Between 2013 and the time of writing, UKCEH has deployed 51 COSMOS-UK environmental monitoring sites across the UK (Fig. 1) (Boorman et al., 2020). Two sites, Wytham Woods and Redmere, have been decommissioned during this time due to changes to site conditions and access. A summary of each site's main characteristics is included in Table 1, and a record of any changes to site land cover is provided in Table 2.

The selection of sites within the network has aimed to provide an appropriate spatial coverage for improving understanding of UK soil moisture conditions, including representation of key land cover and soil types. All UK regions are represented, though there are more sites in the south and east of the UK to adequately capture the greater soil moisture variability in these areas. Installation of sites in less represented regions is in consideration but is dependent on the availability of resources.

Specific site locations have been further determined by practical considerations such as long-term permission and reasonable access for instrument installation and maintenance, and mobile phone network coverage. Where possible, site selection has aimed to exploit opportunities for COSMOS-UK data to support independent, existing research projects, e.g. data assimilation for forecasting and prediction, validation of remote sensing data, and support of other monitoring programmes and activities. Similarly, site selection has aimed to create partnerships with farmers and support agricultural research.

Some site characteristics can limit their suitability for CRNS soil moisture measurement, such as proximity to open water or shallow or perched groundwater (such features should not be present within the CRNS measurement footprint) and highly variable topography. Sites have therefore been installed in non-mountainous and largely flat locations with no regular irrigation or close proximity to significant water bodies.

Instrumentation at COSMOS-UK sites is largely standardized (Fig. 2); however differences have arisen for the following reasons.

Instrument performance was reviewed, resulting in subsequent installations utilizing different, higher-performance sensors (e.g. for improved sensor accuracy).

These site differences are detailed in Table 3. For further information regarding individual instruments, a detailed summary is provided in the COSMOS-UK User Guide (Boorman et al., 2020).

Available measurements are described below, and further information regarding variables and recording intervals is provided in Sect. 4. All COSMOS-UK measurements are logged on a CR3000 Micrologger (Campbell Scientific Ltd., Logan, Utah, USA) and telemetered via the 2G, 3G, or 4G mobile network, or Inmarsat BGAN satellite network (Inmarsat Global Ltd., London, UK), to secure servers at UKCEH Wallingford. Telemetry has been achieved using a COM110 (Campbell Scientific Ltd., Logan, Utah, USA), Maestro M100 (Lantronix Inc., Irvine, California, USA), Proroute^® H820 (E-Lins Group, Shenzhen, China), or 9502 BGAN (Hughes Network Systems LLC, Germantown, Maryland, USA) modem.

Sensor calibration coefficients are stored on the CR3000 for measurements such as soil heat flux (G, W m-2) and the four components of net radiation (RN, W m-2). Equipment across the network is promptly replaced when faults are detected, and instruments are tested and re-calibrated on an annual basis under a maintenance contract with the suppliers of the field instrumentation, Campbell Scientific Ltd. A full record of sensor exchanges is maintained by UKCEH.

An in situ soil sampling procedure adapted from Franz (2012) and Zreda et al. (2012) has been completed at each COSMOS-UK site following installation. The results from the sampling are used to determine site-specific soil properties for CRNS calibration: field average soil moisture and dry bulk density, lattice and bound water, and organic matter. Once the CRNS count data have been corrected for atmospheric pressure (Desilets, 2017; Evans et al., 2016), humidity (Evans et al., 2016; Rosolem et al., 2013), and an empirical background neutron intensity factor (adapted from Evans et al., 2016), the calibration data are used to derive N0 on the day of calibration (details in Sect. 3.1). Soil samples for determination of VWC and dry bulk density were taken at 18 representative locations centred on the CRNS: at compass bearings of 0, 60, 120, 180, 240, and 300∘ and at 5, 25, and 75 m radial distance at each of these compass bearings (Fig. 3). For CRNS calibrations before 14 September 2016, samples were taken at 25, 75, and 200 m radial distances. These locations follow Franz (2012), subsequently modified to account for revised CRNS footprint characteristics (Köhli et al., 2015b). In addition, as the 180∘ sample at 5 m distance would fall on a cable run within the CRNS enclosure, this location has been replaced with a sample at either 90 or 270∘ at 1 m distance. At each location volumetric soil samples (using 0.05 m diameter, 0.051 m length rings (Eijkelkamp 07.53.SC sample ring kit and Edelman auger)) were taken at five depths: 0–0.05, 0.05–0.1, 0.1–0.15, 0.15–0.2, and 0.2–0.25 m below ground level (b.g.l.). Soil sampling depths for CRNS calibration were selected to match typical (moist) UK conditions, and higher weighting is later applied to shallow soil layers to ensure appropriate representation of the decreasing contribution of deeper water (Köhli et al., 2015a; Schrön et al., 2017) (see Sect. 3.1 for details). Three locations at different bearings and distances were also selected for an additional soil sample for the determination of lattice and bound water and organic matter. The additional soil samples were taken from 0–0.25 m b.g.l. This therefore gives a total of 90(+3) soil samples for each calibration.

