A 16-year record (2002–2017) of permafrost, active-layer, and meteorological conditions at the Samoylov Island Arctic permafrost research site, Lena River delta, northern Siberia: an opportunity to validate remote-sensing data and land surface, snow, and permafrost models

Most of the world’s permafrost is located in the Arctic, where its frozen organic carbon content makes it a potentially important influence on the global climate system. The Arctic climate appears to be changing more rapidly than the lower latitudes, but observational data density in the region is low. Permafrost thaw and carbon release into the atmosphere, as well as snow cover changes, are positive feedback mechanisms that have the potential for climate warming. It is therefore particularly important to understand the links between the energy balance, which can vary rapidly over hourly to annual timescales, and permafrost conditions, which changes slowly on decadal to centennial timescales. This requires long-term observational data such as that available from the Samoylov research site in northern Siberia, where meteorological parameters, energy balance, and subsurface observations have been recorded since 1998. This paper presents the temporal data set produced between 2002 and 2017, explaining the instrumentation, calibration, processing, and data quality control. Furthermore, we Published by Copernicus Publications. 262 J. Boike et al.: A 16-year record (2002–2017) of permafrost, active-layer, and meteorological conditions present a merged data set of the parameters, which were measured from 1998 onwards. Additional data include a high-resolution digital terrain model (DTM) obtained from terrestrial lidar laser scanning. Since the data provide observations of temporally variable parameters that influence energy fluxes between permafrost, active-layer soils, and the atmosphere (such as snow depth and soil moisture content), they are suitable for calibrating and quantifying the dynamics of permafrost as a component in earth system models. The data also include soil properties beneath different microtopographic features (a polygon centre, a rim, a slope, and a trough), yielding much-needed information on landscape heterogeneity for use in land surface modelling. For the record from 1998 to 2017, the average mean annual air temperature was −12.3 C, with mean monthly temperature of the warmest month (July) recorded as 9.5 C and for the coldest month (February) −32.7 C. The average annual rainfall was 169 mm. The depth of zero annual amplitude is at 20.75 m. At this depth, the temperature has increased from −9.1 C in 2006 to −7.7 C in 2017. The presented data are freely available through the PANGAEA (https://doi.org/10.1594/PANGAEA.891142) and Zenodo (https://zenodo.org/record/2223709, last access: 6 February 2019) websites.


Abstract.
Most of the world's permafrost is located in the Arctic, where its frozen organic carbon content makes it a potentially important influence on the global climate system. The Arctic climate appears to be changing more rapidly than the lower latitudes, but observational 35 data density in the region is low. Permafrost thaw and carbon release into the atmosphere, as well as snow cover changes, are positive feedback mechanisms that have the potential for climate warming. It is therefore particularly important to understand the links between the energy balance, which can vary rapidly over hourly to annual time scales, and permafrost conditions, which changes slowly on decadal to centennial timescales. This requires long-term 40 observational data such as that available from the Samoylov research site in northern Siberia, where meteorological parameters, energy balance, and subsurface observations have been recorded since 1998. This paper presents the temporal data set produced between 2002 and 2017, explaining the instrumentation, calibration, processing and data quality control.
Furthermore, we present a merged dataset of the parameters, which were measured from 1998 45 onwards. Additional data include a high-resolution digital terrain model (DTM) obtained from terrestrial LiDAR laser scanning. Since the data provide observations of temporally variable parameters that influence energy fluxes between permafrost, active layer soils, and the atmosphere (such as snow depth and soil moisture content), they are suitable for calibrating and quantifying the dynamics of permafrost as a component in earth system models. The data also 50 include soil properties beneath different microtopographic features (a polygon center, a rim, a slope, and a trough), yielding much-needed information on landscape heterogeneity for use in land surface modeling.
For the record from 1998 to 2017, the average mean annual air temperature was -12.3 °C, with mean monthly temperature of the warmest month (July) recorded as 9.5 °C and for the coldest 55 month (February) -32.7 °C. The average annual rainfall was 169 mm. The depth of zero annual

Introduction
Permafrost, which is defined as ground that remains frozen continuously for two years or more, 65 underlies large parts of the land surface in the northern hemisphere, amounting to about 15 million km 2 (Aalto et al., 2018;Brown et al., 1998;Zhang et al., 2000). The temperature range and the water and ice content of the upper soil layer of seasonally freezing and thawing ground (the active layer) determine the biological and hydrological processes that operate within this layer. Warming of permafrost over the last few decades has been reported from 70 many circum-Arctic boreholes (Biskaborn et al., 2018;Romanovsky et al., 2010). Warming and thawing of permafrost and an overall reduction in the area that it covers have been predicted under future climate change scenarios by the CMIP5 climate models, but at widely varying rates (Koven et al., 2012;McGuire et al., 2018). Continued observations, not only of the thermal state of permafrost but also of the multiple other types of data required to understand the 75 changes to permafrost, are therefore of great importance. The data required include information on conditions at the upper boundary of the soil (specifically on snow cover), on atmospheric conditions, and on various subsurface state variables (such as, e.g., soil volumetric liquid water content and soil temperature). The seasonal snow cover in Arctic permafrost regions can blanket the land surface for many months of the year and has an important effect on the thermal regime 80 of permafrost-affected soils . The soil's water content determines not only its hydrological and thermal properties, but also the energy exchange (including latent heat conversion or release) and biogeochemical processes.
In view of these dependencies, the data sets presented here, including snow cover and the thermal state of the soil and permafrost, together with meteorological data, will be of great value 85 (i) for evaluating permafrost models or land surface models, (ii) for satellite calibration and validation (cal/val) missions, (iii) in continuing baseline studies for future trend analysis (for example, of the permafrost's thermal state), and (iv) for biological or biogeochemical studies.
The Samoylov research site in the Lena River Delta of the Russian Arctic has been investigated by the Alfred Wegener Institute Helmholtz Center for Polar and Marine Research (AWI), in 90 collaboration with Russian and German academic partners, since 1998. The land surface characteristics and basic climate parameter data collected between 1998 and 2011 have been previously published in Boike et al. (2013). Major developments in earth system models, for example through the European PAGE21 project (www.page21.org), the Permafrost Carbon Network projects (www.permafrostcarbon.org), satellite calibration and validation missions, 95 and observations through the Global Terrestrial Network on Permafrost (GTN-P) have subsequently led to sustained interest from a broader modelling community in the data obtained.
In this publication we provide information on the research site and a full documentation of the data set collected between 2002 and 2017, which can be used for forcing and validation of earth system models (see e.g. Chadburn et al., 2015;Chadburn et al., 2017;Ekici et al., 2014;Ekici 100 et al., 2015). We present data that incorporate subsurface thermal and hydrologic components, of heat flux as well as snow cover properties, and meteorological data from the Samoylov research site, similar to the data published previously for a Spitsbergen permafrost site . 105 The Samoylov research site is located within the continuous permafrost zone on Samoylov Island in the Lena River Delta, Siberia (Figure 1). It has been a site for intensive monitoring of soil temperatures and meteorological conditions since 1998 .

