L-Band Radiometry of Alpine Seasonal Snow Cover: 4 Years at the Davos-Laret Remote Sensing Field Laboratory

This study reports on four consecutive winter campaigns (2016–2020) at the “Davos-Laret Remote Sensing Field Laboratory” in the Swiss Alps to gain insight into the L-band microwave emission of ground covered with seasonal snow. Close-range L-band Brightness temperatures <inline-formula><tex-math notation="LaTeX">$T_\mathrm{{B}}^{p,\phi }(\theta)$</tex-math></inline-formula> were measured over the site scanning different observation nadir angles <inline-formula><tex-math notation="LaTeX">$\theta$</tex-math></inline-formula> and azimuth angles <inline-formula><tex-math notation="LaTeX">$\phi$</tex-math></inline-formula> at horizontal and vertical polarization <italic>p</italic> = {H,V}. State parameters (SPs) of the snowpack (e.g., height, density, and snow water equivalent) and the subnivean soil (permittivity, temperature) were measured quasi-simultaneously using in-situ sensors and sampling, as well as meteorological data. In each campaign, <inline-formula><tex-math notation="LaTeX">$T_\mathrm{{B}}^{p,\phi }(\theta)$</tex-math></inline-formula> were measured over a “natural area” and a “reflector area” with a metal mesh reflector laid on the ground before snow accumulation. The radiometer measurements over “reflector area” allowed to retrieve the time-series of Snow liquid Water-content <inline-formula><tex-math notation="LaTeX">$W_\mathrm{{S}}$</tex-math></inline-formula> and Snow liquid Water-Column <inline-formula><tex-math notation="LaTeX">${WC}_\mathrm{{S}}$</tex-math></inline-formula>, which are employed as “derived measurements” to support interpretation of <inline-formula><tex-math notation="LaTeX">$T_\mathrm{{B}}^{p,\phi }(\theta)$</tex-math></inline-formula> measured over “natural areas” during different winter phases. The detailed approach for the estimation of <inline-formula><tex-math notation="LaTeX">$W_\mathrm{{S}}$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">${WC}_\mathrm{{S}}$</tex-math></inline-formula> using L-band radiometer data is presented. The data and analyses in this article address the following major points: 1) determination of the characteristic features of measured <inline-formula><tex-math notation="LaTeX">$T_\mathrm{{B}}^{p,\phi }(\theta)$</tex-math></inline-formula> during different periods in each of the four winter campaigns; 2) effects of dry and wet snow precipitation on L-band radiometer data compared to corresponding simulations; 3) effect of removal and compression of the snowpack on <inline-formula><tex-math notation="LaTeX">$T_\mathrm{{B}}^{p,\phi }(\theta)$</tex-math></inline-formula>; 4) effects of spatial heterogeneity on brightness temperatures. Finally, the study is concluded with recommendations relevant for future close-range remote sensing campaigns.


I. INTRODUCTION
M ICROWAVE radiometry was first performed on controlled and characterized snow samples to support the development and validation of emission models [1], [2], [3], [4]. Gradually, longer measurement campaigns were carried out to study the effects of temporal variations of snow cover and subnivean ground states on microwave emission. For example, long-term radiometry in the frequency range 5-100 GHz was performed in Weissfluhjoch (at 2550 m.a.s.l. near Davos, Switzerland) to study microwave signatures of snow under various conditions and to develop algorithms for snow-type classification, snow-mapping, and snow liquid water-content detection [5], [6]. Another example is four sessions of activeand passive close-range remote sensing (RS) together with a detailed characterization of snow cover reported in [1] and [2]. Each session lasted between one to three days and the purpose of the study was the exploration of active and passive microwave responses to snow parameters, such as wetness.
In the 1980s and 1990s, a lot of research has been invested in low-frequency passive RS at the L-band (1-2 GHz), including the development of corresponding microwave radiometers (RMs) and radiative transfer models, development of retrieval methods to estimate geophysical parameters based on L-band brightness temperatures, and the conduction of dedicated RS experiments. Most of these activities were aimed at gaining information about soil moisture, vegetation properties, and ground freeze/thaw. Notable examples of early RM-system developments include the National Aeronautics and Space Administration's (NASAs) PushBroom Microwave Radiometer (PBMR) [7], and the Electronically Scanned Thinned Array Radiometer (ESTAR) [8]. In connection with early microwave RS campaigns in support of model development and validation, the long series of Radiobrightness and Energy Balance Experiments [9], [10], [11], [12], [13], [14], [15] was conducted. Results of the mentioned activities culminated in the famous book series "Microwave Remote Sensing-Active and Passive" [16], [17], [18], which is still considered as reference literature in the field of microwave RS. Furthermore, the entirety of these early works laid the ground for later spaceborne microwave missions aiming to estimate soil moisture at large spatial scales. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ Examples of such current operational missions are the European Space Agency's (ESAs) Soil Moisture and Ocean Salinity (SMOS) [19] and the NASAs Soil Moisture Active Passive (SMAP) [20].
However, the topic of L-band radiometry in application to snow-covered areas did not receive momentum until the 2000s. This was mainly because of the small magnitude of the interaction between L-band radiation and dry snow, resulting in a very large penetration depth (>100 m [5], [21]). In consequence, RS researchers concentrated on higher frequencies where interaction with snow is significantly stronger, and the scientific consensus established that L-band radiometry could not provide information on snow cover properties. This narrative was challenged with the development of the L-band Specific Microwave Emission Model of Layered Snow (LS-MEMLS) in 2014 [22]. LS-MEMLS simulations showed that, despite the contemporary understanding that dry snow is "invisible" at L-band, it is merely "transparent", and thus, it influences Brightness temperature T p B (θ) at different observation angles θ relative to nadir at horizontal and vertical polarization p = {H,V} via impedance matching and refraction.
With the launch of the L-band missions SMOS [19] and SMAP [20], long-term L-band radiometry of the Cryosphere became more interesting in the scientific community and led to the theoretical study [23] proposing an approach to retrieve Snow density and Ground permittivity (ρ S , G ) from L-band T p B (θ). Such developments laid ground for long-term towerbased L-band radiometry in multiple climate regions including the Finnish Meteorological Institute's Arctic Research Center (FMI-ARC) since 2009 [24], the Tibetan plateau since 2016 [25], and in Poland since 2017 [26].
The measurements at FMI-ARC experimentally confirmed the model prediction that dry snow affects T p B (θ) and that (ρ S , G ) can be retrieved from L-band RM T p B (θ) data. Accordingly, in Autumn 2016, the Davos-Laret Remote Sensing Field Laboratory [27] was established for the tower-based RS of seasonal snow cover. The Davos-Laret site was complementary to the other sites with long-term microwave RS in terms of climate and instrumentations; it was an alpine site at an altitude of 1420 m.a.s.l. and it included multifrequency active and passive RS instruments. In addition, the Davos-Laret site included artificially prepared areas (e.g., a reflective wire grid) to study specific radiation transport mechanisms (e.g., absorption of the snow) from the start.
