Three years of soil moisture observations by a dense cosmic-ray neutron sensing cluster at an agricultural research site in north-east Germany

. Cosmic-ray neutron sensing (CRNS) allows for the estimation of root-zone soil water content (SWC) at the scale of several hectares. In this paper, we present the data recorded by a dense CRNS network operated from 2019 to 2022 at an agricultural research site in Marquardt, Germany - the first multi-year CRNS cluster. Consisting, at its core, of eight permanently installed CRNS sensors, the cluster was supplemented by a wealth of complementary measurements: data from seven additional temporary CRNS sensors, partly collocated with the permanent ones, 27 SWC-profiles (mostly permanent), two groundwater 5 observation wells, meteorological records, and global navigation satellite system reflectometry (GNSS-R). Complementary to these continuous measurements, numerous campaign-based activities provided data by mobile CRNS-roving, hyperspectral imagery via unmanned aerial systems (UAS), intensive manual sampling of soil properties (SWC, bulk density, organic matter, texture, soil hydraulic properties), and observations of biomass and snow (cover, depth, and density). The unique temporal coverage of three years entails a broad spectrum of hydro-meteorological conditions, including exceptional drought periods, 10 extreme rainfall, but also episodes of snow coverage, as well as a dedicated irrigation experiment. Apart from serving to ad-vance CRNS-related retrieval methods, this data set is expected to be useful for various disciplines, e.g. soil and groundwater hydrology, agriculture, or remote sensing. Hence, we show exemplary features of the data set in order to highlight the poten-1

1 Introduction 1.1 Towards closing the scale gap in soil moisture observation A large body of literature highlights the significance of soil moisture as a key state variable of the earth system. Yet, soil moisture appears to elude our attempts to obtain representative observations: point-based measurements lack both representativeness and coverage (Blöschl and Grayson, 2000) while remote sensing struggles with issues such as small penetration depth 20 and low overpass frequencies (Peng et al., 2021). Since Zreda et al. (2008), cosmic-ray neutron sensing (CRNS) has emerged as a promising option to address these issues, and hence to close the scale-gap between point measurements and large-scale soil moisture retrievals. The advantage of the CRNS sensor is its considerable horizontal (100-200 m) and vertical (20-50 cm) footprint (Schrön et al., 2017). It is hence considered to efficiently average across small-scale heterogeneity, allowing to obtain continuous estimates of "root-zone" soil moisture at the scale of several hectares (see e.g. Zreda et al., 2008;Desilets et al., 25 2010; Köhli et al., 2020, for further details). To that end, soil moisture is estimated from the epithermal neutron intensity by means of a conversion function. This typically involves the calibration of one parameter, N 0 , on the basis of a sufficient number of soil moisture measurements in the sensor footprint (Schrön et al., 2017;Köhli et al., 2020).
Furthermore, different application scenarios of the CRNS technology have been developed in order to obtain spatial soil moisture estimates beyond the isolated footprint of a single, stationary CRNS sensor. The application of CRNS roving, for 30 instance, involves a mobile neutron detector that is moved within a study area (Desilets et al., 2010;Schrön et al., 2018). That way, the spatial distribution of soil moisture along the roving transect can be inferred in terms of a snapshot in time, given that potential sources of bias (e.g. from road material or biomass effects) are sufficiently addressed (Fersch et al., 2018;.
As an alternative, dense clusters of stationary CRNS were proposed as an option to retrieve the spatial and temporal distribu-

Three years of dense CRNS observations: the Marquardt Cluster
In August 2019, the research unit Cosmic Sense, funded by the German Research Foundation (DFG), launched a dense CRNS cluster at an agricultural research site in Marquardt (about 10 km northwest of Potsdam, Germany) -the first of its kind designed sensor, and, on the other hand, to obtain information about the spatial variability of soil moisture at a length scale that is smaller than the diameter of a single CRNS footprint.
55 Figure 1 provides a graphical overview of the MqC. There are various features which are specific to this endeavour, and which make it unique in comparison to both previous CRNS clusters, hence allow for new research opportunities: -With a density of 8 CRNS sensors per 10 ha, MqC features a significant leap in density compared to previous clusters (3 and 2.4 sensors per 10 ha in Heistermann et al. (2022a) and Fersch et al. (2020), respectively), and hence a considerable overlap of footprints; this aims for an improved identification of heterogeneity at the sub-footprint scale.