Site description
The region is characterized by an Arctic continental climate with low mean annual air temperature of below -12 °C, very cold minimum winter air temperatures (below -45 °C), and 110 summer air temperatures that can exceed 25 °C, a thin snow cover and a summer water balance equilibrated between precipitation input and evapotranspiration .
The study area of the Lena River Delta has permafrost to depths of between 400 and 600 m (Grigoriev, 1960). The active layer thawing period starts at the end of May and active layer thickness reaches a maximum at the end of August/beginning of September. Marked warming 115 of this area over the last 200 years has been inferred from temperature reconstruction using deep borehole permafrost temperature measurements in the delta and the broader Laptev Sea region (Kneier et al., 2018). Samoylov Island is located within a deltaic setting, consists of a flood plain in the western part of the island and a Holocene terrace characterized by ice-wedge polygonal tundra and larger 120 waterbodies in the eastern part ( Figure 1).
The area is generally characterized by ice-rich organic alluvial deposits, with an average ice content in the upper meter of more than 65% by volume for the Holocene terrace and of about 35% for the flood plain deposits (Zubrzycki et al., 2013). The Holocene terrace is dominated by ice wedge polygons so that a considerable volume of the upper soil layer (0-10 m) is 125 characterized by excess ground ice . Degradation of ice wedges, as observed throughout the Arctic (Liljedahl et al., 2016), occurs at only a few, localized parts of the research site (Kutzbach, 2006). The recent work by Nitzbon et al. (2018) shows that the spatial variability in the types of ice-wedge polygons observed at this study area can be linked to the spatial variability in the hydrological conditions. Furthermore, wetter hydrological 130 conditions have a destabilizing effect on ice wedges and enhance degradation.
The total mapped area of the polygonal tundra on Samoylov Island (excluding the floodplain) is composed of 58% dry tundra, 17% wet tundra and 25% water surfaces, of which 10% are overgrown water and 15% open water (Muster et al., 2012, Figure 3a). The landscape is characterized by polygonal tundra, i.e. a complex mosaic of low-and high-centered polygons 135 (with moist to dry polygonal ridges and wet depressed centers) and larger waterbodies (Muster, 2013;Muster et al., 2012). The polygonal tundra microtopography, polygon rims, slopes, and depressed centers are clearly distinguishable. Depressed polygon centers are typically watersaturated or have water levels above the ground surface (shallow ponds). High-centered polygons have inverse microtopography, i.e. drier elevated centers and wet surrounding 140 troughs. Polygonal ponds and troughs make up about 35% of the total water surface area on the island .

Data description 155
This paper presents for the first time a complete data archive and descriptions in he form of the following data sets: (i) a full range of meteorological, soil thermal, and hydrologic data from the research site covering the period between 2002 and 2017 ( Figure 2), (ii) high spatial resolution data from terrestrial laser scanning of the research site completed in 2017, with resulting data sets for a digital terrain model and for vegetation height, (iii) time-lapse camera 160 images, and (iv) a data set containing specially compiled or processed data sets for those parameters that were measured in the period from 1998 to 2002, thus extending the record to form a long-term data set, as initiated in Boike et al. (2013). The processing and level structure is described in detail in Section 4. Additional data such as soil properties and soil carbon content are also included in this paper in order to provide a complete set of data and parameters suitable 165 for earth system, conceptual and land surface modeling. All of these data are archived in the PANGAEA data libraries and the measuring principles and analysis are described in this paper.
Data logging between 2002 and 2013 at the research site was powered by a solar panel and a wind turbine generator and the data was retrieved manually during site visits once or twice a year, when visual inspections were also made of the sensors. Data gaps prior to 2013 resulted 170 mainly from problems with the site's energy supply, such as problems with the solar/wind charge controller. No other gap filling has been undertaken, but previous publications (e.g. Langer et al., 2013) suggest that reanalysis data, such as ERA-Interim, could be used for this purpose. In Chadburn et al. (2017), a method for correcting reanalysis data to better represent the site is described and applied. The gap-free meteorological dataset that was produced and 175 used in Chadburn et al. (2017) is now available on the PANGAEA database (Burke et al. 2018), making it easy for modellers to begin running the Samoylov site and therefore to make good use of our data.
Since 2013 the research site has been connected to the main electricity supply of the new Russian Research Station, resulting in much improved data collection with almost no data gaps.