The first winter Campaign (C1) was conducted in 2016/2017 and resulted in increased understanding of the effect of geophysical and instrumental noise on two-parameter retrievals (ρ S , G ) [28], and the retrieval of Snow liquid Water-content W S [29]. Additionally, sensitivities of L-band T p B (θ) and derived retrievals (ρ S , G ) to various snow-and ground-SPs were analyzed. The examples include the effect of W S and its distribution pattern across the snowpack [27] and the effect of subnivean layer permittivity and roughness [23]. Some of the synthetic sensitivity analysis previously available in the literature lacked the experimental pairs simply due to lack of observations. This included for instance the observations over unfrozen ground covered with snow.
We present results from four consecutive winter Campaigns, C1 to C4, for winter seasons between 2016 to 2020, which provide experimental validation to some of the previously theoretically inspected cases. Accordingly, first, we provide an overview of the measurements performed during C1 to C4. To do so, a description of the site together with its setup schematics for different campaigns is given in Section II. Section III-A provides the introduction to the employed L-band RMs and the methods for the computation of calibrated Brightness temperatures T p B (θ) and for quantification of non-thermal Radio Frequency Interference (RFI). Sections III-B and III-C focus on the in-situ soil, snow, and on-site meteorological measurements. Considering the large volume and variety of data, Section IV provides an overview of the measurements, which are available through this publication [30]. To help better understand the L-band emission from natural ground covered with snow, we use Snow liquid Water-content W S and thereon based Snow liquid Water-content W C S . The latter are considered as "derived measurements" because they are retrieved from L-band brightness temperature measured over FootPrint (FP) areas with a reflector beneath the snowpack. This approach was first applied in [29] using brightness temperatures measured during C1 at the Davos-Laret site. Later studies [31], [32], [33] utilized conceptually similar retrieval methods to estimate snow melt over the Greenland ice sheet. However, due to the central importance of W S and W C S derived for all four campaigns, their underlying retrieval approach is described in Sections IV. Section V consists of graphic overviews of the core measurements with a brief description of the main features of each campaign. Further information and measurements are included in the data package linked to [30].
Furthermore, four more detailed example analyses are provided to study the effect of precipitation (Section VII), snow removal and compression (Section VIII), spatial heterogeneities over the site (Section IX), and the intercomparison of angular patterns of T p B (θ) for all campaigns (Section VI). It is noteworthy that, in addition to their scientific value, these analyses are examples of how one can employ the data collected during these campaigns for various scientific investigations.

II. SITE DESCRIPTION
The measurements reported and used in this article were conducted at the Davos-Laret Remote Sensing Field Laboratory [27], henceforth called "Davos-Laret RS site" or "site." Situated at the Laret valley (46 • 50 43 N, 9 • 52 19 E) at an altitude of 1420 m a.s.l., this site was established in autumn 2016 for long-term close-range active and passive microwave RS of alpine snow. The site extends over ∼50 × 50 m; it is relatively flat with small undulations of less than a meter height surrounding all but the northeastern side. Fig. 1 shows an aerial image of the site highlighting its position with respect to the neighboring Lake-Schwarz, residential area, and the forested regions. The location of the microwave RM tower and multiple sets of FP areas over the site during the winter Campaigns 2016/2017 (C1), 2017/2018 (C2), and 2018/2019 (C3) are depicted in Fig. 1. The site is normally covered with grass, which was cut at the end of the autumn each year in preparation for the winter measurement campaigns. Depending on the weather conditions, ground freeze/thaw status, and the duration between the last grass harvest and snow appearance, the grass could grow before the onset of snow. Fig. 2 [29], the "reflector area" allows for the measurement of snowpack's self-emission in presence of snow liquid water content. As is illustrated in Fig. 2(b), arrangements of FPs observed during the fourth Campaign C4 were different with respect to C1-C3. During C4 in winter 2019/2020, the ESA's Wide-Band SCATterometer (WBSCAT) [35], [36], [37] was used in addition to the L-band RM to perform simultaneous active (L-to Ka-band) and passive (L-band) close-range microwave measurements of alpine snow cover [38]. In C4, a double-layer artificial area was designed and prepared. The first layer was composed of charcoal mixed with soil and on top of it was a surface layer of fine-grained sand. This "charcoal-sand area", shown in white in Fig. 2(b), was designed such that both WBSCAT and the L-band RM pointing at φ = 75 • and θ = {35 • , 40 • , 45 • } could observe it. The reason for this experimental setup was to decrease the backscattering from the Snow-Ground (S-G) interface and, thus, to study the backscattering of the snow with reduced blending of backscatter originating from the S-G interface. Additionally, a small reflector with fine mesh-grid cells (∼1 mm) was placed underneath WBSCAT along its boresight to allow for measurement of snow backscatter without any influence from the underlying ground. The photo at the center-right of Fig. 2(b) shows the "charcoal-sand area" and the "small reflector area" in C4. As   TABLE I  INFORMATION ON THE POSITIONS OF THE SOIL IN-SITU SENSORS DEPLOYED  ALONG TRANSECTS (T#) SHOWN IN FIG. 2 DURING THE CAMPAIGNS C1-C4  WITH INCREASING DISTANCE FROM THE RM TOWER ALONG THE T#, IDS OF  SMT-100 AND 5TE FORK SENSORS ARE INDICATED AS S# AND F#, RESPECTIVELY explained in the following, in-situ sensors were used to monitor properties of these artificial FP areas. Several Transects (T) are shown in Fig. 2 to depict the location of deployed in-situ sensors measuring dielectric permittivity G (real part) and temperature T G of the Ground. Two types of in-situ sensors were buried ∼ 5 cm below the soil surface: 1) SMT-100 and 2) 5TE fork sensors depicted in Fig. 2 with green circles and black crosses, respectively. A green circle with a black cross is used to indicate where an SMT-100 and a 5TE sensor were placed at the same location. Examples for this case are Transects 2 and 3 during Campaign 2 indicated with T2(C2) and T3(C2), respectively. In Fig. 2(b), green dots encircled by green circles are used to indicate the pairs of SMT-100 sensors over T1(C4) deployed at ∼15 cm and ∼3 cm below the surface to measure the charcoal-soil mix and sand layer properties, respectively. In addition to the sensors buried in the ground, during C1, sensors were installed on a wooden stick [orange circle indicated in Fig. 2(a)] to measure permittivity and temperature of the snow. The respective 5TE and the two SMT-100 sensors were installed 10-cm, and 15-, 50-cm above ground. Data measured by these three sensors are not reported here but included in the data package made available in [30]. Table I contains the sensor IDs for each of the Transects (T) in each Campaign (C) ordered based on increasing distance from the RM tower. In Table I, the IDs given for SMT-100 and 5TE Fork sensors begin with letters "S" and "F", respectively, followed by the sensors' identifying letter or number. Further information about the deployed in-situ sensors and their measurements is provided in Section III-B. It is noteworthy that in this article, we only show permittivity G measured by SMT-100 sensors. This is because SMT-100 directly output G while 5TE sensors include an internal transformation to output soil moisture.
