The Greenland Firn Compaction Verification and Reconnaissance (FirnCover) Dataset, 2013-2019

. Assessing changes in the density of snow and firn is vital to convert volume changes into mass changes on glaciers and ice sheets. Firn models simulate this process but typically rely upon steady-state assumptions and geographically and temporally limited sets of field measurements for validation. Given rapid changes recently observed in Greenland’s surface mass balance, a contemporary dataset measuring firn compaction in a range of climate zones across the Greenland ice sheet’s 15 accumulation zone is needed. To fill this need, the Firn Compaction Verification and Reconnaissance (FirnCover) dataset comprises daily measurements from 50 strainmeters installed in boreholes at eight sites on the Greenland ice sheet between 2013 and 2019. The dataset also includes daily records of two-meter air temperature, snow height, and snow temperature from each station. The majority of the FirnCover stations were installed in close proximity to automated weather stations that measure a wider suite of meteorological measurements, allowing the user access to auxiliary datasets for model validation 20 studies using FirnCover data. The dataset can be found here: https://www.doi.org/10.18739/A25X25D7M (MacFerrin et al., 2021).


Introduction 25
Mass loss from the Greenland ice sheet (GrIS) is currently one of the largest direct contributors to sea-level rise (IPCC, 2013), and the majority of that loss since the early 2000s has been due to significant increases in surface melt and runoff (Velicogna et al., 2014, van den Broeke et al., 2016Mottram et al., 2019). In Greenland's accumulation zone, which covers approximately 80% of the ice sheet (Box et al., 2006), annual snow accumulation is buried and densifies until it becomes glacial ice (Bader, 1954;Benson, 1962;Herron and Langway, 1980). Greenland's firn layer can be up to ~70 m thick (Schwander et al., 1997). 30 The GrIS's firn layer has been the subject of recent research for multiple reasons. First, assessments of Greenland's total mass balance using altimetry products use satellite-derived measurements of surface height to assess ice sheet volume, but need to resolve the evolution of the firn's porosity before converting volume change into mass change (i.e. Zwally et al., 2011;Shepherd et al., 2012;Csatho et al., 2014;McMillan et al., 2016, Smith et al., 2020. Second, the firn is able to retain part of the meltwater generated at the surface and buffer sea level rise (Pfeffer et al., 1991). The firn's retention capacity depends on 35 both the pore volume available for meltwater storage (Harper et al., 2012), which is decreasing (Vandecrux et al, 2019), and the firn's cold content, which is the energy required to bring the firn to the melting temperature (Vandecrux et al., 2020). Third, near-surface ice slabs have formed in western Greenland's firn. These slabs block percolation and reduce the buffering capacity, and thus promote lateral runoff (Machguth et al., 2016, MacFerrin et al. 2019. The development of these features is the result of increased melt volume (MacFerrin et al. 2019), increased near-surface firn densities, and sufficient cold content 40 to sustain meltwater refreezing (Vandecrux et al., 2020). Finally, knowing the depth and age of the firn-ice transition is important for the interpretation of climate records from ice cores (Schwander and Stauffer, 1984;Adolf and Albert., 2014). In all these cases, knowledge of the firn's compaction rate is crucial, yet to date there are relatively few in situ measurements of firn compaction, and there is no single, widely accepted model to simulate it. In this paper, we present the Firn Compaction Verification and Reconnaissance (FirnCover) dataset, which comprises measurements of firn compaction, depth-density 45 profiles, and temperatures from eight sites on the GrIS.