180
Details of the sensors used are provided in the following sections, as well as descriptions of the data quality and cleaning routine (Section 4). The instruments can be divided into above-ground sensors (meteorological) and below-ground sensors (e.g. soil sensors). Further detailed information on the sensors can be found in Table 2, which summarizes all of the instruments and relevant parameters, as well as in the appendices B to H (metadata, description of 185 instruments, and calculations of final parameters). Figure 2 presents a time series of selected parameters measured between 2002 and 2017.

Meteorological station data
The standard meteorological variables described in this section were averaged over various intervals (Table 2) with the averages, sums, and individual values all being saved hourly until 190 2009 and half-hourly thereafter. The sampling intervals changed as a result of different logger and sensor setups and different available power sources. Sensors were connected directly to data loggers. A number of different data logger models from Campbell Scientific were used over the years (CR10X between 2002and 2009, CR200 between 2007 since 2009), together with an AM16/32A multiplexer.

Air temperature, relative humidity
Air temperature and relative humidity were measured at 0.5 m and 2 m above the ground (starting with hourly averages at 2.0 m until 30 June 2009 and at 0.5 m until 26 July 2010, with half-hourly averages thereafter) using Rotronic and Vaisala air temperature and relative humidity probes protected by unventilated shields ( Figure B1 and Table 2). According to the 200 sensor's manuals, the HMP45 sensors have a measurement limit of -39.2 °C, but we recorded data down to -39.8°C. During extreme cold air temperature periods, for example, between 1 February and 15 March, 2013, constant air temperature values were recorded at the sensor's output limit. These data periods were manually flagged (Flag 6: consistency; Table 3) using a lower temperature limit of -39.5 °C.

205
Also of importance is the decrease in accuracy of the air temperature and humidity data with decreasing temperature and moisture content. For example, the accuracy for the HMP45A sensor at 20 °C is ±0.2 °C, but at -40 °C it is ±0.5 °C. Campbell Scientific PT100 temperature sensors were installed on 22 August 2013 alongside the temperature and humidity probes, at the same heights but in separate unventilated shields, in order to circumvent this problem. Since 210 17 September 2017 Vaisala HMP155A air temperature and relative humidity probes were installed which enable the full range of temperatures (below -40 °C). The uncertainty in all the temperature measurements ranges between 0.03 and 0.5 °C, depending on the sensors used; the uncertainty in the relative humidity measurements ranges between 2 and 3%. The measurement heights were not adjusted with respect to the snow surface during periods of snow cover 215 accumulation or ablation. The lower probes (at 0.5 m) were only completely snow-covered during two months of the 2017 winter season (16 April-11 June 2017), as observed in photographic images, and therefore this time period is flagged in the data series (Flag 8: snowcovered; Table 3).

220
The wind speed and direction were measured using a propeller anemometer (R.M. Young Company 05103, Figure B2), which was calibrated towards geographic north. This was done by orienting the center line of the sensor towards true north (using a GPS reference point) and then rotating the sensor base until the datalogger indicated zero degrees. The averaged wind direction, its standard deviation, and the wind speed were all recorded at hourly intervals until 225 30 June 2009 and at half-hourly intervals thereafter. Since August 2015, wind maximum and minimum wind speed are also recorded. The mean wind speeds and directions were calculated using every value recorded during the measurement interval. The standard deviation of the wind direction was calculated using the algorithm provided by the Campbell Scientific data logger.

230
The net radiation was measured between 2002 and 2009 using a Kipp & Zonen NR LITE net radiometer; outgoing longwave radiation was also measured using a Kipp & Zonen CG1 pyrgeometer. Since 2009, various 4-component radiometers were used (Table 2). The averaged values were stored at hourly intervals until 30 June 2009 and at half-hourly intervals thereafter.
Further details of the measuring periods and the specifications for the different sensors can be 235 found in Table 2. Although all radiation sensors were checked for condensation, dirt, physical damage, hoar frost, and snow coverage during the regular site visits, the instruments were largely unattended and their accuracy is therefore estimated to have been ±10%. Our quality analysis also includes flagging the data during those periods where short wave incoming radiation was lower than shortwave outgoing radiation by 10 W m -² using Flag 6 (plausibility, Kipp & Zonen CNR4 four-component radiation sensor is operative, together with a CNF4 ventilation unit to prevent condensation ( Figure B3). The additional heating available for the CNR4 sensor was never used.

Rainfall
Un-heated and un-shielded tipping bucket rain gauges (Environmental Measurements ARG100 and R. M. Young Company model 52203) were installed directly on the ground on 31 August 2002 (ARG100) and 26 July 2010 (52203). The Environmental Measurements ARG100 liquid precipitation probe was damaged during the winter of 2009/2010. By installing the gauge close 250 to the ground the risk of wind-induced tipping of the bucket which would lead to false data records, can be reduced (as observed by Boike et al., 2018). Due to the typically low snow heights, the risk of snow coverage of the instrument is also very low.
The instruments measure only liquid precipitation (rainfall) and not winter snowfall. The tipping buckets were checked regularly during every summer by pouring a known volume of 255 water into the bucket and carrying out frequent visual inspections for dirt or snow during each site visit. These calibration data are flagged with Flag 3 (maintenance periods).