The neighboring SCATerometer (SCAT) tower, depicted in Fig. 2, hosted the SCATterometers SnowSCAT [39] during C1, C2, and C3 and WBSCAT [40] during C3 and C4 [38]. The emitted signals from these scatterometers could cause RFI disturbing the brightness temperatures measured by the RM. Furthermore, the towers were made of metal parts and thus could reflect the RFI signal into the RMs' antenna whenever the RM would point (θ) are considered with care, and only used for specific case studies and to investigate the effect of spatial heterogeneities on the measured brightness temperatures.

III. EXPERIMENTAL SETUP
In addition to L-band radiometry, meteorological as well as in-situ snow and soil measurements were carried out during each campaign. Here, we provide a description of each of the experimental setup elements.

A. L-Band RMs
Throughout the four winter campaigns, two different microwave RMs were used to measure Brightness temperatures The Radio Frequency (RF) frontend implemented in ELBARA-II and OMRA is literally the same resulting in similar performances in terms of accuracy (∼ ±1K) and sensitivity (∼0.1 K). Both RMs split the antenna signals in two 11-MHz frequency channels ch = {1,2} to enable detection of RFI in the frequency domain. Furthermore, the same Picket-horn antenna [34] connected with low-loss coaxial Feed Cables (FC) was used. The aperture size of the antenna was ∼1.4 m and its Half-Power (−3 dB) Full Beamwidth (HPFB) was ∼ 12 • . Accordingly, the directional sensitivity of the shared Pickethorn antenna is strongly concentrated around its boreside, and sensitivities of side-and back lobes are less than ∼−14.3 dB relative to the boreside sensitivity. The exact dimensions and the experimental characterization of the Picket-horn directivity is shown in [41].
As shown at the top left in Fig. 2(a), ELBARA-II or OMRA was mounted on a scaffold equipped with an automatic elevation-azimuth tracker. This tracking system, controlled by the instrument computer of the respective RM, allowed for regular scanning of the site along multiple nadir and azimuth angles θ and φ, respectively (Fig. 2). Installation heights of the RMs were consistent in all campaigns, such that the phase center of the Picket-horn antenna was ∼8 m above ground. However, ELBARA-II and OMRA differ in 1) the internal calibration sources; 2) the temperature stabilization; 3) the data acquisition and the instrument computer: 1) ELBARA-II: ELBARA-II [41] used in C1 and C2 is a temperature stabilized L-band RM equipped with an internal matched resistive noise source of known physical and equivalent noise temperature T R , and with an Active Cold Source (ACS). Noise temperature T ACS of the ACS is determined by regular measurements toward the sky (φ = 90 • and θ = 140 • ) whose L-band brightness temperature T sky can be computed using existing models, for instance [42]. Calibrated T p,ch B are achieved from ELBARA-IIs raw data (voltages) measured toward FPs, and T R and T ACS associated with corresponding voltage readings. Here, we refrain from further explanation on the calibration of ELBARA-II, as this is already described in [27], for instance. In [43], it is shown that absolute uncertainty of RFI-free ELBARA-II measurements T p,ch B is ∼±1 K for the range of brightness temperatures expected from observed FPs.
2) OMRA: OMRA used in C3 and C4 is developed from a JÜLBARA [44] upgraded with a "Raspberry Pi" instrument computer to have digital data acquisition and interfaces to control the automatic elevation-azimuth tracker. JÜLBARA and, therefore, also OMRA took over the RF frontend from ELBARA-0 [45]. Similar to ELBARA-II, the RF Assembly (RFA) in OMRA (= JÜLBARAs RFA) is installed on a metal plate with high thermal inertia to minimize thermal gradients. However, unlike to ELBARA-II, the RFA in OMRA is not accurately temperature stabilized. Instead, the entire JÜLBARA RF frontend was enclosed in a weatherproof housing equipped with heating elements that switch ON automatically when internal temperature drops below ∼20 • C. This means that OMRAs RFA temperature variability in response to ambient temperature is at least limited. Without the additional enclosure, operation of the original JÜL-BARA RF frontend under harsh environmental conditions would have been a problem. For instance, the measure taken prevents RFA temperature from dropping below the dew point and, therefore, the formation of condensation. Furthermore, OMRA is equipped with matched resistive cold and hot noise sources (no Active Cold Source (ACS) as in ELBARA-II). Physical temperatures of the resistive cold and hot noise sources are stabilized to T cold ≈ 4 • C and T hot ≈ 44 • C, respectively. The accurate match of both resistive sources with the 50-Ω impedance of the RF frontend input ensures that physical temperatures T cold and T hot correspond with respective noise temperatures.
It is obvious that the mentioned technical differences between OMRA and the more recent ELBARA-II require different methods to achieve calibrated Brightness temperatures T p,ch B from the raw data (voltages) of the two RMs. Since the calibration method used with OMRA was newly developed, it is briefly outlined.
The original method to obtain calibrated T p,ch B from OMRA is the same as the one employed with ELBARA-0 [45], detailed in [46], [47], and [48], and further applied in [49]. That approach uses T cold and T hot , which are relatively close in magnitude compared to ELBARA-II, and thus relies on extrapolation to compute T p,ch B . Such an extrapolation and lack of characterization of temperature-dependent performance of the RFA can lead to inaccurate T p,ch B . In this work, we introduce an alternative approach that improves the calibration of T p,ch B . The first calibration step uses sky measurements performed at φ = 90 • and θ = 140 • . From each undistorted sky measurement j, four data-pairs (U p,ch sky,j , T sky,j ) are created corresponding to p = {H,V} and ch = {1,2}. Downwelling sky radiance T * sky,j at the antenna aperture (*) at recorded T air,j , θ = 140 • and the site altitude 1420 m a.s.l. is simulated with the model presented in [42]. Subsequently, thermal noise emitted by each of the two FC, connecting the horizontal and the vertical ports of the antenna, is added to achieve T sky,j at the corresponding OMRA input port Transmissivity t FC = 10 −L FC /10 of each FC is related to its transmission loss specified as L FC = 0.3 dB. FCs physical temperature is represented by measured air temperature T FC = T air,j . In addition to the now available data (U p,ch sky,j , T sky,j ), there exist further 2×2 data-pairs (U ch cold,j , T cold,j ) and (U ch hot,j , T hot,j ) from measuring the cold and hot resistive sources (at ch = {1,2}) kept at the physical temperatures T cold = 4 • C ≈ T cold,j and T hot = 44 • C ≈ T hot,j equivalent to their respective noise temperatures.
