To provide consistency with CAMELS-GB v1, we derive the daily timeseries from the same underlying meteorological datasets; rainfall from UK Centre for Ecology & Hydrology Gridded Estimates of Areal Rainfall (CEH-GEAR) dataset, potential evapotranspiration and temperature from the Climate hydrology and ecology research support system (CHESS). These datasets were selected due to their high spatial resolution (1 km2), long temporal coverage (>50 years) and basis on the UK climate monitoring network. However, these meteorological datasets are no longer consistently updated and do not cover the full time period required. Consequently, in CAMELS-GB v2 we also provide meteorological timeseries of catchment average precipitation, potential evapotranspiration and temperature from a new UK dataset of gridded climate observations (HadUK-Grid; Hollis et al., 2019). A national comparison of the different products is shown in Fig. 1, while a more detailed comparison of the different products for two catchments can be found in Figs. S1 and S2.

Daily rainfall timeseries were derived from two national products; the CEH Gridded Estimates of Areal Rainfall dataset (CEH-GEAR; Keller et al., 2015; Tanguy et al., 2021) and the HadUK-Grid dataset (Hollis et al., 2019). Both consist of 1 km2 gridded estimates of daily rainfall and are based on quality-controlled precipitation data from the Met Office UK rain gauge network. However, the two datasets cover different time periods and use different interpolation methods. CEH-GEAR covers 1890–2019 whereas HadUK-Grid rainfall is available from 1836–2023. The CEH-GEAR dataset uses natural neighbour interpolation, whereas HadUK-Grid uses inverse-distance weighted interpolation to generate the daily rainfall grids. Given the similarities in their underlying datasets, the difference between the two rainfall products is small for most CAMELS-GB catchments (Figs. 1a and S1). However, there can be differences of up to 20 % in mean annual rainfall totals and larger differences in daily totals for individual catchments (Fig. S2).

Daily temperature timeseries were derived from the Climate Hydrology and Ecology research Support System meteorology dataset (CHESS-met; Robinson et al., 2017a) and the HadUK-Grid dataset (Hollis et al., 2019). CHESS-met contains 1 km2 gridded estimates of daily mean air temperature (K) from 1961–2019 derived from the Met Office Rainfall and Evaporation Calculation System (MORECS) dataset (Hough and Jones, 1997). MORECS is a 40 km gridded dataset of daily temperature derived from Met Office synoptic stations. For the temperature data in CHESS-met, the MORECS temperature data was interpolated from 40 km resolution to 1 km resolution using a bicubic spline and then the temperatures were adjusted to the elevation of each 1 km grid using the same lapse rate. HadUK-Grid contains 1 km2 gridded estimates of daily maximum and minimum air temperature (°C) from 1960–2023 derived by interpolating temperature observations from climate observing stations in the Met Office's Integrated Data Archive System (MIDAS). In the HadUK-Grid daily climate variables, the maximum air temperature is measured between 09:00 UTC on day D and 09:00 UTC on day D+1, while the minimum air temperature is measured between 09:00 UTC on day D-1 and 09:00 UTC on day D (Robinson et al., 2023c). Therefore, daily mean temperatures have been calculated by averaging maximum air temperature on day D and minimum air temperature on day D+1 to ensure both values represent the same 24 h period for each day. On average, the difference in mean daily temperature between the two products is relatively small (0.14 °C; Fig. 1d); however, differences can be larger for individual timesteps (Figs. S1 and S2).