The field soil samples were returned to the laboratory for analysis. VWC and dry bulk density were determined for the 90 volumetric samples using oven drying (∼ 36 h at 105 ∘C). Following analysis, a ∼ 2 g sub-sample was taken from each sample and aggregated to form a composite sample for lattice and bound water and organic matter determination. The three additional soil samples from the field were air-dried (on the lab bench or in the oven at 30 ∘C) for around 3 d. The additional samples, along with the composite, were then crushed to pass a ∼ 0.4 mm sieve and subsequently air dried at 105 ∘C for ∼ 36 h. Soil organic matter was then estimated for a ∼ 3 g air dried sub-sample (with six replicates per additional sample, i.e. 24 sub-samples) using loss on ignition at 400 ∘C for 16 h in the furnace (following Nelson and Sommers, 1996). Following cooling in a desiccator and weighing, the sub-samples were then returned to the furnace to estimate lattice and bound water by loss on ignition at 1000 ∘C for 4 h (following Pansu and Gautheyrou, 2006). For use in the CRNS calibration calculation, soil organic carbon was estimated as 50 % of soil organic matter (Nelson and Sommers, 1996). Pansu and Gautheyrou (2006) note that loss on ignition removes organic matter at 300–500 ∘C and lattice and bound water at 350–1000 ∘C. The procedure outlined above therefore follows the 400 ∘C temperature recommendation by Nelson and Sommers (1996), which removes organic matter but causes minimal dehydroxylation of clay minerals. The CRNS calibration procedure uses the mean soil organic carbon and mean lattice and bound water from the 24 sub-samples along with the mean dry bulk density from the 90 volumetric samples. The field average reference VWC for the day of calibration is then calculated as a radial and vertical weighted mean following Köhli et al. (2015). Planned work includes obtaining site bulk density using this weighting function. The soil properties and soil moisture results for calibrating each site are available in Table 5.

Repeat calibrations using secondary samples have been conducted at two COSMOS-UK sites to explore the accuracy of the derived VWC obtained on a particular day using this methodology. There was < 0.03 cm3 cm-3 difference in VWC between the soil moisture determined from the second calibration and the corresponding daily VWC value derived using the site's first calibration. Considering the estimated errors in soil sampling and (to a lesser extent) laboratory procedures, the difference in calibrations is considered to be within the uncertainty of the reference soil moisture determined from secondary sampling and the predicted VWC from the CRNS and its original calibration. Additional repeat calibrations are planned across the network to help further analyse the current methodologies and assess sensor performance over time.

Field-scale soil moisture (CRNS VWC) is derived from the corrected CRNS epithermal neutron counts, which inversely correlate with hydrogen present at the land surface (soil, vegetation, and any other water sources; Zreda et al., 2008, 2012). Incoming epithermal neutrons collide with hydrogen nuclei at the land surface and are therefore moderated by the hydrogen present in water molecules, thereby enabling an indirect measurement of soil moisture (Rivera Villarreyes et al., 2011). VWC is calculated using the following formula where coefficients are determined for a basic silica soil (Desilets et al., 2010; Evans et al., 2016). 1VWC=0.0808NcorrN0-0.372-0.115-τ+SOCρbdρw

In Eq. (1), Ncorr denotes the corrected counts, τ is the reference lattice and bound water, SOC is the reference soil organic carbon, ρbd is the reference bulk density, and ρw is the water density equal to 1 g cm-3. τ, SOC, and ρbd are determined on the calibration day by field and laboratory analysis (Evans et al., 2016; Franz, 2012; Franz et al., 2013; Zreda et al., 2012). Ncorr is obtained by aggregating raw neutron counts from each site to a 60 min interval and correcting for atmospheric pressure (Desilets, 2017; Evans et al., 2016), humidity (Evans et al., 2016; Rosolem et al., 2013), and background neutron intensity variations (adapted from Evans et al., 2016) using in situ measurements. The atmospheric pressure correction uses instantaneous barometric attenuation lengths (Desilets and Zreda, 2003) calculated for COSMOS-UK sites according to http://crnslab.org/util/intensity.php (last access: 22 April 2021), and the correction uses a fixed reference pressure value of 1000 Pa. The background neutron intensity correction uses the publically available Jungfraujoch (JUNG) data (http://nmdb.eu/station/jung/, last access: 22 April 2021) provided by the Physikalisches Institut, University of Bern, Switzerland. Normalized count rates from JUNG are retrieved and used in sub-daily calculations to produce near-real-time COSMOS-UK datasets; the period of record is subsequently updated for any changes to JUNG data on an annual basis. Where data are unavailable from the JUNG detector the period is infilled with appropriately scaled values from alternate monitors: another counter at Jungfraujoch (JUNG1), Newark in the USA (NEWK) provided by the University of Delaware Department of Physics and Astronomy and the Bartol Research Institute, or Apatity in Russia (APTY). When choosing the most suitable neutron monitors for COSMOS-UK data, these monitors were identified as well-maintained with high levels of data completeness. The geomagnetic cut-off rigidity of the available monitors' locations was also considered when identifying suitable monitors. Normalized count rates are not greatly affected by cut-off rigidity except for during significant space weather events, when magnetic field disturbances may result in a change to a location's cut-off rigidity. A comparison between JUNG and monitors with cut-off rigidities similar to COSMOS-UK sites presented good agreement between the normalized counts and associated trends. Following this correction for background neutron intensity, counts are then calibrated to the site's specific soil, using the soil calibration values determined by UKCEH laboratory analysis.