60
-The two aforementioned clusters span a few months each; the MqC is the first dense CRNS cluster operated over several years, spanning all seasons and allowing to observe more diverse conditions and processes, including periods of drought and snow cover as well as a set of heavy rainfall events at different durations.
-Focus on an agro-ecosystem with very diverse management: the site comprises traditional field crops, meadows, biomass crops and orchards, irrigated as well as rainfed management, and adjacent forests.

65
-A large number of vertical soil moisture profile probes not only allows to study the effect of the vertical soil moisture distribution on the CRNS signal, but also, together with groundwater observations, to investigate the processes of infiltration and groundwater recharge.
-The data set includes a dedicated irrigation experiment: a selected plot was intensively irrigated while being monitored with a cross-scale combination of sensors, including hyperspectral UAS-based remote sensing and CRNS roving.

70
-On-site muon-monitoring allows to study novel methods for the local correction of the incoming neutron flux.
-MqC is a typical lowland site in the transition from maritime to continental climate, thus representing a landscape typical for large parts of northern Central Europe.

Structure of this paper
This paper presents the MqC data acquired between August 2019 and November 2022. The study area is introduced in Sect. 2; 75 the acquisition and, partly, the processing, of different subsets of the data is documented in Sect. 3. Other relevant data from third parties are addressed in Sect. 4. In Sect. 5, we highlight selected features of the obtained data with regard to various event types (snow, drought, heavy rainfall, irrigation) while the conclusions in Sect. 6 outline research perspectives with regard to the published data set.
2 Study site 80 The study area in Marquardt is a lowland agricultural site in the north-east of Germany. Various practical features made it a suitable candidate location for MqC, e.g. the vicinity to the participating institutions, the integration in an existing research context, and the security provided by a complete fencing of the perimeter. The diversity of agricultural land use and management types, together with its high-level documentation, was a desirable feature with the regard to the investigation of a heterogeneous landscape.  DWD station, the average annual precipitation is 584 mm , the average annual temperature is 9.3°C (Cfb according to Köppen's climate classification). The Marquardt site is located at about 40 m a.s.l. It sits on a gentle hillslope sloping westwards, with distances to the unconfined groundwater table ranging from around 1.5 to 10 m. The soils are dominated 90 by periglacial sand deposits over glacial till. The sand content ranges between 68 and 91 %, silt content is 8-27 %, and clay 0.6-4.4 %. The soil organic matter content ranges between 0.4 and 17.3 %.
The Marquardt site comprises approx. 20 field plots, which host annual and permanent crops. In the observation period of 2019-2022, the core area covered by the MqC was dominated by orchards (cherry, apple), field crops (cereals, alfalfa, sugar beets, maize), meadows and a biomass plantation stocked with young poplars. Details are contained in the data repository. Germany) and Finapp S.r.l. (San Pietro in Cariano, Italy). Table 2 provides an overview of the detectors and their sensitivities relative to a reference device (see Heistermann et al., 2022a, for details). Dividing by the sensitivity, neutron count rates  (Gianessi et al., 2022). In addition to epithermal neutrons, two devices also recorded thermal neutron count rates which might have the potential to support the separation of signals from soil moisture, vegetation, or snow (Tian et al., 2016;Jakobi et al., 2018). Further recorded variables include relative humidity, air temperature, and barometric pressure.
The locations of the CRNS sensors are shown in Fig. 2. Sensor placement was guided by various criteria: (i) to create 115 significant overlap in the core area of the cluster, (ii) to cover the site along the hillslope gradient, (iii) to place some sensors close to the irrigated orchards, (iv) to minimize interference with agricultural management operations (sensors could not be placed on cropped fields).
The prime observation variable of the CRNS sensors are count rates of detected neutrons. The sensitive energy range for the individual detectors is different though . The stochastic uncertainty of the observed neutron count rate N 120 at an arbitrary integration interval ∆t amounts to √ N / √ ∆t . The uncertainty of soil moisture estimates based on CRNS, however, is subject to a wide range of effects, including the correction for effects of the atmosphere and other hydrogen pools, the collection and weighting of calibration measurements, and the propagation of errors through the non-linear conversion from neutron intensities to soil moisture (see e.g. Jakobi et al., 2020;Weimar et al., 2020;Baroni et al., 2018;Iwema et al., 2021, for a discussion).
125 Bold-IDs mark the permanent core cluster, non-bold IDs were only operated during shorter time periods (see Fig. 3).