Daily potential evapotranspiration (PET) timeseries were derived from the Climate Hydrology and Ecology research Support System Potential Evapotranspiration dataset (CHESS-PE; Robinson et al., 2017b, 2023b) and the Hydro-PE HadUK-Grid dataset (Hydro-PE; Brown et al., 2023; Robinson et al., 2023c). Both datasets consist of daily 1 km2 gridded estimates of potential-evapotranspiration for Great Britain calculated using the Penman–Monteith equation for well-watered grass. They also both provide daily potential evapotranspiration with (PETI) and without (PET) an interception correction. Core differences between the datasets are that the PET datasets cover different time periods; 1969–2022 for Hydro-PE and 1961–2019 for CHESS-PE. They also provide different PET estimates, with the Hydro-PE HadUK-Grid dataset providing higher mean annual estimates of PET (on average 0.1 mmd-1 higher across the CAMELS-GB catchments) and PETI (on average 0.2 mmd-1 higher across the CAMELS-GB catchments) for most CAMELS-GB catchments (Fig. 1b and c). This is due to differences in the underlying data and methodologies used to derive the PET estimates. CHESS-PE is derived from CHESS-met variables (Robinson et al., 2017a) that have been downscaled from the Met Office Rainfall and Evaporation Calculation System (MORECS) dataset (Hough and Jones, 1997), whereas Hydro-PE is derived from HadUK-Grid meteorological data (Hollis et al., 2019). Wind speeds are higher and specific humidity is lower in the HadUK-Grid dataset, and many of the variables in the Hydro-PE HadUK-Grid dataset have been temporally downscaled from monthly to daily using a simple smooth interpolation (for a full discussion of the differences, see Sect. 5.1 in Robinson et al., 2023c). This leads to different estimates of daily PET and PETI between the different datasets across all CAMELS-GB catchments (Figs. S1 and S2, see Sect. 3.3 for guidance on dataset selection).

As the meteorological timeseries from HadUK-Grid covers the full time period, we use these data to derive the climate catchment attributes described in Sect. 5.2.

Daily streamflow data for the 671 gauges were taken from the UK NRFA on the 7 January 2025 (https://nrfaapps.ceh.ac.uk/nrfa/nrfa-api.html, last access: 22 January 2025). Streamflow data on the NRFA are provided by measuring authorities who operate the river flow monitoring network, including the Environment Agency (EA) in England, Natural Resources Wales (NRW) in Wales and Scottish Environmental Protection Agency (SEPA) in Scotland. The streamflow data undergo quality control by the measuring authorities and the UK NRFA before being uploaded to the NRFA site. This quality control process removes flow values that are perceived to be erroneous, while retaining as complete a series as possible through the use of a range of infilling techniques (Dixon et al., 2013). Streamflow data in CAMELS-GB v2 are provided as volumetric discharge (m3s-1) and specific discharge (mmd-1).

Figure 2 shows the daily flow data availability for all gauges contained in the CAMELS-GB v2 dataset (Fig. 2a) and how this availability changes over time (Fig. 2c). Nearly all (666) of the gauges have at least 20 years of daily flow data, and 86 % (577) of the gauges have at least 40 years of daily flow data (Fig. 2a). Overall, there is good spatial coverage of long flow time series across Great Britain, with slightly shorter time series concentrated in the north, Midlands and south-east GB. Data availability increases over the time period with 60 % of the daily flow data available in 1970, peaking to 99 % in the early 2000's and slightly dropping to 96 % by 2022 (Fig. 2c).

Hourly rainfall timeseries are derived from two national products so users have the choice from multiple products.

The UK Centre for Ecology and Hydrology Gridded Estimates of hourly Areal Rainfall for Great Britain (CEH-GEAR1hr; Lewis et al., 2018, 2022) consists of 1 km2 gridded estimates of hourly rainfall from 1990–2016. The hourly rainfall estimates are derived from the temporal disaggregation of the CEH-GEAR daily data (described in Sect. 3.1.1) using hourly gauge data from the Met Office, EA, NRW and SEPA. The hourly gauge data are quality controlled to identify and reject erroneous hourly values in the gauge rainfall input dataset by comparing the gauge data with the CEH-GEAR daily dataset and by implementing a series of quality control tests (Lewis et al., 2018). The nearest neighbour interpolation methodology was used to generate the gridded hourly estimates which were subsequently used to disaggregate the daily data.