The linear model below is adapted to the data-pairs (U p,ch sky,j , T sky,j ) and (U p,ch cold,j , T cold,j ), (U p,ch hot,j , T hot,j ) resulting in the optimal fit-parameters A p,ch j and B p,ch j for each sky measurement j The values of these 2×4 fit parameters change with ambient temperature because OMRAs RF frontend is not accurately temperature stabilized. Instead, it is built from JÜLBARAs RF frontend enclosed in an additional housing, as discussed above. However, for each measurement j, performed at T air,j , temperature T RFA,j is measured at the thermally inert metal plate carrying temperature-sensitive RFA components (amplifiers and attenuators, for instance). This ensures that temperatures of these components are very similar and well represented by measured T RFA,j . With these considerations, it is clear that the available 2×4 datasets (T RFA,j , A p,ch j ) and (T RFA,j , B p,ch j ) specify OMRAs residual noise and gain, respectively, at recorded T RFA,j and p = {H,V} and ch = {1,2}.
The following second-order polynomials, fit to each of the datasets (T RFA,j , A p,ch j ) and (T RFA,j , B p,ch j ) derived from sky measurements, yield the 2×4 models that represent OMRAs residual noise and gain at temperature T RFA at p = {H,V} and ch = {1,2} With the models A p,ch (T RFA ) and B p,ch (T RFA ), noise temperatures at the input ports (*) of OMRA are computed as: FP are raw data measured when OMRA is pointing toward the FP of interest and RFA temperature T RFA is measured simultaneously.
Finally, in analogy to (1), calibrated Brightness temperatures T p,ch B at the antenna aperture (also called "antenna temperature") is computed from T * p,ch B at the OMRAs inputs by adding the thermal noise of the FC As above, the physical temperature of the FC is considered as From (3), (4), and (5) it is clear that the accuracy of OMRAs calibration depends on the uncertainties ΔA p,ch and ΔB p,ch . The latter are estimated as the median (med) of the absolute values of associated j residuals (res) Ares p,ch j and Bres p,ch j for p={H,V} and ch={1,2} The result is ΔA p,ch ≈ 2 K and ΔB p,ch ≈ 0.5 KV −1 leading to uncertainties ΔT p,ch B ≈ 2 K in OMRAs calibrated Brightness temperatures T * p,ch B at the input ports computed with (5) 3) RFI: As described above, raw data (voltages) are used to compute the calibrated T p,ch B for both ELBARA-II and OMRA. Measurement at two frequency channels ch={1,2} within the protected part of the L-band enables detection of narrow-band RFI. A measurement, which is distorted by narrow-band RFI, is identified when channel difference ΔT p,ch | exceeds a certain threshold. For ELBARA-II and OMRA, measurements exhibiting ΔT p,ch B ≥ 1.5 K and ΔT p,ch B ≥ 5 K, respectively, are excluded from the data shown here but are still included in the data package [30]. This data include ΔT p,ch B values.
An additional method used here to identify non-thermal RFI is the Probability Density Function method (PDF-method) developed and described in [27]. Its concept consists of the comparison between the PDF of raw-data voltage samples measured (at the sampling frequency of 800 Hz and a 3-s integration time) and the Gaussian distribution expected for samples associated with undisturbed, purely thermal microwave radiation. Compared to conventional "normality" tests (kurtosis and skewness [50]), the "PDF-method" offers significant advantages: First, it is more responsive to certain types of time-domain RFI. Second, it allows quantifying the extent ΔT p B,non−thermal to which T p,ch B is distorted by non-thermal radiance. In the data shown here, the measurements with ΔT p B,non−thermal ≥ 1.5 K are omitted.

B. In-Situ Soil and Snow Measurements
As explained in Section II (Table I and Fig. 2) in-situ SMT-100 and 5TE fork sensors were used in each campaign. The SMT-100 sensors [51] measure temperature and real dielectric permittivity at f ≈ 100 MHz. The permittivity is used as a proxy to also output volumetric water-content. Additionally, five 5TE sensors [52], [53] were used to measure temperature, volumetric soil water-content, and electrical conductivity. The sensor measurements were conducted using a DT80 data logger every 5 min. There exist some data gaps in the recorded data due to technical problems with the data logger during some campaigns. More information is provided in Section V.
Furthermore, weekly manual snow characterization was carried out on the site to keep track of snow profile variations. In the case of heavy snow falls, snow characterization was conducted immediately afterward to capture resulting abrupt changes. Snow density ρ S was measured with a 3-cm density cutter. Additionally, Snow Water Equivalent (SWE) and Snow temperature T S were recorded next to Near Infra-Red (NIR) images of the profile's cross section. The latter helps identify snow layering and internal macrostructural features, which can be relevant in the analysis of L-band brightness temperatures. Photos taken by several webcams provide another useful source of information for the qualitative assessment of snowpack conditions. Time-lapse videos generated from these photos are part of the data package made available [30].

C. Automatic Weather Station
During the winter campaigns, several meteorological, snow, and ground parameters were recorded hourly by an on-site Automatic Weather Station (AWS). These parameters include air temperature T air at 2 m, relative humidity, melted equivalent Precipitation Prec. (rain or snow), incoming and reflected shortwave radiation, Snow height h S and Snow Surface temperature T SS , wind speed and direction, as well as SWE. In this article, we report and use T air , h S , and Prec., which are available throughout the four Campaigns. The AWS used during C1 and C2 did not include a snow balance. This AWS was upgraded for the following Campaigns C3 and C4. Accordingly, hourly AWS recordings of SWE were available only during C3 and C4. Nevertheless, SWE derived from weekly snow characterization is available for all four campaigns. Table II provides a summary of important dates and main characteristics of all Campaigns C1 to C4. Most notably the length of campaigns increased from 109 days in C1 to 223 days in C4. However, C4 was the only campaign in which the measurements stretched in time from before snow cover to after complete melt. In C1 and C3, at least one snow-free period exists while C2 began and ended in presence of snow cover. Finally, it is noteworthy that both situations of unfrozen ( G 5) and  [27] occurred within C1 to C4.

