left column) is encouraging and suggests that even in the English lowlands river flows are generally reflecting changes driven by climate, rather than from artificial sources (e.g., from groundwater and/or surface water abstraction) which, while controlled as far as possible in the benchmark designation, cannot be ruled out in the catchments flagged as 'caution'. However, it is challenging based on these results alone to provide clear guidance regarding potential long-term implications for water resources management, so future work that combines innovative observational and modelling approaches using several lines of hydroclimatic inquiry is still needed. It is also noted that the majority of studies examine changes in low flows, rather than actual drought 'events', and so such event-based analyses should be another research priority. catchments >1,500 km 2 so any study, including the trend analysis here, will be biased towards medium and small catchments, particularly in the south and east of England as abstractions and discharges are less prevalent in headwater catchments. There is also a dearth of very small catchments; only four catchments within the UKBN2 dataset have areas <10 km 2 and only one of those can be considered upland (elevation >300 m a.s.l.). Therefore, processes operating only at these scales would not be captured.
On the other hand, RHNs can provide a near-natural baseline for comparing with human-influenced sites (e.g., using paired 'impacted' catchments as in Prosdocimi et al. ) or for modelling studies, so can play a vital role even in efforts to quantify human disturbances on the hydrological cycle.
The second iteration of the UKBN has made several improvements since UKBN1, but there are many potential further improvements that could be made to future iterations of the dataset and to how users access it. For example, we anticipate future analytical efforts will undertake comprehensive homogeneity testing and infilling, while a particular focus will be efforts to improve the assessments of AIs. One of the most challenging aspects of the UKBN update was the fragmented quality and availability of information on AIs, especially access to water abstractions and discharges. While some datasets were consulted (e.g.,

CONCLUDING REMARKS
The first designation of the UKBN has proven a valuable dataset that has fed into many national and international scientific studies, several of which are relied upon for making policy and water management decisions on future flood design and long-term drought planning. Results from the trend analysis of the updated UKBN2 have reinforced previous findings. We recognise the UKBN will always remain a work-in-progress as new information about gauging stations and the catchments they drain comes to light, or new techniques for assessing benchmark suitability developed. A benchmark version control system has been instigated to ensure minor and major network changes are recorded in a transparent way, the datasets are easily accessible, and studies using previous versions reproducible.
Further information about the UKBN2 and how to access the data can be found here: http://nrfa.ceh.ac.uk/benchmark-network.
A community effort involving both those who collect the data (Measuring Authorities) and those who use it (e.g., researchers and practitioners), would make the process of UKBN evolution and updating more efficient and comprehensive. We hope by releasing the UKBN2 we present an opportunity for the hydrological community to provide ideas, novel methods, and feedback on the current version. . We therefore invite users to provide information on these catchments, or others that may be candidate benchmark catchments, via contacting the NRFA (nrfa@ceh.ac.uk). We are also interested to understand the range of uses of the network, and invite users to engage with the NRFA team about current and future applications of the dataset.