Muons as a reference for incoming neutron intensity
Recent studies have indicated that local counts of muon and gamma particles might have the potential to support methods to correct CRNS data for the variability of incoming neutrons (Stevanato et al., 2022;Gianessi et al., 2022). Conventionally, neutron monitor data are used for that purpose (see Sect. 4.2). In order to allow for further benchmarking studies with regard to correction approaches, a muon and gamma detector (as part of the scintillator-based sensor FINAPP3, see CRNS sensor 11a in 130 Tab. 2) was operated from May 2021 until May 2022. For detector-related technical details, we refer to Gianessi et al. (2022).

Roving CRNS
Roving CRNS snapshots were acquired as part of the irrigation experiment between July and August 2020 (see Sect. 3.15).
The UFZ Hydroinnova rover is a moderated CRNS unit (Hydroinnova LLC, Albuquerque, USA) based on 3 He gas and has been used by car, by train, by aircraft, by handwagon, or carried by hand in many previous studies Fersch 135 et al., 2020;Heistermann et al., 2022a). Here, we placed the detector in a car to survey the whole MqC area, and then used it on a handwagon to map the irrigated plot and nearby fields inaccessible by car. The uncertainty of the neutron count measurements from roving follows the same counting statistics as outlined for stationary sensors (Sect. 3.2).
The same applies to the uncertainty of corresponding soil moisture estimates. For roving, specific uncertainties arise from the spatial heterogeneity of hydrogen pools or soil properties, the effect of roads , and the trade-off between 140 integration time and spatial resolution . All these uncertainties depend on the chosen processing methods and should be discussed by users of this data set.
Measurements were conducted on four days, one before the irrigation events (July 13 th ), and three right after the irrigation events (Jul 23 th , Aug 6 th , 11 th , see Tab. 3). The handwagon has been used in stop-and-go mode with typically 5-10 minutes residence time per point. Raw data in this repository have been cleaned and contain detector-relevant variables, GPS records, as 145 well as meteorological observations from an external mobile weather sensor mounted on the handwagon. In order to visualize the observations (see Sect. 5.2), the data recorded at intervals of 10 seconds were smoothed temporally with a moving average window of 1 minute, and spatially within a 5 meter radius using the distance-weighting function W r (Schrön et al., 2017).
Further corrections and the conversion to soil moisture followed the procedures outlined in Schrön et al. (2018). Since the sensor on a handwagon is not shielded by car material, we used a slightly larger calibration factor N 0 = 13447 cph compared 150 to other studies. The GNSS antenna receives the direct signal as well as the signal reflected by the earth's surface. For the purpose of soil moisture retrieval, Larson et al. (2008b) suggested to use the signal-to-noise ratio (SNR) recorded by the antenna, as it is independent of the effects of orbits, atmosphere, or clocks. Eq. 1 in Larson et al. (2008b) provides the key relationship between the SNR and a phase offset ϕ. This offset directly relates to the apparent reflection depth of the GPS signal which, in turn, 160 depends on permittivity and hence soil moisture. Accordingly, relative soil moisture changes can be retrieved by comparing different phase offsets ϕ, assuming that other surface properties remain constant. Please note that some more filtering is required to take into account the dominance of the direct signal in high elevation angles (see Larson et al., 2008a, for details).

GNSS
Actual SWC can then be estimated by relating these relative changes to representative SWC measurements on the ground.
In the MqC context, the obvious choice for such measurements would be the four collocated TDR profiles at a depth of 9 cm 165 (see Sect. 3.8).
Among all constellations, the Global Positioning System (GPS) is the most suitable for GNSS-R soil moisture applications.
The repetition period of the orbits is one sidereal day (23 h 56 min 4.0905 s), and each satellite yields two individual tracks over each location. In combination with the track splitting into two arcs (ascending and descending), all GPS satellites provide more than 100 reflection paths per day, each of them potentially providing a soil moisture estimate. The footprint of the reflection 170 corresponds to a projected ellipse on the ground and is not constant in time. Its shape and size depend on the antenna height, the wavelength of the carrier frequency, and the elevation angle of the satellite. Its position is related to the unique orbit of the satellite and the antenna height . For a 3 metre high antenna, the L2 GPS signals yield an ellipse of 33.66 m by 2.93 m size, when the satellite is located 5°above the horizon. The present setup hence accumulates observations from a circular area of approximately 33 m radius around the station, recorded with a Delta TRE-G3T receiver (JAVAD GNSS, Inc.,