Snow depth
The snow depth around the station has been continuously monitored since 2002 using a Campbell Scientific SR50 sonic ranging sensor ( Figure B4). The sensor measures distance 260 between the sensor and an object or surface which could be the upper surface of the snow (in winter), or the water surface, ground surface, or vegetation (in summer). On 17 July 2015 a metal plate was placed directly beneath the ultrasonic beam to reduce the amount of noise in the reflected signal due to surface vegetation ( Figure B4). The acoustic distance data obtained from the sonic sensor were temperature-corrected using the formula provided by the 265 manufacturer (Appendix C) using the air temperature measured at the Samoylov meteorological station.
To obtain the snow depth, the distance of the sensor from the surface was recorded over the summer and the mean calculated. The recorded (corrected) winter distances are then subtracted from this mean (previous) summer value to obtain snow depth. Due to seasonal thawing the 270 ground surface can subside by a few centimeters over the summer season (and therefore no longer be set to zero) resulting in negative heights for the ground surface level being computed.
In contrast, vegetation growth and higher water levels (e.g. as observed in 2017) will result in positive heights. The distance measurements collected during the snow-free season are not removed from the series or corrected since they provide potentially useful information about 275 these processes.
The SR50 sensor acquires data over a discoidal surface with a radius that ranges from 0.23 m (0.17 m 2 ) in snow-free conditions to 0.19 m (0.12 m 2 ) with 20 cm of snow. This footprint disk is located in the center of a low-centered polygon for which the spatial variability of snow has been investigated by Gouttevin et al. (2018). The microtopography of this polygonal tundra 280 (characterized by rims, slopes and polygon centers) was identified as a profound driver of spatial variability in snow depth: at maximum accumulation in 2013 rims typically had 50% less snow cover and slopes 40% more snow cover than polygon centers. However, the snow cover within each topographical unit also exhibited spatial variability on a decimeter scale , probably resulting from underlying micro relief (notably vegetation 285 tussocks) and processes such as wind erosion. This variability can affect the representativity of the SR50-measured snow depth data and visual data obtained from time-lapse photography can therefore be extremely important (see next section).

Time lapse photography of snow cover and land surface
In order to monitor the timing and pattern of snow melt an automated camera system (Campbell 290 Scientific CC640) was set up in September 2006 to photograph the land surface in the area in which the instruments were located ( Figures B5 to B7). The images are used as a secondary check on the snow cover figures obtained from the depth sensor and are also valuable for monitoring the spatial variability of snow cover across polygon microtopography. During the polar night the image quality was found to be somewhat reduced and a second camera with a 295 better resolution (Campbell Scientific CC5MPX) was therefore installed in August 2015 to record high-quality images in low-light conditions over the winter period.

Atmospheric pressure
A Vaisala PTB110 sensor in a vented box was installed next to the data loggers at the meteorological station ( Figure B1) in August 2014 to measure atmospheric pressure.

Water levels
The suprapermafrost ground water level, i.e. water level of the seasonally thawed active layer above the permafrost table within one polygon, was estimated using Campbell Scientific CS616 and CS625 water content reflectometer probes installed vertically in the soil and air, with the sensor's ends standing upright (Appendix D). The advantage of this method is that the sensor 305 can remain in the soil during freezing and subzero temperatures, whereas pressure transducers need to be removed over winter and then reinstalled. For the unfrozen periods, the soil as measured by a dielectric device is a mixture of air, water, and soil particles.The sensor outputs a signal period measurement from which usually the bulk dielectric number is calculated. The dielectric number (also referred to as the relative permittivity or dielectric constant) is then used 310 to calculate the volumetric water content using an empirical polynomial calibration provided by the manufacturer. We use the signal period output of the CS616 and CS625 water content reflectometer probes (Campbell Scientific, 2016) and a site-specific calibration to convert to water level with respect to the sensor base (Appendix D).

Instrument installation at the soil station and soil sampling
In order to take into account any possible effects of heterogeneity in vegetation and microtopography at the research site (e.g. due to the presence of polygons), instruments for measuring the soil's thermal and hydrologic dynamics (Table 2)  Sensors were installed to cover the entire depth range of the profile, i.e. from the very top, through the active layer and into the permafrost soil. The sensors were positioned according to 335 the soil horizons so that every horizon in the profile contained at least one probe.
Sensors were installed horizontally into the undisturbed soil profile face beneath different microtopographical features and the pits were then backfilled ( Figures B10 and B13)).
Soil samples were collected before instrument installation so that physical parameters could be analyzed. Soil properties within the soil profiles, including the soil organic carbon (OC) content,

Soil temperature
Soil temperature sensors were installed over vertical 1D profiles in 2002 beneath a polygon center, slope, and rim. A measurement chain of temperature sensors was also installed in the ice wedge down to a depth of 220 cm. Their positions are shown in Figure B13. The temperatures were initially measured using Campbell Scientific 107 thermistors connected to a 350 Campbell Scientific CR10X data logger with a Campbell Scientific AM416 multiplexer.
Campbell Scientific's "worst case" example, with all errors considered to be additive, is given as ±0.3 °C between -25 and 50 °C. The average deviation from 0 °C determined through ice bath calibration prior to installation was 0.008 °C (maximum: 1.0 °C; minimum: -0.56 °C, standard deviation: 0.33 °C). The sensors cannot be re-calibrated once they have been installed.