Background
Firn densification characterizes a general increase of the firn's bulk density and encompasses multiple processes. Firn compaction refers specifically to the compression of the firn due to overburden stress. Firn compaction occurs due to processes 50 operating at the grain scale such as grain boundary sliding, sintering mechanisms including dislocation creep and lattice diffusion, and plastic deformation (Herron and Langway, 1980;Morris and Wingham, 2014). Meltwater refreezing increases the firn density when surface meltwater or rain refreezes in the firn's pore space. This occurs primarily in the warmest regions of the ice sheet's accumulation area. The two above-mentioned phenomena are interconnected because meltwater refreezing releases latent heat and increases the firn temperature, which accelerates compaction of surrounding firn. In the highest-55 elevation zones of the ice sheet, where firn densification mainly occurs through compaction, the compaction rate in the nearsurface firn varies seasonally due to the fluctuating temperature; the deeper firn does not experience this seasonal variation in compaction rate. Long-term changes in climate (temperature and accumulation rate) may take many decades before they affect compaction rates over the full depth of the firn column (Li and Zwally, 2015). In the percolation zone, the seasonal cycle in near-surface firn compaction rate is also present. However, the infusion of meltwater can change the compaction rate on much 60 shorter timescales (days to weeks) as latent heat rapidly warms the firn, and rapid densification can occur when the refrozen meltwater fills the pore space. This firn may then compact more slowly in the future because of its higher density. In this realm, a single anomalous melt season can significantly affect the depth-density profile (Brown et al., 2012).  Langway, 1980;Arnaud et al., 2000;Zwally et al., 2011, Arthern et al., 2010Ligtenberg et al., 2011;Morris et al., 2014). On yearly and longer time scales, firn depth-density profiles and compaction rates can be estimated reasonably well using the mean annual air temperatures and accumulation rates (Herron and Langway, 1980). These firn-model results can be used e.g.
to simulate the long-term evolution of the firn-ice transition depth for ice-core studies (Goujon et al., 2003;Rasmussen et al., 2013). On shorter (monthly, daily, or sub-daily) time scales, firn models can be forced with weather data and/or outputs from 70 regional climate models (RCMs) to simulate the firn temperature, density, and thickness change (Vandecrux, et al., 2020). Results from these model runs can be used to correct repeat surface-elevation measurements from altimetry for firn changes (e.g. Smith et al., 2020). Numerous recent studies have coupled meltwater-percolation schemes to firn-compaction models (e.g. Reeh, 2008;Kuipers-Munneke et al., 2015;Vionnet et al,, 2012;van Pelt et al., 2016;Verjans et al., 2019;Vandecrux et al., 2020) to simulate liquid water content, refreezing, and runoff in the firn. 75 Most firn-densification schemes have generally been developed using density profiles observed in firn cores (Herron and Langway, 1980;Sørensen et al., 2011;Kuipers-Munneke et al., 2015). By assuming that the firn is in steady state, a dated depth-density profile can be converted to a densification rate. There are several potential issues with this method. First, it is not necessarily safe to assume that the firn at a given site is in steady state. Even if the firn is in steady state, a compaction rate 80 derived from the depth-density profile does not provide information about the firn's response to a transient climate or how its compaction rate varies on sub-annual timescales. Additionally, density profiles from the percolation zone cannot disentangle contributions of firn compaction and meltwater refreezing, which makes it difficult to assess these two processes in firn models.
Finally, a densification model that is tuned to match firn-density observations may be biased if there are errors or biases in the climate forcing that is used to tune the model. 85 Among the numerous firn models, none is broadly accepted as a definitive model. Lundin et al. (2017) showed that these models agree neither in steady-state nor in transient modes. Further, certain firn models are tuned specifically for Greenland or Antarctica, despite the fact that the physical processes driving densification should not vary solely due to geographic location. Vandecrux et al. (2020) compared numerous firn-meltwater models to observations and found that while different 90 models accurately simulated physical characteristics of different firn zones in Greenland, no single model accurately represented firn density, temperature and water content at all sites.
The uncertainties associated with firn-model development and the disagreement among the existing models underscore the need for direct measurements of firn compaction. Such measurements are currently rare and from only isolated regions of an 95 ice sheet (Hamilton et al., 1998;Arthern et al., 2010;Morris and Wingham 2014;Hubbard et al. 2020). To fill this knowledge gap and increase our understanding of firn densification, we present data from the Firn Compaction Verification and Reconnaissance (FirnCover) project, which monitored firn compaction between 2013 and 2019 at eight stations on the GrIS. Each station monitors firn compaction with strainmeters installed over boreholes at various depth ranges, 100 as well as firn temperature, air temperature and surface height. Additionally, we measured depth-density and stratigraphy profiles of recovered cores and in snow pits during each field visit. In this paper, we describe the FirnCover stations (Section 3) and the dataset organization (Section 4), and then we present a preliminary analysis of the dataset (Section 5).

The FirnCover stations and dataset
The eight FirnCover stations are located in various climate zones of the ice sheet accumulation area ( Figure 1, Table 1) Each station included a suite of instruments, which we detail below, and was equipped with a tower to hold instrumentation, a data logger (Campbell CR800), a solar panel, and a battery. Borehole strain rates were recorded daily, while air temperature, surface height, and firn temperature measurements were recorded hourly. During most years, summary data from the 115 instruments was transmitted from each station once per day using an Iridium short-burst data modem. Full data tables were saved on the data logger and were read from each station upon visits in the field, which usually occurred in late April or early May.