UAS-based hyperspectral remote sensing
In order to be able to observe high resolution SWC patterns and to analyze plant-soil interactions, hyperspectral imagery was acquired during the irrigation experiment (section 3.15). Table 3

185
We performed the following steps to process the hyperspectral data: -Conversion of the raw data to radiance and reflectance using the Headwall SpectralView software and a reflectance tarp, which was scanned during the image acquisition.
-Geo-referencing of the reflectance data using the Georeferencer tool of ArcGIS (Esri Inc., USA) and high resolution multi-spectral UAS data as a basemap. Details of the basemap creation can be found in Döpper et al. (2022).

190
-Correction of the spectral signal for spikes and drops.
-Spectral filtering of the data with the Savitzky-Golay filter (Savitzky and Golay, 1964) as implemented in the Python SciPy module (Virtanen et al., 2020). We applied a window width of seven and a second order polynomial smoothing.
Besides the processed hyperspectral imagery, we provide the spatial SWC products based on a data-driven approach and a hybrid approach as described in Döpper et al. (2022). For interpretation and accuracy of the products, we refer to Döpper et al.

Leaf area index measurements
Over vegetated areas, the hyperspectral signal is dominated by variations of the leaf area index (LAI). We provide LAI measurements as a resource for interpreting the hyperspectral data and disentangling soil moisture-related signals from LAI signals.
The LAI was sampled on the four dates of the UAS-based hyperspectral image acquisition (see Tab. 3) and comprise 31 to 40 200 measurements per campaign. We measured the LAI using a LAI-2200C Plant Canopy Analyzer (LI-COR Biosciences GmbH, Germany). In order to avoid direct sunlight scattering, we sampled the LAI at dawn, starting shortly after sunset. We randomly located the measurements to reduce damages of the vegetation due to repetitive sampling. Each measurement resulted from the mean of 5 above and below canopy measurements at each location. The location was recorded using a Leica Zeno GG04 (Leica Geosystems AG, Switzerland) DGPS antenna with accuracies at centimeter level.

Soil moisture profiles
A variety of soil moisture profile measurements was implemented on the premise, covering different measurement depths and technologies (with a total number of 27 individual profiles, 23 of them with more than two years of data). That way, we could obtain detailed records on the vertical SWC dynamics related to infiltration and drying, which are critical to the retrieval and interpretation of CRNS-derived soil moisture (see e.g. Scheiffele et al., 2020). Furthermore, 5 profiles of 20 single TDR soil moisture probes (TDR100, Campbell Scientific Ltd., UK) in close vicinity to the permanent GNSS-R antenna are available (Fig. 2). The probes are installed in 9, 11, 25, 45 and 75 cm depth, and operate using the conversion suggested by the manufacturer, i.e., the approach after Topp et al. (1980), to derive volumetric soil water content in m 3 /m 3 in 15-minute intervals.

225
Finally, three profiles with measurements in 5, 10, 20, 40 and 60 cm depth were equipped with 5TE sensors (5TE, Decagon Devices, Inc., Pullman, USA) and single ML2x ThetaProbes in the same depth. The single ThetaProbes were calibrated with the above approach and removed in 2020. During installation of the 5TE sensors, soil samples were taken to determine volumetric soil moisture (section 3.11) and derive a linear relationship between sensor permittivity and soil moisture. The sensors also provide soil temperature measurements; electrical conductivity values were regarded as unreliable and are not provided here. surveyed by dGPS (see Fig. 2 for an overview). In October 2020, we retrieved another 29 soil cores specifically for the analysis 255 of bulk density, organic matter, lattice water content and texture. Additionally, during the installation of the tensiometers (see Sect. 3.10) and the 5TE sensors (Sect. 3.8), soil cores were collected from the installation depths down to to 2 m, allowing the retrieval of the mentioned parameters for greater depths.
Soil cores were extracted and treated as described in Heistermann et al. (2022a). Water content and bulk density were derived by oven drying, organic matter and lattice water determined on subsamples by loss-on-ignition. The texture analysis was done 260 on untreated soil samples by wet-sieving (for gravel, sand) and laser diffraction (silt, clay). The manual measurements with ThetaProbes, too, followed the procedure outlined by Heistermann et al. (2022a), using portable ThetaProbes with sensorspecific calibrations and a site-specific conversion of permittivity to soil moisture, which resulted into an RMSE of 0.03 m 3 /m 3 for the soil moisture estimates.
In addition to the two campaigns in November 7, 2019 and October 10, 2022, extensive near-surface measurements (ThetaProbes) 265 were conducted on 13 days between July 18, 2019, and August 17, 2020. These comprised fivefold replicates at 35 to 71 random locations with electrodes inserted from the surface (i.e., effective measurement depth 2.5 cm), processed as described above.