355
Phase change temperatures during spring thaw and fall refreezing are stable (the zero-curtain effect in freezing and thawing soils of periglacial regions). Assuming that freezing point depression (due to the soil type and soil water composition) does not change significantly from year to year, these periods can be used to evaluate sensor stability. Between 2002 and 2009 the data logger and multiplexer were not replaced which resulted in a reduced accuracy of up to 360 ±0.7 °C during the winter freeze-back periods in 2009 for two of the sensors near to the surface (center of the polygon at -1 cm, rim of the polygon at -2 cm below ground surface, respectively).
The zero curtain period during fall -winter, where temperatures in the ground are stabilized at 0°C during phase change, offers an accuracy test for sensors that cannot be retrieved. For the remaining sensors the accuracy was better, up to ±0.5 °C. The affected data are flagged in the 365 data series (Flag 7: decreased accuracy; Table 3). The data quality improved greatly following the installation of a new data logger and multiplexer system (Campbell Scientific CR1000 data logger, AM16/32A multiplexer) in 2010 and the maximum offset at 0 °C during freeze-back was ±0.3 °C. reflectometer was used together with an SDMX50 coaxial multiplexers, custom made 20 cm TDR probes (Campbell Scientific CS605) connected to a Campbell Scientific CR10X data 380 logger between 2002 and 2010 and to a Campbell Scientific CR1000 data logger thereafter. All TDR probes were checked for offsets following the method described in Heimovaara and de Water (1993) and in Campbell Scientific's TDR100 manual (Campbell Scientific, 2015). The calibration delivered a probe offset of 0.085 (an apparent length value used to correct for the portion f the probe rods that is covered with epoxy) which was used instead of the value of 0.09 385 suggested by Campbell Scientific. The dielectric number ε (dimensionless) and the computed volumetric liquid water values θl (volume/volume) in frozen and unfrozen soil are provided as part of the time series data set. The calculation for volumetric liquid water content takes into account four phases of the soil medium (air, water, ice, and mineral) and uses the mixing model from Roth et al. (1990)

Soil dielectric number, volumetric liquid water content, and bulk electrical
The data are generally continuous and of high quality, and the absolute accuracy is estimated to be better than 5%. This is estimated from the maximum deviation of calculated volumetric liquid water content below and above the physical limits (between 0-1 or 0-100%). A probe located at 0.37 m depth beneath the polygon rim showed a shift of about 3% (up and down) in the volumetric liquid water content during the summers of 2009, 2013, and 2014, for which we 395 could not find any technical explanation. This shift is flagged in the data series (Flag 6: consistency; Table 3).
Time-domain reflectometry was also used to measure the bulk soil impedance, which is related to the soil's bulk electrical conductivity (BEC). These data were used to infer the electrical conductivity of soil water and solute transport over a twelve-month period in the active layer 400 of a permafrost soil (Boike et al., 2008a). The impedance can be determined from the attenuation of the electromagnetic wave traveling along the TDR probe after all multiple reflections have ceased and the signal has stabilized. The bulk conductivities were recorded hourly using the TDR setup described above in this section. Because no calibration was done, and the TDR probes were custom made to 20 cm, a probe constant (Kp) of 1 was used for BEC 405 waveform retrieval; Campbell Scientific suggests a Kp for the CS605 probes of 1.74.
Measurements of electrical conductivity and the dielectric number were affected by irregular spikes and possibly also by sensor drift similar to that in the soil temperature measurements and thus flagged until August 2015 (Flag 6). Data quality improved significantly after August 2015 when the Campbell Scientific coaxial SDMX50 multiplexers were exchanged for SDM8X50 410 and the electrical grounding system was improved. The dielectric numbers, computed volumetric liquid water contents, and soil bulk electrical conductivities can be found in the time series data set.

Ground heat flux
Two Hukseflux HFP01 heat flux plates were installed on 24 August 2002 and recorded ground The second PVC tube was used for comparison measurements at the same depths in the borehole. The differences between the calibrated reference thermometer (PT100) showed values between ±0.03 and ±0.33 °C (Appendix E, Table E1).
The data record shows that depth of zero annual amplitude (ZAA, where seasonal temperature 445 changes are negligible, ≤0.1 °C) is located below 20.75 m. At 26.75 m, temperatures fluctuate with a maximum of 0.05 °C. The annual mean temperatures between the start and end of the time series, as well as minimum and maximum temperatures, are displayed in Figure 3 ("trumpet curve"). The permafrost warms at all depths within this 10-year period, most pronounced at the surface. At 2.75 m, the mean annual temperature increased by 5.7 °C (from 450 -9.2 to -3.5°C), at 10.75 m by 2.8 °C (from -9.0 to -6.2 °C) and at ZAA of 20.75 m by 1.3 °C (from -9.1 to -7.7 °C).

Active layer thaw depth
Active layer thaw depth measurements have been carried out since 2002 at 150 points over a 27.5×18 m measurement grid (Boike et al., 2013, Figure 12;  To assist in the interpretation of active layer thickness data, surface elevation change measurements (subsidence measurements) have been collected since 2013 at three locations (two wet centers, one rim) using reference rods installed deep in the permafrost (Figure 1).

465
These measurements show that a net subsidence of about 15 cm occurred between 2013 and 2017 at the rim, and smaller subsidence (-1 cm and -3 cm) at the wet centers. A net subsidence of between -1.4 to -19.4 cm between 2013 and 2017 was reported by Antonova et al. (2018) for the Yedoma region of the Lena River Delta. Subsidence monitoring will in future be incorporated into the observational program on Samoylov Island so that active layer thaw 470 depths can be more accurately interpreted taking into account surface changes due to subsurface excess ice melt.