Hourly rainfall time series are also derived from the Gauge-Radar Precipitation Dataset (1 h and 1 km) for Great Britain, GRaD-GB(1H1K), which takes advantage of the accuracy of gauge rainfall and the spatial information of radar rainfall field (Qiu et al., 2026a). The dataset consists of 1 km2 gridded estimates of hourly rainfall for Great Britain from 1 January 2006 to 31 December 2023 and is produced by blending 5 min NIMROD composite radar rainfall (Met Office, 2024) with sub-hourly rainfall observations of ∼1800 rain gauge network from the UK Met Office, EA, NRW and SEPA. To produce the hourly rainfall dataset, the radar rainfall and sub-hour rainfall observations are first aggregated to hourly data. Then a quality control framework is applied to improve the underestimation (radar beam blockage) and overestimation (radar malfunction, ground clutter, and random noise) issues in the radar rainfall data (Qiu et al., 2026b). The quality control procedure that was employed in CEH-GEAR1hr (Lewis et al., 2018) is applied to the gauge rainfall data. A Gauss Blending method was then used to merge the radar rainfall with gauge rainfall. Maintenance work on the rainfall radars means that 3.5 % of the hourly timeseries are missing for the CAMELS-GB catchments due to missing radar data. Analysis of GraD-GB(1H1K) shows that the dataset can capture extreme rainfall events missed by rain-gauges (i.e. severe flash flooding in Coverack, Cornwall, 18 July 2017) (Qiu et al., 2026a).

A comparison of the hourly rainfall timeseries from CEH-GEAR1hr and GRaD-GB(1H1K) (Fig. 3) shows that on average, GRaD-GB(1H1K) hourly rainfall estimates are 10 % higher than CEH-GEAR1hr (range of -25 %–65 %) when calculating the average hourly rainfall using the full timeseries. There is more variability between the two products in the north of Great Britain (Fig. 3a). The higher average rainfall for the GRaD-GB(1H1K) dataset is partially explained by a higher proportion of wet hours (an hour with >0.1 mm of rainfall) in GRaD-GB(1H1K) (Fig. 3b), however, when the average hourly rainfall is calculated using only hours with >0.1 mm of rainfall, the relationship reverses and CEH-GEAR1hr has higher average hourly rainfall despite a lower proportion of wet hours (Fig. S3). The CEH-GEAR1hr data are based on gauge data corrected to the daily total and it specifically seeks to preserve intense sub-daily rainfall characteristics in the interpolation method it uses. Therefore, it has fewer wet hours and more intense hourly rainfall than GRaD-GB(1H1K). There is also a relationship with elevation where catchments with lower median elevation have higher average rainfall in the GRaD-GB(1H1K) dataset compared to catchments with higher median elevation (Fig. 3c). An example of the differences between the hourly rainfall datasets for individual catchments is shown in Figs. S4 and S5.

Hourly river flows and levels are provided for 664 and 570 gauges respectively in CAMELS-GB v2 from 1 October 1990–30 September 2022. These hourly river flow and level data combined with the hourly rainfall data provide a wealth of additional information beyond the daily data, particularly for flood analyses, convective storm responses, and other short-duration extremes. Below, we describe the source datasets of the hourly hydrological timeseries, how the timeseries were produced and the quality control applied to the dataset. Further details on the quality flags can be found in the Supplement (Sect. S1, Fig. S6 and Tables S1–S3).

Sub-daily river flows and levels are collected by the measuring authorities. The level data were obtained from SEPA via the timeseries data service (https://timeseriesdoc.sepa.org.uk/, last access: 23 September 2025), from EA primarily through the Hydrology Data Explorer (https://environment.data.gov.uk/hydrology, last access: 23 September 2025) and, where unavailable, with staff assistance, and from NRW entirely with staff assistance. The hourly flow data were obtained from the same sources but derived from the UK-Flow15 dataset (Fileni et al., 2025, 2026a). UK-Flow15 is a quality-controlled 15 min flow dataset for the UK, using records from over 1300 gauging stations including the EA, SEPA and NRW, in addition to the Department for Infrastructure in Northern Ireland and the UK Centre for Ecology & Hydrology.