Evaporation experiment to determine soil hydraulic parameters
Two undisturbed soil cores of 250 cm 3 were taken during the installation of the three tensiometer and deep soil moisture profiles 270 at each installation depth (see section 3.10). The resulting 24 soil cores underwent soil hydraulic analyses in the laboratory.
First, saturated hydraulic conductivity was measured by the constant head method (Klute and Dirksen, 1986). Second, the cores were used to determine the unsaturated soil hydraulic properties within an evaporation experiment (Wind, 1968;Schindler and Müller, 2006). This was done using a kupF MP10 device (UGT, Umwelt-Geräte-Technik GmbH, Müncheberg, Germany). At an interval of ten minutes the device records the weight of the soil core as well as the matric potential at two tensiometers 275 vertically inserted at depths of 1.25 and 3.75 cm. Afterwards, dry weight and bulk density of the samples were determined.
Additionally, the water content at a pF-value of 4.2 (or suction of 15,000 hPa) was determined using a ceramic plate and pressure chamber on two small subsamples per soil core (Brooks and Corey, 1964;Klute, 1986). The data can be used to determine the retention curve and unsaturated hydraulic properties of the soil (e.g. Peters and Durner, 2008).

Measurement of snow depth, density and cover 280
In MqC, an appreciable snow cover can be expected every two to three years. In February 2021, a snow layer with a maximum depth of approx. 10 cm persisted for about one week. During this time, its main properties were measured: snow depth (continuously and campaign based), precipitation, snow cover and density (campaign-based).
Snow depth was monitored with an ultrasonic temperature-compensated distance-meter (SR50-45, CampbellScientific, Inc., Logan, USA) looking downward on a representative spot of shortly-clipped grass. Simultaneously, close to three CRNS sensors

285
(1, 2 and 11, see Fig. 2) wildlife cameras (SECACAM HomeVista, VenTrade GmbH, Köln, Germany) with infrared night vision took hourly images of sets of ten 50-cm snow stakes, distributed in clusters of approx. 10 m diameter each. The length of the stakes protruding from the snow was determined manually on the resulting images and used for a simple photogrammetric calculation of time series of the snow depths at the stakes.
Shortly before and during the snow period, total precipitation and air temperature was additionally measured using a Manual snow sampling included snow depth measurements with a ruler and density measurements. These were conducted using cylinder cores or collecting (sweeping up, rolling) all snow from a designated area and successive weighing in the field.

Land use and biomass
Hydrogen as contained in biomass affects both epithermal and thermal neutron count rates. The corresponding biomass pools 300 in the sensor footprints should hence be accounted for when interpreting the CRNS signal.
For the agricultural plots, the crops, date of operation and their yields have been determined by weighing the harvest. From these yields, total biomass inventories were computed using literature values for water content and harvest index (Munns et al., 2018;Stöckle et al., 2022;Taes et al., 2022;Kuai et al., 2015).
The highest biomass densities in and around the study area are within the surrounding forests in the west and north (Fig. 2), 305 followed by the orchards (cherry and apple plantations) and a short rotation plantation of poplar trees. For these areas of higher biomass density, the determination of the average above-ground dry biomass (AGB) was further refined by using allometric functions. In the forest stands and the poplar plantation, we used in situ measurements of diameter at breast height (DBH, for Populus was taken from Zell (2008). On average, the biomass in the poplar stand was estimated to be 1.3±0.6 kg/m 2 .
-For the cherry and apple orchards, we refer to Richter (2021) who measured the average AGB per tree from a representative sample of single trees. Combined with the total number of trees per plot and the plot size, the average AGB density yields to 3.8±0.4 kg/m 2 for the cherries and 0.49±0.08 kg/m 2 for the apples.
330 Table 3. Data acquisition during irrigation experiment, July-August 2020 July August 13 14 15 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 irrigation -Other plots: The walnut plot and a row of cornel cherries are the only appreciable woody vegetation for which no biomass or allometric data were collected as part of this study. Nevertheless, for the purpose of correcting the CRNS signal, we used the biomass estimate of the cherry plot as an acceptable approximation.