Data quality control
An overview of the periods of instrumentation and parameters is provided in Figure 4.
Quality control was carried out as outlined in Boike et al. (2018) for the data set compiled from 475 the Bayelva site, which is located on Spitsbergen. Quality control on observational data aimed to detect missing data and errors in the data, in order to provide the highest possible standard of accuracy. In addition to the automated processing, all data have been visually controlled and outliers have been manually detected, but it cannot be ruled out that there are still unreasonable values present which are not flagged accordingly. We differentiate between Level 0, Level 1, 480 and Level 2 data (Table 3). Level 0 are data with equal time steps (UTC), data gaps filled with NA and standardized into one file format. These data, as well as raw data, are stored internally at AWI and are not archived in PANGAEA. Level 1 data have undergone extensive qualitycontrol and are flagged with regards to equipment maintenance periods, physical plausibility, spike/constant value detection, and sensor drift (Table 3). Level 2 data are compiled for special 485 purposes and may include combinations of data series from multiple sensors and gap-filling.
Examples in this paper of Level 2 data are soil temperature and meteorological data (air temperature, humidity, wind speed, and net radiation) recorded between 1998-2002  that have been combined with a data set since 2002 into a single data series, in order to obtain a long term picture (documentation of source data is provided in the PANGAEA data 490 archives).
Nine types of quality control (flags) have been used (Table 3). Data are flagged to indicate where no data is available, or system errors, or to provide information on system maintenance or consistency checks based on physical limits, gradients, and plausibility.
Due to the failure of some sensors that cannot be retrieved for repair or re-calibration (e.g. 495 sensors installed in the ground), the initial accuracy and precision of the sensors may not always be maintained. In the case of soil temperature sensor accuracy can be estimated by analysis of temperatures relative to the fall zero-curtain effect, assuming that the soil water composition is similar from year to year. Our temperature data have been checked against the fall zero-curtain effect and information on any reduction in accuracy is flagged in the data set (Flag 7: decreased 500 accuracy; Table 3). These checks are essential if subtle warming trends are to be detected and interpreted. The suitability of flagged data therefore depends on what it is to be used for and the accuracy required.
The local differences between the sensor locations from 1998 and 2002 (even though less than 50 m meters apart), as well as differences between sensor types and accuracies, need to be 505 considered when interpreting longer term records. For example, relative air humidity data show marked differences between the earlier data set (1998)(1999) compared to the later data set (starting in 2002). Net radiation between 1998 and 2009 showed lower values during the summer periods compared to the summer periods between 2009 and 2017. One reason could be the change in sensor types: during the first period, a net radiation sensor was in place, 510 whereas during the second period a four component radiation sensor was used.

Summary and Outlook
The climate of the period between 1998 and 2017 can be characterized as follows: The average mean annual air temperature is -12.3 °C, with mean monthly temperature of the warmest month (July) recorded as 9.5 °C and for the coldest month (February) as -32.7 °C. The average annual Since the installation in 2006, permafrost has warmed by 1.3 °C at the zero annual amplitude depth at 20.75 m. Permafrost in the Arctic has been warming and the rate of warming at this borehole is one of the highest recorded (Biskaborn et al., 2018). Mean annual permafrost season's active layer thaw depth shows a marked interannual variation. Further analysis is required to disentangle the relationships between meteorological drivers, permafrost warming, and active layer thaw depths at this research site. The data sets described in, and distributed through, this paper provide a basis for analyzing this relationship at one particular research site 525 and a means of parameterizing earth system modelling over a long observational period. The newly collated data set will allow multi-year model validation and evaluation that includes the small-scale microtopographic effects of permafrost-affected polygonal ground. Landscape heterogeneity (such as, e.g., in soil moisture) is particularly poorly represented in earth system models and yet exerts a strong influence on the greenhouse gas balance (e.g. Kutzbach et al., 530 2004;Sachs et al., 2010). As such, this data set allows the distinction between microtopographic units (wet vs. dry) to be incorporated into modelling. This makes this an important dataset for modellers. We will continue to update these data sets for use in baseline studies, as well as to assist in identifying important processes and parameters through conceptual or numerical modeling.

Data availability
The data sets presented herein are freely available as a download from PANGAEA.
Permafrost temperature and active layer thaw depth data are also available through the Global Terrestrial Network for Permafrost (GTN-P) database (http://gtnpdatabase.org). All data are provided as ASCII files and are freely available through the following data provider and links:    (1998-2011Boike et al., 2013). Continuous data (light and dark colored data sets, e.g. wind speed and direction) are combined in the Level 2 product as one continuous data series for the period 1998-2017. Details of parameters for all sensors can be found in Table 2.
Note that the color bars describe the sensor installation period, but data might not be available 575 in the published data set due to sensor malfunction/failure. Note that the measuring period for the Vaisala HMP155A only started 17 September, 2017, which is why the bar appears very thin. Recording of all parameters is still continuing at present.  (1) derived as mean vegetation height within a radius of 3 cm -center: mean 5.4 cm/standard deviation 2.0 cm -rim: mean 4.6 cm/standard deviation 2.1 cm (2) derived as maximum vegetation height (99th percentile) within a radius of 3 cm -center: mean 11.7 cm/standard     Modified data compiled for special purposes such as combined data series from multiple sensors and gap-filled data 0 Good data All quality tests passed 1 No data Missing value 2 System error System failure led to corrupted data, e.g. due to power failure, sensors being removed from their proper location, broken or damaged sensors, or the data logger saving error codes 3 Maintenance Values influenced by the installation, calibration, and cleaning of sensors or programming of the data logger; information from field protocols of engineers 4 Physical limits Values outside the physically possible or likely limits

Appendix C: Calculation and correction of soil and meteorological parameters C1 Calculation of soil volumetric liquid water content using TDR
The apparent dielectric numbers were converted into liquid water content (θl) using the semi-630 empirical mixing model in Roth et al. (1990). Frozen soil was treated as a four-phase porous medium composed of a solid (soil) matrix and interconnected pore spaces filled with water, ice, and air.
The TDR method measures the ratio of apparent to physical probe rod length ( ) which is equal to the square root of the bulk dielectric number ( ).