Here, the 15 min flow and level data have been aggregated to hourly using a next-hour resample (e.g. 10:00 UTC flow value is the mean of flow recorded at 09:15, 09:30, 09:45 and 10:00 UTC). Most gauges (650) have at least 20 years of hourly flow data and 86 % (579) of the gauges have at least 30 years of hourly flow data (Fig. 2b). There is good spatial coverage of long hourly flow time series across Great Britain, with shorter time series concentrated in the Midlands and south-east of Great Britain. Similar to the daily data, there is greater missing flow data in the earlier part of the record (1990–1995, Fig. 2c). There is also good availability of level data where 80 % (542) of gauges have at least 20 years of hourly level data and 74 % (499) of gauges have at least 30 years of hourly level data.

The hourly flow and level timeseries are also provided with quality control flags. The hourly flow and level data have been quality controlled using both visual/manual inspection and automated quality control, including novel quality assessment techniques to assess the plausibility of extreme flow events (Fileni et al., 2026b). This quality control process is different to the quality control process applied to the daily flow data. Erroneous daily flow data are removed as part of the quality control process by measuring authorities and the UK NRFA, thus the daily flow data are the best estimate available and considered suitable for analysis and modelling without reference to flags. However, the quality control process for the hourly flow and level data acknowledges that some of these data may be incorrect and addresses this by applying flags to alert users of data issues, rather than removing data. It is therefore important that users use the flag-based system on the hourly flow and level data to identify, remove, or interpolate potentially problematic data as per their study requirements.

The quality flag for the corresponding hour in the flow and level data was selected according to an order of priority (i.e. which flag was deemed to be most important; see Sect. S1 and Fig. S6). No flow or level data have been removed or modified by the quality control process so users can decide which data they want to include as part of their analyses. The flags are grouped into three categories; (1) comparison with other data products (such as the daily NRFA data), (2) traditional QC checks (such as negative values, truncated low/high flows, spikes), and (3) high-flow QC checks (with comparison to antecedent rainfall and assessment of unrealistically high values) (see Tables S1–S3 for a full description of the quality control codes). The levels data underwent less extensive quality control than the flow dataset, and do not include quality control flags for some anomaly checks, comparisons with other UK hydrological products and hydrological similarity flags (see Sect. S1).

The most common flag for hourly river flow (Fig. S8) are where there is a mismatch of >5 % between the 15 min values recorded in UK-Flow15 (aggregated to daily) and the National River Flow Archive daily values – this flag is recorded in the timeseries of 539 stations and can appear for over 90 % of the timeseries for some gauging stations. The hourly flow data will not always be consistent with the daily NRFA flow data because of rating curve changes, version control inconsistencies, and station-specific issues.

The other flags typically affect a relatively small proportion of the timeseries (this is expected as many flags focus on extremes) but will be important for users to identify, remove, or interpolate potentially problematic data as per their study requirements.

Users seeking a simpler way to select a subset of stations that are suitable for a range of applications are recommended to use the “station quality” categorical information contained within the hydrometry attribute file (see Sect. 5.7).

The UK Benchmark Network contains 146 catchments where human impacts on flow regimes are assumed to be minimal Harrigan et al., 2018). All CAMELS-GB catchments are flagged according to whether they belong to the UK Benchmark Network, providing users with an indication of which catchments are relatively near-natural and therefore more suitable for studies requiring minimal human impact. Users should be aware that to ensure coverage in the south and east of GB (where there are lots of human influences on river flows), some human impacts were accepted.

Average daily abstraction and discharge rates are provided again in CAMELS-GB v2 but based on a new dataset of abstractions and discharges (Rameshwaran et al., 2025). This new dataset is based on the same underlying data as used in CAMELS-GB v1 but underwent additional quality control and is now available open source.

The abstraction data consist of the total water quantity (in most cases measured using a water meter) that has been abstracted for each license and each month from 1999 to 2014 in England on a 1 km grid. These monthly abstraction data were averaged to provide a mean monthly abstraction from 1999 to 2014 for each abstraction licence and then aggregated for each catchment to provide a mean daily abstraction rate for all English catchments in CAMELS-GB v2 for groundwater and surface water sources. The use of the abstracted water (agriculture, amenities, environmental, industrial, energy, or water supply) is also provided and how much of the abstracted water is consumed/lost (high – 100 %, medium – 60 %, low – 3 % and very low – 0.3 %). For example, cress pond throughflow is described as very low loss, whilst farming and water supply is classed as medium loss, and trickle irrigation is classed as high. These loss factors are only used for billing purposes and are therefore indicative of the true water consumption.