Irrigation Experiment (July-August 2020)
An irrigation experiment was carried out in order to observe the response of various sensors to pronounced soil moisture 335 contrasts. Between July 22 and August 11, a plot of 35 m × 100 m (Fig. 2) was repeatedly irrigated with a centre pivot sprinkler (Irriland, Italy): During three phases, 40-50 mm were applied to 3/3, 2/3 and 1/3 of the plot, respectively, i.e. the southernmost part receiving three pulses of water (see Tab. 3). Three rain gauges and one pan (41 cm × 72 cm) served for verifying the respective amount of applied water. Accompanying measurements included the following features (see also Tab. 3): -Three CRNS sensors were operated at close proximity to the irrigated plot, including sensors 1 and 22. The CRNS 340 sensor from location 26 was temporarily relocated to the eastern edge of the plot (location 30) since the CRNS sensors at locations 11 and 13 were not operated at the time.
-Nine campaigns of manually measured soil moisture at the surface (at 46-71 locations, see "near surface measurements" in section 3.11), and in 20 PR2-access tubes. The access tubes were visited nine times with a mobile PR2-profile-probe, taking three readings at 0, 120 and 240 degree orientation. Three of the 20 access tubes were also continuously monitored 345 throughout the experiment.
-Four mobile CRNS roving campaigns (see section 3.4 for details).
-Four UAS flights to acquire hyperspectral imagery (see section 3.6) for potential mapping of SWC patterns within the irrigated field.
-Four leaf area index measurement campaigns (see section 3.7) to complement the remote sensing observations since LAI 350 is one of the most dominant variables in the spectral signal.

Relevant data provided by third parties
The following subsections highlight relevant data sets which have been published already or are provided by other organisations or channels, but which we consider as potentially helpful for users of the data presented in section 3.

Weather data 355
A climate station with an heated tipping bucket rain gauge is located in the north-eastern part of the study area (see Fig. 2) and recorded, at an hourly resolution, standard climate variables, including air temperature, relative humidity, precipitation, soil temperature (at 5, 10 and 30 cm), solar irradiation, as well as wind speed and direction. The original data is openly available at http://technologygarden.atb-potsdam.de/bsa_wetter.aspx.
The closest climate station of the German Weather Service (Deutscher Wetterdienst, DWD) is located in Potsdam about 360 12 km south-east of the study site (station ID 03987). The data is available via DWD's open data repository https://opendata. dwd.de/climate_environment/CDC/observations_germany/climate/.
For convenience, we included subsets of both station records in the published dataset, spanning the MqC study period.

Incoming neutron flux
Variations of the incoming cosmic-ray neutron flux on Earth are recorded by neutron monitors. The corresponding data are  A comprehensive presentation and discussion of this large and diverse data set is beyond the scope of this paper. Still, this section provides an overview, and also highlights selected details. This involves some level of data processing, specifically 380 with regard to the correction of neutron counts from stationary sensors and CRNS roving, as well as the calibration of the relationship between neutron counts rates and volumetric soil moisture. Details with regard to the corresponding processing steps are available e.g. in Heistermann et al. (2021Heistermann et al. ( , 2022a. The study period coincided with three major drought years (2019, 2020 and 2022). This becomes apparent from the cumulative difference of precipitation and reference evapotranspiration (Fig. 3a) as well as the soil moisture obtained from the CRNS (Fig. 3b) and the soil moisture profiles (Fig. 3c). While the 2019 drought period was only captured in late summer with the launch of MqC, the droughts in 2020 and 2022 were covered in their full extent, starting already in May and ending in 390 September.