IV. DATASETS
The presented data comprise six categories through all four campaigns. These are described below, with detailed explanations regarding the Snow liquid Water-content W S retrieved from T p R (θ) of the "Reflector area." The first category is time-series of Brightness temperatures T p,φ B (θ) at polarization p={H,V} measured for multiple azimuth angles φ and nadir angles θ = {30 • , 35 • , . . . ,60 • } in 5 • steps (Fig. 2). We focus here on measurements along φ = 90 • for C1-C3 and φ = {75 • , 125 • } for C4. As mentioned in Section III-A3, RFI corrupted T p,φ B (θ) are filtered and excluded from the analysis. The second set is Precipitation (Prec.) measured as rain or melted snow. To qualitatively assess the wetness of the Precipitation, we distinguish between Prec. measured at T air < 0 • C and T air ≥ 0 • C. The third and fourth sets of data are recorded Snow height h S and air temperature T air . The fifth set is the in-situ measured Ground permittivity G averaged over the sensors deployed on the corresponding Transects (T) during a Campaign (C) indicated in Fig. 2, with sensors IDs provided in Table I. The sixth dataset is a time-series of volumetric liquid Watercontent W S of Snow, and Snow liquid Water-Column W C S = W S ·h S . As outlined in [29] W S is retrieved from the multiangle, measurement-based brightness temperatures T p R (θ) emitted by solely the Reflector area (R) and, therefore, understood as a "derived measurement." The latter help to better understand responses of L-band T p,φ B (θ) measured over "natural areas" to changing S-G conditions. The usefulness of the retrieved W S to explore active RS signature's response to snow melt and wetness variations is demonstrated in [38]. In the following, we outline the procedure for obtaining W S and explain the campaign-specific considerations: As explained in Section II, the RM antenna was pointed toward the "reflector area" (φ = 145 • in C1-C3 and φ = 180 • in C4) to measure the snowpack's own emission at L-band. However, due to inevitable measurement nonidealities such as antenna's finite directivity, the RM would measure emission originating from both the artificially prepared "reflector area" and the surrounding "natural area" with no reflector laid on the ground before snow deposition. Therefore, the first step toward obtaining W S is to compute and deduct the partial contribution of the surrounding "natural area" from the total recorded measurement. This leads to the emission T p R (θ) exclusively from the snowpack over the "Reflector area." Therefore, one must compute the weighting-factors Assuming the reflectivity of the metal mesh reflector placed on the ground as one [43], [54], this yields As sketched in Fig. 2(a), for C1-C3 the azimuth angle of the "Reflector area" was φ R = 145 • and the azimuth angle of the chosen "natural area" was φ N = 90 • . For C4, the respective azimuth angles are φ R = 180 • and φ N = 125 • , as shown in Fig. 2(b). For snow-free or dry-snow conditions T p R (θ) is assumed as the reflected downwelling sky radiance T sky (θ) simulated with the model presented in [42]. Accordingly, μ p,θ R can be computed from (10) using T p,φ R B (θ) and T p,φ N B (θ) measured during "μ-calibration periods" defined by snow-free or dry-snow conditions.
The third step is the actual retrieval of Snow liquid Watercontent W S from T p R (θ). A W S -retrieval is achieved by minimizing the sum of squared differences between simulated T p R,sim (θ) and measurement-based T p R (θ). Accordingly, the W S is retrieved by minimizing the following cost function: The single-layer configuration of the L-band-Specific Microwave Emission Model of Layered Snowpacks "LS-MEMLS" [22] is used to simulate T p R,sim (θ, W S ) as a function of the retrieval parameter W S and the auxiliary model parameters: s p S , T S , ρ S , h S , T air , and s p G = 1 representing the reflectivity of the metal-mesh reflector. The respective modeling equations are provided in [27]. Snow temperature is considered consistently as its maximum possible value T S = 0 • C, because for T S < 0 • C, the snowpack is transparent at L-band, and therefore not emitting. In-situ measured Snow density ρ S is used for the computation of W C S . Air temperature T air used to simulate T sky is either taken from the PT-100 sensor attached to the ELBARA-II RM (during C1 and C2), or from the AWS (during C3 and C4).
It is noteworthy that W S retrieved with the aforementioned approach is only qualitatively accurate, mostly because of the simplicity of the single-layer microwave emission model used to simulate T p R,sim (θ, W S ). However, retrieved W S is useful to reliably classify the snowpacks main wetness states, such as dry, moist, and wet. As mentioned earlier, this information is helpful to explore responses of RS signatures to snow melt and wetness variations. The time-series of T p R (θ) and weighting-factors μ p,θ R are included in the data package made available.

V. OVERVIEW OF MEASUREMENTS
Each of the four conducted winter campaigns lasted several months and covered multiple phases of S-G system states. To simplify the inspection and analysis of L-band Brightness temperatures T p,φ B (θ) response to various SP, we define and consistently use a set of characteristic "phases", which are as follows.
"snow-free": At the beginning or the end of Campaigns, except for C2, measurements were performed in the absence of snow cover. The defining condition for "snow-free" phase is h S = 0 m.
"early-winter": Sometimes the beginning of a stable snow cover period was unclear leading to a period of h S ≤ 0.4 m where snow height fluctuated between zero and approximately 0.4 m. Air temperature in "early-winter" often rose above 0 • C during the day and precipitation could be both snow and rain. With these varying snow cover conditions and yet nonnegligible effect of thin snow on L-band measurements T p,φ B (θ), we define "early-winter" as a separate phase. "snow accumulation": Also called "acc." in the following graphs, is the period over which Snow height at least doubled from the maximum h S = 0.4 m in "early-winter" to >0.8 m. This phase is characterized by gradually decreasing T air and mainly snow precipitation. The duration of the "snow accumulation" phase varies greatly among different winters. During the first Campaigns C1 (winter 2016/2017), it was very short and, therefore, not shown in the corresponding Fig. 3. During winter 2017/2018 (C2), the "snow accumulation" phase lasted for more than a month, as shown in Fig. 4. During the winters 2018/2019 (C3) and 2019/2020 (C4), the respective "snow accumulation" phase lasted around 10 and 14 days as is seen in Figs. 5 and 6, respectively. "mid-winter": This period is characterized by the stable presence of a generally dry (W S = 0 m 3 m −3 ) snowpack reaching its maximum seasonal h S with T air < 0 • C most of the time. "early-spring": Increased thermal energy input brought by increasing T air (via convection and conduction) and absorption of sunlight during longer days eventually overcomes ice latent heat and causes partial melt in snow. Accordingly, this phase is characterized by daily fluctuations of W S without rapid decrease in h S . "isothermal": With T air > 0 • C during daytime, the snowpack reaches isothermal state at the maximum possible temperature T S = 0 • C. The key feature of this phase is that the snowpack does not refreeze overnight, rather due to very high values of W S , it becomes opaque at L-band. This phase can take several weeks accompanied by snow and rain precipitations. Liquid water discharge from the snowpack can also influence the underlying Ground permittivity G .