635
The bulk dielectric number is then calculated from the volumetric fractions and the dielectric numbers of the four phases using A value of 0.5 was used for α.
It is not possible to distinguish between changes in the liquid water content and changes in the ice content with only one measured parameter (εb). Equation C1 was therefore rewritten in terms 640 of the total water content (θtot) and the porosity (Φ) as Note that Equation C2 assumes the densities of liquid and frozen water to be the same, which is clearly incorrect for free phases and probably also in the pore space of soils. However, the density ratio can be absorbed into the dielectric number εi, which we do below. The resulting fluctuation of εi is presumed to be small compared to other uncertainties. 645 We use to obtain the equation For temperatures above a threshold freezing temperature (T > Tf), all water is assumed to be unfrozen (θtot = θl ). Equation C5 then reduces to: For temperatures equal to or below the threshold freezing temperature (T ≤ Tf) it was assumed that the total water content (θtot) remained constant and only the ratio between volumetric liquid water content (θl) and volumetric ice content (θi) changed. This is a rather bold assumption as freezing can lead to high gradients of matric potential, as well as to moisture redistribution.
However, since the dielectric number of ice is much smaller than the dielectric number of liquid 655 water, the error in liquid water content measurements is still acceptable (which is not the case for ice content measurements). Under these assumptions we obtained the following equation for calculating the liquid water content of a four-phase mixture: The error of the volumetric water content measurements using TDR probes was estimated to be between 2 and 5%, which is in agreement with Boike and Roth (1997).

47
The availability of reliable temperature data is crucial in this approach. The liquid water content is first calculated for all times when the soil temperature was above the freezing threshold, using Equation C5. When the soil temperature was below the freezing threshold the water content immediately prior to the onset of freezing was determined and used as the total water content (θtot) for calculating the liquid water content during the frozen interval with Equation C7.

665
Since water in a porous medium does not necessarily freeze at 0 °C but at a temperature that depends on the soil type and water content, estimating the threshold temperature is a crucial part of this approach. If the freezing characteristic curve is known for the material then the threshold temperature can be determined from the soil volumetric liquid water content. To avoid interpretations of frequent freezing and thawing due to soil temperature measurement errors, 670 short-term temperature fluctuations were smoothed by calculating the mean of a moving window with an adjustable width. The smoothed temperatures were then used to trigger the switch from one equation to the other, rather than using the original temperature time series.
The porosity values for volumetric liquid water content calculations were obtained from laboratory measurements (Appendix F) and adjusted for probe location, if necessary 675

C2 Snow depth correction for air temperature
The acoustic distance sensor (Campbell Scientific SR50) measures the elapsed time between without requiring the presence of any personnel. A major disadvantage of using a common pressure transducer sensor to measure the water level is that such a device cannot withstand the long frozen arctic winter and is therefore not suitable for use when the presence of personnel is limited due to expedition schedules being restricted to summer period. A setup that can remain installed and withstand the cold winter temperatures therefore has a great advantage. 695 We apply vertically installed soil moisture probes to estimate water level, as described in Thomsen et al. (2000). Campbell Scientific CR200 data logger was connected to a Campbell Scientific CS625 probe (15 cm below the ground surface) to record the water level and two Campbell Scientific T109 sensors (1 cm and 6 cm below the surface) for temperature measurements. Since 2010 the setup has been connected to the main Campbell Scientific CR1000 logger of the meteorological 705 station and the CS625 probe was therefore exchanged for a Campbell Scientific CS616 probe, installed 11.5 cm below the ground surface. Due to a change in data loggers in the summer of 2010, we have two setups with minor differences in the measurement probes and their 50 installation depths, which is detailed below and visualized in Figure D1. The difference between the two water content reflectometers is the electrical output voltage, which had to be changed 710 in order to meet the requirements of the logger. A third T109 probe was also installed 3 cm below ground surface in 2010. This setup is still in operation. These temperature data are only used to distinguish between periods of frozen and unfrozen surface conditions. The unfrozen period, for which water levels were computed, was defined as the period for which soil temperatures at 6 cm below surface are > 0.4°C during spring, and > 0.1°C during fall. Below 715 these temperatures, no water level data are provided.
To obtain a better field calibration of the water content reflectometer a Schlumberger Mini-Diver pressure water level sensor was installed in a well in the same polygon for 68 days of the non-frozen vegetation period in 2016. Measurements obtained from the Diver were compensated for changes in air pressure using data from the meteorological station's barometric 720 pressure sensor (Vaisala PTB110).

D1 Calculation and correction of water level measurements
The measured output, signal period (SP) from the Campbell Scientific CS616 or CS625 probes were converted into the height of the water level above the sensor base (WL) using two polynomial functions derived from an empirical field experiment to determine the correlation between results from the CS616/CS625 probes and those from a Mini-Diver.