An overview of the entire study period
The dynamics of the groundwater level data (Fig. 3d) are consistent with both the cumulative sum of P-ET 0 and the soil moisture estimates. Specifically, the groundwater level at the hill foot (ID 2) shows distinct seasonal dynamics, with groundwater recharge typically starting in September. In August 2022, an exceptional heavy rainfall event caused an immediate groundwater level response, which is discussed in section 5.3 in further detail.

395
In February 2021, the CRNS-based soil moisture estimates (Fig. 3b) exhibited exceptionally high values. This was caused by a period of substantial snow cover, the only such episode in the study period. The effects of snow have not yet been corrected for in the CRNS-based soil moisture estimation. However, independent snow measurements (see section 3.13) in space and time are available to explore options for estimating both soil moisture and snow water equivalent from CRNS data, e.g. by exploiting thermal neutron counts. In contrast, even small rainfall events caused remarkable rise in the CRNS-based soil moisture, demonstrating its pronounced sensitivity to changes in shallow soil depths. The neutron response to the irrigation, in turn, remains relatively small: even for the first irrigation event, the irrigated plot only constitutes a small fraction of the entire CRNS footprint. This becomes even more evident for the second and third irrigation events. Prospective research will show whether it is, despite this low signal to noise ratio, possible to resolve the subfootprint heterogeneity introduced by irrigation (see also Brogi et al., 2022).

Heavy rainfall
The observation period comprised six heavy rainfall events with daily totals over 30 mm (Fig. 5a-f), including an extreme event with almost 100 mm per day and a maximum hourly depth of 60 mm in August 2022 (Fig. 5f). Not surprisingly, all events clearly show in the signal of the CRNS sensors ( Fig. 5g-l). The change in CRNS-based soil moisture not only depends on the total precipitation depth of the event, but also on the effective event duration: e.g. the short event on August 4, 2021, causes 420 a smaller soil moisture change than the event on June 30, 2021, although the total depth of 53 mm is the same. Interestingly, the soil moisture differences between some CRNS locations can be subject to considerable intra-event dynamics. While sensor 2 tends to be the wettest location before the event, it often ranks closer to the median of the ensemble after the rainfall while sensor 22 shows the opposite behaviour. It remains to be shown whether these changes are in fact caused by location-specific hydrological processes or whether they could also result from specific parameter constellations along the conversion from 425 neutron count rates to volumetric soil moisture.  (Fig. 5x). Rainfall events of this magnitude could produce surface runoff at MqC, however, only sporadic visual evidence is available, e.g. for the August 26 event. dropped below zero and precipitation started after February 7. While the weighing-based pluviometer (rain gauge 1 in Fig 6a) registered 5 mm, the permanent precipitation gauge (rain gauge 2) recorded significantly less, probably because of insufficient or malfunctioning heating and resulting of wind drift from the device (Fig. 6a). The resulting snow cover at the snow gauge peaked at around 10 cm (Fig. 6d), while the maxima at the snow stakes varied between 8 and 27 cm. The snow cover caused a marked decrease in the CRNS count rates (Fig. 6b). In contrast, the prior rainfall event (February 4) which yielded almost 440 three times the precipitation height, resulted in a much lower reduction in the neutron counts. During the approximately 10 days of snow cover, snow depth, soil moisture (Fig. 6c)  GW level (m) Figure 5. The six events with the highest daily rainfall depths during the study period. a-f) cumulative hourly rainfall depth; g-l) soil moisture from CRNS-probes; m-r) soil moisture from profile probes, averaged over all locations; s-x) groundwater (GW) levels.

Snow
margins of the MqC and some leeward structures), resulting in another increase in soil moisture by melt water, very similar to the response of the prior rainfall event. The spatial heterogeneity of snow depth in the thawing phase is represented not 445 only by manual measurements, but also by two UAS-borne acquisitions of optical imagery on February 17 and 18 ( Fig. 6e and   f). Altogether, the comprehensive monitoring of a complete buildup and thawing cycle should provide an excellent research opportunity to investigate the interplay of vertical soil moisture distribution and snow cover on the CRNS signal.