We acknowledge that the parameter thresholds given in our definitions of different winter "phases" are not strict, rather approximate indicators of the S-G system state. In the following sections, we present  Fig. 3(a) is measured over the "natural area" at φ = 90 • as indicated in Fig. 2(a). As shown in Fig. 3(f), average Ground permittivity decreased from G > 14 at the beginning of C1 to G ≈ 6 by 3 January 2017. This is because a prolonged "snow-free" phase extending to January 2017 accompanied by low T air resulted in frozen ground condition at least to the depth of ∼5 cm of the installed in-situ sensors used to measure G . In absence of snow cover the ground was exposed to changing air temperature, incoming radiation, and rain. Therefore, G fluctuated daily and responded to major changes such as increased soil moisture caused by rain on 26 December 2016 as apparent from Fig. 3 [22], and the following theoretical study [23] that laid the ground for snow retrievals based on L-band radiometry. It concluded that: "The fact that dry snow is almost fully transparent at wavelengths around 21 cm does not imply that dry snow can be treated as "invisible" for passive L-band measurements." From 31 January 2017, the "early-spring" phase starts with W C S first limited to <2 mm and later fluctuating daily between 8 and 20 February 2017. The "isothermal" phase started on 22 February and continued until the end of the campaign with increasing W C S such that the snowpack was consistently wet and mostly opaque at L-band. The liquid water discharge from the snowpack simultaneously increased the moisture of the subnivean Ground and, therefore, increased its permittivity [ Fig. 3(f)] from G ≈ 6 to G ≈ 14. Fig. 4 provides an overview of the measurements conducted during C2. This campaign was the only one with no L-band radiometry in "snow-free" phase, as is apparent from h S > 0 m shown in Fig. 4(d). Therefore, C2 starts on 17 November 2017 with "early-winter" phase followed by "snow accumulation" phase between 15 December 2017 to 12 January 2018. Due to  Fig. 4(d)]. Accordingly, Snow liquid Water-Column W C S has been reduced to an absolute minimum [ Fig. 4(c)], implying that the snowpack has become highly transparent. Accordingly, upwelling emission from the unfrozen ground below the snowpack [ Fig. 4  Measured G in Fig. 4(f) shows 1) the Ground was unfrozen during C2, and 2) even a thin snowpack is sufficient to thermally insulate the subnivean layer.

B. Second Campaign (C2, 2017/2018)
"Mid-winter" phase stretched from 12 January 2018 to 9 March 2018 where average daily T air was below 0 • C and W C S was mostly 0 mm. The only exception to the latter is 16 and 17 February 2018 where T air > 0 • C accompanied by precipitation [in the form of rain or moist snow marked red in Fig. 4(b)] caused increased Snow liquid Water-content. Careful inspection of W C S [Fig. 4(c)] reveals toothlike fluctuations of Snow liquid Water-Column between 27 and 31 January 2018 in response to daily T air > 0 • C. The L-band brightness temperatures' response to these marginal W C S fluctuations is best visible in The campaign's last phase was "isothermal" which consisted of positive average daily T air reaching > 12 • C, increasingly higher W C S , and decaying Snow height h S . We have not identified an "early-spring" phase in C2 simply because virtually immediately after the "mid-winter" phase the snowpack became consistently wet and, therefore, isothermal at 0 • C. From 3 April 2018, despite the presence of snow cover, G in Fig. 4(f) seems highly correlated primarily with daily T air fluctuations such that the peak of G takes place between three to five hours after maximum T air in the afternoon. This is due to extra liquid water discharged from the snowpack during the accelerated melt in the afternoon.  Table II and in Fig. 5, C3 started on 13 December 2018 with an already existing snow cover of h S ≈ 0.4 m. During the "snow accumulation" and "mid-winter" phases, the snowpack grew thicker while average T air dropped below 0 • C. As shown in Fig. 5(c), the snowpack was nearly consistently dry during "mid-winter" phase. In response to consistently dry snow condition, T p,90 • B (θ) became stable and free of strong fluctuations. The "early-spring" phase started on 19 February 2019 with daily snowpack melt/refreeze cycles. The "isothermal" phase was exceptionally long in that it lasted almost three months from 26 February 2019 to 19 May 2019. According to Fig. 5(c), during this phase increased W C S implies opaqueness of the snowpack at L-band. However, constant snow melt was partially compensated for by snow falls [Fig. 5(b)], during the "isothermal" phase whose signatures are best visible in local peaks of h S in Fig. 5(d Despite the data gaps in G due to technical malfunctioning of the DT80 data logger, Fig. 5(f) shows that the subnivean ground was unfrozen during the entire campaign with an average G ≈ 21 until the end of March 2019. With the melt water from the snowpack percolating down to the ground, Ground permittivity increased to G > 26. Between 18 and 26 April 2019, G fluctuated daily, which was the result of accelerated snow melt during the peak air temperatures.
It is noteworthy that in response to the varying W C S , Brightness temperatures T p,90 • B (θ) fluctuate. However, in "isothermal" phase with consistent W C S > 0 mm such a one-to-one traceability is missing. The interpretation of T p,90 • B (θ) time-series during the "isothermal" phase is complicated because the permanent presence of increasing snow liquid-water and its changing distribution across the snow profile causes partial or complete opaqueness at L-band. Nevertheless, the response of T p,90 • B (θ) to some major changes in the S-G system can be identified. Fig. 6 provides an overview of the measurements conducted during C4. This last Campaigns was the most comprehensive of all in that it was both the longest and encompassed most of the winter phases defined. Fig. 6(a) and (b) shows the time-series of T p,φ B (θ) measured over "charcoal-sand area" at φ = 75 • and over the "natural area" at φ = 125 • , respectively. In Fig. 6(a), only T p,75 • B (θ) at θ = 30 • and θ = 40 • are shown because as illustrated in Fig. 2(b), the "charcoal-sand area" was designed to fully cover L-band RMs −9 dB FPs associated with 30 • θ ≤ 45 • . It is evident from Fig. 6, especially during "snow-free" and "mid-winter" phases, polarization differences in brightness temperatures are more pronounced for T p,75 • B (θ) over the "charcoal-sand area" than for T p,125 • B (θ) measured over the "natural area." This is best visible during the "snow-free" phase at the beginning of the campaign where for a given nadir angle θ differences T V,75 (θ) over the "natural area" is ∼ 30 K [ Fig. 6(b)]. This is a direct effect of impedance matching and lower surface roughness provided by the "charcoal-sand area" compared to the "natural area." C1 and C4 are similar in that they both start with the "snowfree" phase. However, as seen in Fig. 6(f), air temperature was virtually constantly above T air = 0 • C, which prevented the ground from freezing. Correspondingly, mean Ground permittivities of "Natural area" (N) [measured with sensors along T2(C4) and T3(C4) in Fig. 2] are G,N > 20 [ Fig. 6(g)] and, therefore, prove the unfrozen state of ground. In contrast, mean permittivities G,CS within the "Charcoal-Sand area" (CS) are considerably lower, sometimes G,CS < 5. However, this does not mean that the "charcoal-sand area" froze at any time; instead, the generally low G,CS are due to the low water-holding capacity of the surface layer of this area.