735
The two regressions represent different water level regimes (low and higher water levels) recorded by the CS616/CS625 sensor. The results of this experiment showed a low accuracy for very low water levels (1.5 cm or less above the sensor base) resulting in output periods of

53
Note that WL is given relative to the sensor base in the time series data and reported in meters.
To obtain water level relative to the ground surface (WLgs) from Level 1 data, the following calculation is suggested: for We compared the calculated WL with manual distance measurements taken in the field over the years (n = 12). The largest differences between TDR derived and manual measurements was 2 cm. This includes all measurement errors, such as sensor movement (probes are not anchored into the permafrost, they can potentially move with the seasonal heaving, subsiding  A thermistor chain with 24 temperature sensors (RBR thermistor chain with an XR-420 logger) was inserted into a close fitting PVC tube (4 cm inside, 5 cm outside diameter) and installed in 775 the borehole on 21 August 2006, down to a depth of 26.75 m ( Figure E2).
A second PVC tube with the same dimensions as the first tube was also inserted into the borehole to permit additional (geophysical and calibration) measurements to be made in the future. The remaining air space in the borehole was backfilled with dry sand. The outside metal pipe (used for drilling and to prevent inflow of water) which stands 0.5 m above ground surface, 780 was closed at the top and was covered with a wooden shield, which was renewed in 2015.
The accuracy of the temperature sensors of the thermistor chain is reported by RBR to be ±0.005 °C between -5 °C and 35 °C. However, direct comparison with a high precision reference PT100 temperature sensor (certified to be accurate to ±0.01 °C between -20 and 30 °C) at six different depths in the borehole between 9 and 17 August 2014 showed the accuracy 785 of the RBR XR-40 temperature sensors to be approximately ±0.03 °C at depths ≥8.75 m (Table   E1). The deviation increased with decreasing depth, e.g. between -7.75 m and -1.75 m the deviation was ±0.33 °C and at -0.75 m it was ±0.65 °C. This increase in deviation towards the surface may be because (a) the chain was installed in sand whereas the calibration thermometer was in air and could therefore possibly have been affected by air circulation, or (b) the 790 temperature gradient becomes steeper with decreasing depth below the surface and thus small differences between the measuring heights of the two sensors will have a larger impact on temperatures as the surface is approached. The offset of the reference thermometer at exactly 0 °C was 0.01 °C, and the average statistical accuracy (Uk=2) is given by the manufacturer as 0.1083 °C. During calibration in the borehole the temperature was given time to stabilize (i.e. 795 until the recorded temperature change was less than ±0.03 °C) before being recorded (Table   E1).
The data from these sensors have not been flagged as they are of high quality, but they may not provide an accurate reflection of the actual soil temperatures.

815
The metal pipe extends 0.5 m above ground surface. Note the differences in scale between above and below ground surface.   Table F1. Soil data from the BS-1 (polygon rim) and BS-3 (polygon center) soil pits, which 830 were sampled and had instruments installed in 2002. The location of the soil profiles is described in  and shown in Figure B8 and B9. Photos of the soil profiles can be seen in Figure F1 below. Grain size classification is according to Folk (1954)  according to DIN 19683 (1973). OC and N were determined following removal of inorganic carbon and dry combustion at 900 °C (DIN ISO 10694). The organic carbon density in bulk soil CDbulk (kg m -3 ) was calculated using the mass fraction of organic carbon in soil OC, the average dry bulk density ̅ bulk , and the following formula: The organic carbon content for each soil horizon SOCC (kg m -2 ) was calculated using the mass 845 fraction of organic carbon in soil OC, the average dry bulk density ̅ bulk , the horizon thickness, and the following formula:   The raw data set was filtered using a statistical outlier removal (SOR, Rusu and Cousins, 2011) to remove spatially isolated points as outliers from the point cloud, with the number of neighbors set to 10 and the standard deviation multiplier threshold to 1.0.

885
A digital terrain model (DTM) representing the ground surface elevation was derived from this pre-processed data set. To determine the ground surface elevation the 3D TLS points were first classified into ground and non-ground points. For this we used a minimum approach, classifying all points within a search radius of 2.5 cm that were at less than 5.0 cm vertical distance from the minimum point elevation as ground points. This vertical distance threshold is 890 included to take into account position uncertainties of the TLS acquisition. The ground points in the 3D TLS data set are subsequently rasterized into the final DTM (with a cell size of 5.0 cm) using a robust moving planes interpolation strategy (TU Wien, 2016).
For evaluation purposes the DTM was compared to 27 GNSS measurements of the ground surface that were obtained during the TLS data acquisition. The data sets were compared by 895 taking the difference between GNSS-based elevation measurements and the corresponding DTM pixel values. Statistical analysis of these differences in ground surface elevation yielded a mean difference of 3.7 cm, a median difference of 1.7 cm, and a standard deviation of 5.1 cm.
Differences were mainly within the accuracy ranges of TLS point cloud registration and GNSS positioning. Larger positive differences (> 2.0 cm) indicated an overestimation of ground 900 surface elevation in the TLS point cloud. Where dense, short vegetation is present an error is introduced to the estimated ground surface elevation as the laser beam does not hit the ground surface at every local area in the site. This is to be expected, particularly for larger distances from the scan positions as the incidence angle from the TLS instrument has a direct effect on the penetration depth of the laser beam (Marx et al., 2017).

905
Relative height above the ground surface was derived as vertical distance of TLS points to the ground surface. The DTM was used to calculate the vertical distance to the ground surface for every 3D point in the TLS point cloud. A raster of relative height values was generated using the 99th percentile of the relative height attribute per raster cell, with a cell size of 5.0 cm.
Furthermore, a raster of mean relative heights above ground surface was generated that could