Conclusions
From August 2019 to November 2022, eight CRNS sensors, 23 permanent and four temporary SWC profiles, two groundwater 450 gauges and one climate station were almost continuously operated in an area of 10 ha at the agricultural research site in Marquardt, Germany. If aggregated, the eight CRNS core sensors -some among the most sensitive ones available for stationary CRNS detectors -provide a neutron count rate about 37 times higher than the one of a conventional Hydroinnova CRS-1000, and hence an unique signal-to-noise ratio for continuous measurements at this spatial scale. The core series were supplemented by a wide range of additional measurements:

455
-Additional CRNS sensors and SWC-profile measurements were implemented for shorter time periods (weeks to months).
Some of these additional measurements aimed to cover, at least temporarily, additional locations or soil depths (e.g., during the irrigation experiment). Others were co-located with the core sensors in order to allow for instrument comparisons; -Various intensive snapshot campaigns, including a large number of manual soil moisture measurements for ground truthing and soil mapping, CRNS roving and UAS-based hyperspectral remote sensing as part of an irrigation experi-460 ment, intensive monitoring and mapping of snow depth, density and coverage, and biomass mapping in areas with an -Detailed records of land use and irrigation management as part of the agricultural operations.
We extensively documented this data set and exemplarily highlighted various interesting features, including the representation of soil moisture from CRNS and profile probes during highly contrasting conditions such as drought, irrigation, heavy rainfall, and snow coverage. The long monitoring period of three years is a prerequisite to explore the sensor response to and the impact of such contrasting conditions with statistical significance.

470
This comprehensive data set provides the opportunity to investigate a diverse set of research problems, such as: the effects of vertical and horizontal soil moisture on the CRNS signal, in combination with the effects of biomass heterogeneity and snow cover; the retrieval of spatial soil moisture patterns from dense CRNS observations under consideration of different governing processes (irrigation, soil variability, cropping patterns) and with different levels of auxiliary data (e.g. CRNS roving, 475 hyperspectral remote sensing, soil water modelling); the response of different CRNS sensor types that were directly collocated (e.g., at location 11); preferential and bypass flow as well as surface to groundwater connectivity under heavy rainfall conditions; the potential of different sensor combinations to produce representative soil moisture estimates for heterogeneous landscapes, which could serve as a reference for remote sensing or hydrological modelling.

480
Consequently, the application of this dataset is not limited to the CRNS community, it can also serve as a valuable resource to various neighbouring disciplines, including soil and groundwater hydrology, agriculture, remote sensing and hydrological modelling. Currently, MqC is in the process of being extended: in addition to the dense core network of eight CRNS sensors, positions are being re-arranged and up to eight additional sensors are in the process of being added to achieve a coverage of a total area of at least 0.5 km 2 . This modification also implies that November 2022 is a natural end point of the McQ data set in 485 the configuration presented here.

Data availability
The published data set is organized along instruments and observed variables, and follows the structure of Sect. 3 of this paper (Tab. 4). Each subset of data is documented in a dedicated meta-data file in "json" format. Format conventions follow Fersch et al. (2020) and Heistermann et al. (2022a) and are summarized in a 'readme' file. We used EUDAT infrastructure 490 (https://eudat.eu), namely the services B2SHARE and B2HANDLE, in order to manage identifiers and guarantee long-term persistence. The repository's reference is https://doi.org/10.23728/b2share.edfdaa0d2a82477fa512bde3f53312f2 (Heistermann et al., 2022b). Author contributions. TF, MH, LS, and SO designed the study and coordinated the instrumentation; TF and MH coordinated the data management and led the writing of the manuscript; TF processed the soil moisture and snow measurements; MH processed the CRNS data 495 and prepared Figs. 2-6; LS processed the groundwater and soil hydraulic data; KDP processed the tensiometer and quantified the biomass; CB was responsible for setting up and maintaining the instrumentation and supported the data management; MS designed the rover campaign and processed the rover data; BT co-designed the instrument network and provided data on land use, yields and irrigation water use; DR and AG provided the data for the TDR-based soil moisture profiles; VD and MF provided the UAS-based hyperspectral data and related soil moisture products and LAI measurements; MK processed the data from the StyX sensors and supported their maintenance; LA developed the concept and implemented Fig. 1; NA processed the GNSS data; MZS supported the biomass quantification in the cherry and apple plantation; SO headed the MqC effort and is the principal investigator of this study; all authors contributed to writing and proofreading the manuscript.
Competing interests. Markus Köhli holds a CEO position at Styx Neutronica GmbH. content in southern England derived from a cosmic-ray soil moisture observing system -COSMOS-UK, Hydrological Processes, 30,