The FirnCover strainmeters
The main components of each FirnCover station were borehole strainmeters, which made daily measurements of borehole 125 lengths. These used the "coffee-can" method (Hulbe and Whillans, 1994;Hamilton et al., 1998) to continuously monitor firn compaction, similar to the method used by Arthern et al. (2010). Each instrument was composed of a line with a weight attached to one end and connected to a spring-loaded potentiometer on the other end. The weight was anchored at the bottom of a borehole, and the potentiometer was placed at the top of the borehole. As the borehole shortened due to firn compaction, the potentiometer reeled in string to maintain tension (Figure 2), and a data logger recorded the length of string that had been 130 reeled in. A total of 50 strainmeters were installed at the 8 FirnCover stations. Table 2 lists metadata for each instrument, including the initial depths of the boreholes. Two of the instruments are missing data altogether and their details are not included in Table   2. 135

Figure 2: FirnCover station (left), strainmeter casing (inset), and strainmeter conceptual design (right)
The potentiometers were high-precision analog HX-PA units from Unimeasure, Inc. (Bend, Oregon) with a 2.032 m range. 140 The end of the potentiometer's steel extension wire was attached to a Vectran string that extends to the bottom of the borehole.
The string was anchored using a 0.226 kg lead weight. Each potentiometer was independently calibrated to accurately measure, within ± 1 mm, the length of the extension wire that was pulled out of the potentiometer. The potentiometer was enclosed in a weatherproof plastic case with an opening at the bottom. To stabilize the instrument atop the borehole, it was installed atop a 0.61 m 2 white PVC plastic platform. A section of PVC pipe (0.1-0.7 m long) was attached to the bottom of the casing and 145 inserted in the top of the borehole to prevent the collapse of the top of the borehole and keep the instrument in place. The line lowered in the borehole was covered with hydrophobic lithium grease to prevent water from refreezing on it and to keep the line from snagging on the instrument or freezing to the side of the borehole.
To install each instrument, a borehole was drilled into the snow and firn using a Kovacs (diameter 9 cm) coring drill. The 150 weighted Vectran string was then lowered into the borehole, and the potentiometer platform was placed atop of the borehole ( Figure 2). The length of the string was set so that the potentiometer's steel cable was near its full extension, maximizing the distance over which the borehole shortening could be observed. Some instruments were installed on the surface and thus (Summit, EastGRIP) all instruments were installed directly on the surface, while instruments in the percolation zone were mixed between surface and snow-pit installations. The depth of each borehole was measured both along the core (by reassembling core segments on the surface) and by using the Vectran line to directly measure the borehole. Instruments #1-10, installed in 2013, use the approximate core length (as borehole length was not measured); the remaining instruments use the measured borehole length. Borehole and core-length measurements typically agreed to within 0-8 cm. 160

Air temperature, surface height and firn temperature observations
Each FirnCover station was equipped with a Campbell L109 air-temperature thermistor with 6-plate radiation shield, which 165 measured air temperature hourly at approximately 2 m ground height. Snow-surface height was measured from 2015 onward with a SR50 sonic-ranging sensor mounted on the tower cross-beam. A string of 24 10-KΩ resistance-temperature diodes (RTDs, from Omega sanitary, Class A, IEC 60751 standard) measured firn temperatures from 0 to approximately 14 m depth (every 0.5 m from 0-10 m depth, every 1 m thereafter). The manufacturer-stated precision of the RTDs is ±0.2 o C. Some RTDstring boreholes were less than 14 m due to challenges clearing chips from the boreholes. RTD measurements are corrected 170 for wire resistance (by measuring across a 25 th bare wire without an RTD), and measured resistances are converted to temperature using formulae provided by the RTD manufacturer. The initial installation depths of each RTD string are noted in the FirnCover_Station_Metadata data table (Table A5). The daily depth of each thermistor is calculated by adding the original installation depth to the snow depth measured by the sonic ranging sensor. 175

Firn core and snow pit observations
Firn cores were recovered from each of the FirnCover strainmeter boreholes. To understand the structure of the firn at each FirnCover instrument, the cores were visually inspected for stratigraphic layers (ice lenses, etc.) at ~1 cm vertical resolution, and cut into segments to measure density at ~10 cm resolution. At some sites where multiple FirnCover instruments were installed in the same visit, only one core was logged and nearby boreholes were assumed to have similar stratigraphy and 180 density profiles. Density profiles from all cores logged by FirnCover field campaigns are included in the 2018 release of NASA's SumUp dataset (Montgomery et al., 2018). The table "Compaction_Instrument_Metadata" (Table 3 and