The discharges data consist of recent actual discharges for England from the WRGIS (Water Resources Geographic Information System). These data represent discharges from sewage treatment works and other “significant” discharges (typically those >20 m3d-1) using an estimate of recent actual summer discharge. For each catchment, we calculate a sum of all the discharges that fall within the catchment boundary and then convert into millimetres per day using catchment area to provide a mean daily discharges rate.

There are several limitations associated with these data. Firstly, these catchment attributes are only available for England and there are many catchments where either (1) no data are available (identified by “NaN”), (2) abstractions or discharges are recorded in zero when in reality they are not, or (3) only a proportion of the abstractions/discharges are accounted for, as the catchments lie on the border of England–Wales or England–Scotland. Secondly, the topographical catchment mask was used to define which abstraction returns were included in each catchment which will not be representative for groundwater abstractions that lie within the topographical catchment but do not have a direct impact on the catchment streamflow or those that lie outside the catchment but have an impact on that catchment's streamflow. Thirdly, this is not the full picture of human influences within each catchment. Not all licence types/holders are required to submit records to the Environment Agency; the abstraction data used here does not hold returns for abstractions less than 20 m3d-1, and from 2008, abstraction licence-holders less than 100 m3d-1 were no longer required to submit records of abstraction. Furthermore, there is large inter-annual and intra-annual variation in the abstraction and discharges data, and its impacts will be different across the flow regime. Finally, while abstractions represent water removed from surface water or groundwater sources, some of this water will be returned to catchment storages. This is partially represented by the loss factor and discharges information but the relationship between discharge consent data and water returned from abstractions will often be more complex than these simple attributes.

For CAMELS-GB v2, several reservoir attributes are derived for each catchment by determining the reservoirs that lie within the catchment mask from the reservoir locations and then calculating (1) the number of reservoirs in each catchment; (2) their combined capacity; (3) the fraction of that capacity that is used for hydroelectricity, navigation, drainage, water supply, flood storage, and environmental purposes; (4) the year when the first and last reservoirs in the catchment were built, and (5) the contributing area and normalised upstream capacity.

The first four sets of reservoir attributes are the same as CAMELS-GB v1 and calculated from the open-source UK reservoir inventory (Durant and Counsell, 2018) supplemented with information from SEPA's publicly available controlled reservoirs register. For CAMELS-GB v2, we excluded 43 reservoirs from the inventory as they could not be placed on the river network largely because their outflow or inflow location was unclear (see Fig. S1 in Salwey et al., 2023) and therefore the new reservoir attributes could not be calculated. This leads to only minor differences with CAMELS-GB v1 (Fig. S10).

Two new reservoir attributes are included in CAMELS-GB v2; contributing area and normalised upstream capacity. These attributes were chosen due to previous studies finding clear links between the size and location of upstream reservoirs and the associated flow alteration for UK catchments (Salwey et al., 2023).

Contributing area describes the percentage of the overall catchment surface area that is drained through reservoirs: 1Contributing area(%)=catchment area drained by reservoirs(km2)total catchment area(km2)×100.

In CAMELS-GB v2, the mean contributing area is 18 %, with a maximum of 100 %. The contributing area is complemented by the normalised upstream capacity, which compares the capacity of a reservoir to the average volume of precipitation received by the catchment in a year: 2Normalised upstream capacity=total upstream reservoir capacity(mm3)total catchment area(mm2)×average annual catchment precipitation(mm).

In CAMELS-GB v2, the average normalised upstream capacity is 0.08 (i.e. the reservoir is large enough to store 8 % of average annual rainfall) with a maximum of 2.5 (i.e. the reservoir is large enough to store 250 % of average annual rainfall). Twelve catchments have a normalised upstream capacity greater than 0.2 where the highest degree of alteration is typically found in the UK (Salwey et al., 2023).