D. Fourth Campaign (C4, 2019/2020)
Both permittivity records, G,N and G,CS , fluctuate in response to precipitations such as on 16, 21, and 29 October 2019 during the "snow-free" period before snow onset. During the "early winter" phase, in-situ G,N and G,CS respond to liquid water brought by major precipitations at T air > 0 • C (for example, on 15 November 2019) and increased W C S (for example, on 23-25 November 2019). Fig. 6(g) shows that from approximately 30 November 2019 ground permittivities of the "natural area" and the "charcoal-sand area" ceased their fluctuations and by the middle of the "snow accumulation" phase they had settled to G,N ≈ 21.8 and G,CS ≈ 5.8, respectively. Despite the gap in the ground permittivities, the next available records on 23 January 2020 in "mid-winter" phase confirm that in presence of thermally insulating snow, subnivean permittivity was stable. The time-series of W C S in Fig. 6(d) shows that the "mid-winter" phase is immediately followed by "isothermal" phase with W C S rapidly increasing to high values making the snowpack opaque even for L-band observations. According to Fig. 6(e), the "isothermal" phase ended with a quick melt enforced with consistent T air > 0 • C such that between 10 and 18 April 2020, h S dropped from ∼0. 45 Fig. 7. Both photos were automatically taken at 10:38 CET after the end of the "isothermal" phase (19 April 2020) and at the end of campaign (7 May 2020). Since the "snow-free" phase after snow clearance lasted for ∼20 days, the effect of growing vegetation on T p,125 • B (θ) is noticeable. Fig. 7(b) shows that by the end of the campaign even the "reflector area" at φ = 180 • [ Fig. 2(b)] is covered with green grass, which grew through the metal-mesh grid. Vegetation did not grow over the artificially created "charcoal-sand area"; therefore, T p,75 • B (θ) does not exhibit an increasing trend toward the end of C4.

VI. MULTIYEAR OVERVIEW OF ANGULAR
DEPENDENCE OF T p,φ B (θ) The time-series of measured L-band brightness temperatures for each campaign was presented and discussed in Section V.
Alternatively, Fig. 8 provides a summary overview of T p,φ B (θ) for all four campaigns as a function of nadir angle θ. The graphs in Fig. 8 are based on T p,φ B (θ) measured for azimuth angles φ = {90 • , 90 • , 90 • , 125 • } along "natural areas" in Campaigns {C1,C2,C3,C4} (Fig. 2). Each column in Fig. 8 refers to one Campaign (C) and the title above the top graph in each column indicates the Campaign number C#, ground freeze/thaw condition (frozen or unfrozen), and the azimuth angles φ of FPs over "natural area" where radiometry was conducted. The first, second, and third rows in Fig. 8 refer to "snow-free" (h S ≈ 0 m), "dry-snow" (h S > 0.05 m and 0 mm < WC S < 0.2 mm), and "wet-snow" (h S > 0.05 m and W C S > 0.5 mm) conditions, respectively. Accordingly, for each nadir angle θ in each panel, the average T p,φ B (θ) for p={H,V} is computed and plotted. The vertical bars at each data point indicate the standard deviation of T p,φ B (θ) and are indicative of the variability of brightness temperatures for a given scenario. For example, since the only condition for "snow-free" graphs is the absence of snow cover (h S ≈ 0 m), T p,φ B (θ) can still refer to nighttime and daytime conditions as well as different meteorological conditions such as precipitation which may change FP emissivity as discussed in Section VII. In the following, we highlight a few key observations based on the angular behavior of T p,φ B (θ) shown in Fig. 8. The "snow-free" graphs in Fig. 8 show that when the ground is frozen [C1 in Fig. 8(a) is larger than what is measured for unfrozen ground conditions [C3 and C4 in Fig. 8(g) and (j)]. Due to smaller permittivities G of frozen Ground compared to G of moist, unfrozen ground this was expected and confirmed by LS-MEMLS simulations [23]. In Campaigns C1, C3, and C4 where "snow-free" conditions were available, the comparison between "snow-free" and "dry-snow" graphs [ Fig. 8(a) and (b) for C1, Fig. 8(g) and (h) for C3, and Fig. 8(j) and (k) for C4] show that dry snow cover increases T p,φ B (θ). Impedance matching caused by the dry snowpack is the reason for higher brightness temperatures during "dry snow" periods than during "snowfree" periods. As anticipated by LS-MEMLS simulations, this increasing effect is stronger at p=H than at p=V.
Comparison between the "dry-snow" and "wet-snow" graphs in each campaign indicates increased brightness temperature To understand this observation, we first reiterate that dry snow does not have noticeable self-emission at L-band; therefore, dry snow cover influences the observed brightness temperatures via impedance matching and refraction effects tuned with snow density. However, in presence of liquid water, snowpack starts to emit at L-band and even small variations of Snow liquid Water-content (for instance W S < 1%) can significantly change the snow cover emissivity. As shown in [27], even the W S distribution pattern across a snow profile can influence the observed Fig. 8(c), (f), (i), and (l) includes measurements in various times and thus a wide range of snow cover properties. As a result, the variability of T p,φ B (θ) is larger for wet snow cover conditions. Fig. 8 and its brief analysis in this section offer an example of how the measurements from C1-C4 in Davos-Laret site can be examined from multiple points of view. The inspection of data in the time domain (Section V) and the angular domain (Section VI) presents different observations which are possible only by changing the analysis domain.

A. Modeled Case
Precipitation can generally alter the existing snow cover properties in two main forms: First, dry snow precipitation at T air < 0 • C increases snow height h S and creates a layer of fresh snow atop the snowpack with noticeably lower density ρ S than the rest of the snow profile. Since dry snow is virtually lossless at L-band (penetration depth is > 100 m [5], [21]), the comparably small change in h S does not affect Brightness temperatures T p B (θ) of a ground covered with dry snow. However, as shown in [22], dry snow influences T p B (θ) via refraction and impedance matching, which is most important at the lower and the upper bound of the snowpack. The second form of snowpack alteration due to precipitation is caused by rain or wet snow precipitation when T air ≥ 0 • C. Rain or wet snowfall introduces increased Snow liquid Water-content W S on the surface of the snowpack. Light rain may initially increase W S in the snowpack surface-layer while stronger or long-lasting rain may infiltrate the snowpack. However, it is obvious that precipitation in any form (dry-or wet snow, or rain) changes the microwave emissivity of the snow-covered ground. Effects of W S on T p B (θ) were investigated for multiple scenarios of wetness distribution across the snow profile [27]. In a wet precipitation scenario, if it happens to be wet snow, we neglect the increased h S and possibly lower ρ S atop the snowpack because the major influential parameter is propagation losses introduced by W S > 0 m 3 m −3 .