Dataset structure and handling
The FirnCover dataset is organized in a single .hdf5 file, which comprises four data tables and three metadata tables. Table 3 gives a summary of the data tables, and tables A1 to A7 detail the variables contained in each table. The first records of compaction for each strain meter may be subject to error due to settling effects. For the analyses presented here, we discard the first month of recordings for each instrument. This number is based on data from an instrument at KAN_U that was installed in a borehole comprising solid ice, which would be expected to have zero compaction but is still subject to instrument settling. That instrument registered a compaction signal for approximately one month before recording zero 195 compaction for the remainder of its observations. We leave data from the initial month in the dataset so that more delicate filtering may allow the recovery of more observation within that period, but we recommend to potential users that it be discarded.
Some compaction data was read directly from the station's data logger in 32-bit floating point format. For measurements where 200 data tables were unable to be directly read due to lack of re-visit, data summaries from Iridium transmissions were used with 16-bit floating point values. Due to the limited data resolution, borehole lengths recorded from Iridium transmissions exhibit a 2 mm stepwise discretization rather than smooth continuous measurements. This can influence compaction rates when computed as derivatives of borehole lengths over time. In the present analysis, we use a two-month-wide running mean to smooth the borehole length. This filtering removes most of the noise, but it may also smooth part of seasonal changes of 205 compaction rates. The dataset includes the unfiltered data to allow users to use their own filtering strategy. are periods of data that we consider suspicious. These are listed in Table 4. We exclude these suspicious data from our analysis in Section 5. They are included in the released dataset, but we advise using caution when using them.

5.
Data overview and preliminary analysis  The difference between sites can be further investigated by looking at the compaction rates, which are calculated by taking the 235 time derivative of the borehole length data ( Figure 4). As discussed above, NASA-SE shows the largest daily changes and KAN-U the lowest. At each site, instruments installed in deeper boreholes show larger magnitude of daily compaction compared to shorter instruments. The faster compaction after the installation of the instrument appears as peaks in firn compaction rates in Figure 4. Since we already discard the settling period of the instrument, we attribute these faster initial compaction rates to the high potential for deformation of low-density, relatively new snow/firn in which the instruments are 240 installed. Faster compaction during the first summer following the installation of the instruments also stems from the conduction of warmer surface temperatures down to the instrument during summer. These warmer firn temperatures during summer increase the firn compaction rates. In spite of the smoothing applied, KAN-U, Crawford Point, EKT and Dye-2 still show notable noise. They are the sites that undergo the highest melt; we hypothesize that this noise is due to interaction between meltwater and the borehole. Targeted noise removal strategies may be necessary at these sites for any potential users. As 245 mentioned previously, daily compaction rates at KAN-U are lower than at other sites due to the presence of a ~5 m thick ice slab at that site. The seasonality of the daily compaction rates is clearly visible at the dry snow sites, Summit and EastGRIP, but also at sites in the percolation areas: NASA-SE, Crawford Point , EKT and Saddle. Daily compaction rates peak in the autumn and reach a minimum at the end of the winter. The delay between the highest (resp. lowest) surface temperatures in summer (resp. winter) and the highest (resp. lowest) compaction rates is due to the time the surface temperatures need to 250 diffuse down to the depth of the firn that the instrument is measuring.  Table 2.

255
The FirnCover dataset also includes air temperature, surface height, and firn-temperature measurements (at all sites except EastGRIP and NASA-SE); these data enable us to relate the compaction rates to each year's weather conditions (Figures 5   and 6).   The firn temperature measurements, in particular, allow analyses using the actual firn conditions rather than using average air temperature as a proxy for firn temperature, which is commonly done. The comparison of average air temperature and interpolated 10m firn temperatures indicate that they are rarely equivalent (Table 4). At Summit, the 10m firn temperatures are 2.6 o C lower than the average air temperature. This is due to strong near-surface atmospheric inversion and radiative cooling 270 of the surface (e.g. Miller et al., 2017). At all the other sites, the 2 m firn temperature is higher than the average air temperature.
We attribute this to meltwater percolation and latent heat release at depth (e.g. Humphrey et al., 2012). This difference is largest at KAN-U where the firn is 7 o C warmer than the average air temperature. At Saddle, the firn temperature is within a degree of the average air temperature. We interpret this as the neutralization of the two processes mentioned above: heat loss through radiative cooling at the surface and latent heat release during meltwater refreezing. This site-specific difference 275 between 10 m firn temperature and average air temperature shows the limitation of firn compaction parameterizations that use air temperature as a proxy for firn temperature and how these parameterizations perform outside of their training site.