COSMOS-UK uses the site-specific N0 method (Desilets et al., 2010) for deriving water content from a site's corrected neutron count data, where N0 is the site-specific neutron counting rate over dry soil under reference atmospheric pressure and solar activity conditions. Alternative methods are described in Baatz et al. (2014), Bogena et al. (2015), and Iwema et al. (2015). A site-specific N0 value is calculated by rearranging Eq. (1) for N0 and substituting the average neutron counts on the day of calibration for N, together with reference soil moisture for VWC. The corrected counts and N0 can then be input into Eq. (1) to produce VWCs. These data are subsequently constrained to the physical range of 0 %–100 % soil water content by determining values of Nmax and Nmin respectively, the maximum and minimum physically admissible neutron count value for each site. Figure 4 shows an example of the calibration curve for the Redhill site, located in South East England.

Once complete, this process produces the hourly CRNS VWC dataset. In a subsequent process, hourly corrected neutron counts are averaged to a daily mean and undergo the same calculations to produce the daily CRNS VWC soil moisture dataset. A minimum of 20 hourly values in a day is set as the requirement to produce a daily soil moisture value. An additional version of the soil moisture dataset is calculated, in which daily CRNS VWC has been adjusted for snow events using site measurements of albedo.

An in-lab cross-comparison was performed on the majority of CRNSs prior to field deployment. Cross-calibration of deployed CRNSs was also carried out at six COSMOS-UK sites; data were captured from two adjacent CRNSs for a period of several months to establish a reliable relationship between their counts using a linear regression model.

Point soil moisture and precipitation data at each COSMOS-UK site provide important ancillary information for assessing the potential accuracy of the CRNS VWC data. Figure 5 shows each of the processing stages for deriving soil water content from neutron counts for the Cochno site in Scotland, alongside soil moisture measured by the 10 buried point sensors and precipitation. This figure clearly shows that daily CRNS VWC data closely resemble the soil moisture dynamics measured by the point sensors, and the response of both VWC measurements to precipitation events.

Some sites may have a higher CRNS VWC measurement uncertainty. For example sites with extensive soil organic matter accumulation (e.g. carbon-dense peatlands) or mature woodlands where CRNS VWC methods might need to be further refined to account for biomass, plant roots, litter-layer thickness, and intercepted water (Andreasen et al., 2017; Baatz et al., 2015; Heidbüchel et al., 2016; Rivera Villarreyes et al., 2011). The contrast of CRNS VWC measurements between sites can be seen in Fig. 6, which displays all data for the period of record as a normalized curve for each site. This figure demonstrates the importance of identifying and understanding localized soil properties and shows how sites in close proximity and experiencing broadly similar weather patterns can exhibit vastly different ranges and extremes in VWC.

The CRNS VWC value is an average soil moisture measurement (%) across an estimated, variable footprint of radius up to 200 m and estimated variable measurement depth of between approximately 0.1 and 0.8 m (following Köhli et al., 2015). Measurement area depends on local soil moisture, humidity, and land cover (Köhli et al., 2015b), whilst penetration depth depends on soil moisture as well as lattice water and soil organic matter water equivalent (Zreda et al., 2008, 2012; Franz et al., 2012). The greater the actual soil water content, the smaller the CRNS measurement area and shallower the penetrative depth. The measurement area of the CRNS was initially believed to have a radius of approximately 300 m (Zreda et al., 2008); however Köhli et al. (2015) report that 50 % of measured neutrons originated within 50 m of the CRNS, and the footprint radius extended to only 240 m in arid climates. The penetration depth of the measurement is greatest near the CRNS and decreases with distance from the sensor; this varying depth across the footprint is provided as “D86”, the depth at which 86 % of the measured neutron counts are estimated to have originated at a given distance (Zreda et al., 2008; Franz et al., 2013). In the COSMOS-UK dataset, D86 is provided at distances of 1, 5, 25, 75, 150, and 200 m from the CRNS. Figure 7 shows the estimated D86 values for a typically drier site, Euston (average soil moisture approximately 15 %), and a typically wetter site, Riseholme (average soil moisture approximately 33 %), for 2018. During this year the UK experienced a cold wave with significant snow in February to March and a heatwave in June to August. This figure presents how measurement depth increases in drier conditions, decreases with distance from the CRNS, and differs between sites.