Summary remarks 280
We present data from 50 strainmeters installed at eight sites located in different climatic zones of the Greenland ice sheet and covering the 2013-2019 period. Additional surface and firn measurements available at each of the FirnCover sites are firn density, air temperature, surface height and firn temperatures. These data will allow future work to investigate the interannual and seasonal response of firn compaction to surface and subsurface conditions. We also note that several other measurements are available at some of the FirnCover sites: at KAN-U the PROMICE automatic weather station has been operating since 285 2009(Ahlstrøm et al., 2008; at Crawford Point, Saddle, NASA-SE, Summit and Dye-2, GC-Net weather stations document the history of these sites back to the 1990s and are still operating. At Summit, extensive instrumentation is measuring the atmospheric conditions and the surface energy budget (e.g. Miller et al., 2017). At Dye-2, upward looking Ground Penetrating Radar (Heilig et al., 2018) and time-domain resistivity probes (Samimi et al., 2020)  investigations of how meltwater affects firn compaction. The FirnCover dataset will help to evaluate and calibrate firn models and help reduce uncertainty when using these models to interpret satellite altimetry measurements or calculating the past, current and future mass balance of polar ice sheets. The dataset can be found here: https://www.doi.org/10.18739/A25X25D7M.

Code Availability
All the scripts used to load, process and plot the FirnCover dataset are available here: 300 https://github.com/BaptisteVandecrux/FirnCover.

Acknowledgements
The majority of this work, including instrumentation and station visits, was funded by NASA awards NNX15AC62G and NNX10AR76G. All authors acknowledge the work and efforts of multiple field logistic partners and team members for their 305 essential help in maintaining the datasets and instruments: K. Alley, C. Charalampidis, W. Colgan, F. Covi, A. Crawford, M. Eijkelboom, S. Grigsby, A. Heilig, D. Hill, H. Machguth, S. Marshall, A. Rennermalm, S. Samimi, T. Snow, A. Sommers, D. van As, the Summit Station scientific team, the EastGrip team and Polar Field Services.

10.
Author contribution MM conducted the conceptualization, funding acquisition, the methodology, the field investigation and the data curation. CMS 310 participated to the conceptualization, funding acquisition, field investigation, formal analysis and visualization. BV participated to the field investigation, formal analysis and visualization. EW and WA participated to the funding acquisition and supervision. All authors contributed to the manuscript preparation.

Competing interests
The authors declare that they have no conflict of interest.    The raw distance measured by the sonic ranger, before temperature correction. SonicRangeAirTemp_C

Appendix
The air temperature at the time of the sonic ranger measurement. SonicRangeDist_Corrected_m The corrected distance measured by the sonic ranger. Accum_Snow_Depth_m The accumulated snow depth since the instruments' installation, corrected for tower raises upon revisits.  latitude The WGS84 latitude of the station upon installation longitude

325
The WGS84 longitude of the station upon installation installation_daynumer_YYYYMMDD The day the station was installed. comments General comments about the station upon its installation.

RTD_stringnumber
The string serial number of the RTD string installed at the station.

RTD_installation_daynumber_YYYYMMDD
The day at the RTD was installed at the station. RTD_top_usable_RTD_num Number (from the top) of the first usable RTD sensor. Non-usable sensors could not be inserted in the snow and were left lying on the surface.

RTD_depths_at_installation_m
The 24-length depths of each RTD at installation.

RTD_direction_from_tower_degrees
The compass direction (non-corrected for declination) of the station tower to the RTD string.

RTD_distance_from_tower_m
The distance from the station tower to the RTD string.

Field Name Comments instrument_ID
Unique identification number of the instrument sitename Name of the FirnCover site installation_daynumber_YYYYMMDD Date that the instrument was installed. borehole_top_from_surface_m The top of the borehole from the surface upon installation, in m. (0 for surface, negative numbers for beneath the surface) borehole_bottom_from_surface_m The depth from the surface to the bottom of the borehole, in m borehole_initial_length_m The length of the borehole, in m instrument_has_wire_correction Whether the instrument installed has a wireresistance correction sensor installed, or not. direction_from_tower_degrees The compass direction (not corrected for declination) from the tower to the instrument. distance_from_tower_m The distance (in m) from the tower to the instrument. borehole_ID The identifying name of the core taken from the borehole, where stratigraphy and density were measured (names consistent with cores in the NASA SumUp dataset). borehole_ID_is_direct A "direct" (True) core density profile came straight from that borehole. If "indirect" (False), that core was not profiled for density directly, and this links to a nearby. Adjacent core measured at the same time, typically within 10-20 meters distance.