Raw data collected at each COSMOS-UK site, comprising the measured variables described above as well as additional diagnostic data from sensors (e.g. internal humidity of the CRNS), are telemetered to UKCEH and stored in an Oracle relational database (Oracle, 2012). When new values are derived following the application of corrections, calibrations, and quality tests, these derived data are stored in separate, secondary tables. These secondary datasets are those that are published.

Data quality assurance (QA) and quality control (QC) are applied to specific variables in the raw data. QC is conducted in two stages.

Automated processing is applied to raw data to provide a quality-assured dataset. Data which fail the tests are flagged and are not written to secondary datasets. These automated tests include pre-processing for known errors and subsequent QC processes for detecting additional erroneous data. These processes are explained below.

Automated processing tasks assess the raw data and create a flagged dataset based on the test results. This enables tracking of data removal and ensures raw data are not lost or overwritten. Raw data are passed through multiple independent QC tests (Table 6). Each test assigns a unique flag value to any raw data which fail. Where data fail multiple tests, the flag values are summed. The summed flag values are unique for each combination of tests, allowing failed tests to be determined from the sum. Where data pass all QC tests, the flag values are assigned “0”. The tests flag issues including data exceeding known thresholds, implausible values, and data where other variables indicate an issue. The secondary dataset comprises all data not flagged by the QC processes.

All derived datasets are obtained using the quality-checked 30 min data. Planned future work includes the development of a tertiary dataset comprising quality-processed and gap-filled data.

In addition to the COSMOS-UK observed soil and hydrometeorological data, the network provides derived datasets including potential evaporation (PE), albedo, snow days, and snow water equivalent (SWE).

PE has been derived from each site's solar radiation, soil heat flux, air temperature, humidity, and wind speed data using the Penman–Monteith method as described by the Food and Agriculture Organization of the United Nations (FAO) (Allen et al., 1998) (Fig. 8). Daily PE data for all COSMOS-UK sites are provided in this dataset.

Snow days have been identified using albedo measurements, and SWE has been determined using the albedo and neutron count data available from the CRNS at each COSMOS-UK site. Neutron counts from both the CRNS and SnowFox sensor are sensitive to all sources of water in the environment, allowing them to be used to estimate the SWE held in a snowpack. First the albedo is used to determine the presence or absence of snow cover, and then, if present, the reduction in neutron count rate from an estimated snow-free value is used to approximate the SWE, following the method of Desilets (2017). This dataset includes CRNS SWE. Methods for estimating SWE are available from Wallbank et al. (2020) and discussed in more detail in Wallbank et al. (2021).

Available derived data are listed in Sect. 4.

Additional information can be derived from the data provided by COSMOS-UK sites. As part of ongoing and planned evolution of the network, the additional data described in this section are not yet included in the published data.

Existing PE data will be complemented by a new derived dataset, which estimates actual evapotranspiration (ET) as the residual term from measurements of net radiation, soil heat flux, and the sensible heat flux derived from sonic anemometer measurements. Modelled energy fluxes, such as latent and sensible heat, have been calculated by utilizing the 20 Hz wind measurements recorded at the majority of COSMOS-UK sites (Crowhurst et al., 2019). This provides a network-wide modelled actual ET dataset for the UK.

In addition to the measurements mentioned previously, COSMOS-UK sites also capture photographs. Sites include a camera for monitoring phenology, a “PhenoCam”, with two hemispheric lenses facing north and south (Fig. 9). Each COSMOS-UK site sends five photographs per day, which capture the full extent of the COSMOS-UK site and surrounding area, thereby providing additional information on local phenology and cloud cover. These PhenoCam images can be used to confirm when site conditions have changed, for example when the land cover has been modified (e.g. ploughing, mowing, grazing, harvesting) or there has been heavy snowfall. PhenoCam photos from COSMOS-UK sites are also currently being analysed to produce a greenness dataset. Using site-specific image masks, RGB (red, green, blue) data can be extracted from each image to determine the greenness of the land cover at each site (Wingate et al., 2015). In 2020 the network's first gauge board was installed at the Cwm Garw site in Wales. Gauge boards indicate height above ground level (cm) against which vegetation height and snow depth can be estimated via PhenoCam images. Further gauge boards are planned at sites across the network.