First, we present the comparative response of T p B,sim (θ) to dry and wet snow precipitation simulated (sim) with LS-MEMLS and shown in Fig. 9. The bold gray lines in Fig. 9 consistently refer to the situation before precipitation (dry uniform snowpack). The red and green dotted lines, in panel (a) and (b), respectively, refer to simulations of brightness temperature for a snowpack  Fig. 9]. The nadir angle range 30 • ≤ θ ≤ 60 • of L-band brightness temperatures measured at the Davos-Laret site is indicated in the right parts of Fig. 9(a) and (b).
To simulate the impact of dry snow precipitation on T p B,sim (θ), we use Fermi-Dirac function (Section 5.2 in [55]) to create a Snow density profile ρ S (h) with decreasing density in the upper part of the snowpack [ Fig. 9(a)]. Concretely, we have chosen a snowpack of total height h S = 1.5 m with densities ρ S (0) = 300 kg m −3 and ρ S (h S ) = 100 kg m −3 at the bottom and at the top, respectively. Transition height h * defined by ρ S (h * ) = (ρ S (0) + ρ S (h S ))/2 = 200 kg m −3 is assumed at height h * = 0.9 · h S = 1.35 m. The profile W S (h) of Snow liquid Water-content [ Fig. 9(a)] is considered as W S (h) = 0 m 3 m −3 to represent dry snow precipitation. Furthermore, ground is assumed to be unfrozen ( G = 20). Simulated T p B,sim (θ) shown in the right panel of Fig. 9(a) demonstrate that the light snow atop the snowpack causes an observation angle-dependent increase in T p B,sim (θ) (red) compared to T p B,sim (θ) (bold gray) simulated for the situation before dry snow precipitation. At horizontal polarization p=H, this increase is stronger at larger nadir angles in the range of 20 • ≤ θ ≤ 85 • where T H B,sim (θ) (red) is increased by ∼ 2 K and ∼ 24 K, respectively. At p=V, this increase is only distinguishable at 60 • ≤ θ ≤ 85 • . The reason why light dry snow atop the snowpack increases brightness temperature is the decreased reflectivity at the snowpack surface, which corresponds to increased emissivity and, therefore, higher T p B,sim (θ). The simulated impact on T p B,sim (θ) induced by rain or wet/moist snow precipitation is demonstrated with Fig. 9 (60 • ) increases by ∼ 5.4 K from the start of precipitation until its peak intensity. This effect is increasingly limited for lower θ relative to nadir (steeper observation directions). However, as predicted by the LS-MEMLS simulations in Section VII-A for 30 • ≤ θ ≤ 60 • , the dry snow precipitation effect is negligible for p=V.
It is noteworthy that dry snow precipitation effect is most notable when there is a heavy snowfall, which quickly forms a distinct lower density layer on top. When the precipitation is prolonged, such as on 26 January 2019 in Fig. 10(b), the effect on T p,φ B (θ), while present, is hardly noticeable. One reason is the gravimetric snowpack compression, which at slow precipitation rates avoids the formation of a top low-density layer.   Fig. 11(d) shows that none of the presented precipitations resulted in noticeable change in h S implying that they were either rain on snow or wet snow with no tangible effect on snow height. Fig. 11(b) and (c) show that wet precipitation in case 1 causes a short-term detectable increase in Snow liquid-Water Column W C S [Fig. 11(c)], which falls back to W C S = 0 mm indicating snowpack refreezing. Fig. 11(a) shows that in response to this short-term increased 0 mm < W C S ≤ 0.3 mm, T p,90 • B (θ) at both polarization increases and then returns to preprecipitation level. This effect is distinctly visible in Fig. 11(a) Fig. 10], which shows an initial increase in brightness temperature for small values of W C S followed by dropping brightness temperature for larger snow liquid water-column.
In case 2, the snow wetness introduced via precipitation reaches W C S = 0.4 mm. Inspection of the grey shaded part (60 • ) for which the response to increased W C S is solely an increase in brightness temperature. This is thought to be due to the spatial heterogeneities in the largest RM FP area associated with θ = 60 • . Case 3 involves a strong prolonged wet precipitation accompanied by about 2.5 days of T air > 0 • C. Accordingly, only the first two hours are considered here starting from 21 February 2017 08:00 where precipitation introduces W C S = 0.8 mm. This quick wetting of the snowpack's surface and the measurement limited temporal resolution of one hour explains the immediately decreasing response of T p,90 • B (θ) without the initial marginal increase, as predicted by simulations in [27].
This strong precipitation and positive air temperature transfer the snowpack from "early-spring" phase with fluctuating snow melt/refreeze cycles to "snow-melt" phase with consistent W C S > 0 mm. One must note that for such high snow liquid water-columns the analysis of T p,90 • B (θ) with respect to S-G SPs becomes exceedingly difficult. Nevertheless, the clear post-case 3 trend in Fig. 11 is gradually increasing T p,90 • B (θ) after 23 February 2017 12:00 in response to partial refreezing of the snowpack (decreasing W C S ).
Simulations presented in Section VII-A and in [27] as well as experimental examples, such as in Fig. 11, indicate the complexity of the L-band brightness temperatures' response to snow wetness and melt. Therefore, the accuracy of snow melt detection methods, which are solely based on brightness temperature thresholds and/or air temperature measurements, (θ) measured for the three consecutive measurement scenarios "removed snow", "fresh dry snow", and "compressed snow". is questionable. This is because, depending on the temporal resolution and time of the observations, these methods may miss the response of T p B to short-term change cycles of W C S . This is of utmost relevance when estimating the number of melt-days where it is possible to miss the signature to small amounts of liquid water content.

VIII. SNOW REMOVAL AND COMPRESSION
In a field experiment in February 2015 performed at the Finnish Meteorological Institute's Arctic Research Center (FMI-ARC), snow was removed from the FP area of ELBARA-II to demonstrate for the first time the effect of dry snow on L-band brightness temperatures. The results, published in [56], corroborate the theoretical predictions of the LS-MEMLS developed in 2014 [22]. As shown in Fig. 12(a), we performed a similar experiment on 10 February 2017 during Campaign C1 in which dry snow cover was removed from ELBARA-IIs −9 dB FP areas corresponding to φ = 200 • and 35 • ≤ θ ≤ 60 • [ Fig. 2(a)]. As shown in the inset of Fig. 12(b), after a few weeks the FPs were covered with about 31 cm of fresh dry snow. Therefore, on 8 March 2017, we performed a snow compression experiment whereby we walked over the aforementioned FPs with snowshoes to mimic the situation of increased Snow density ρ S . It is reiterated that as explained in Section V, snow compression is a relevant phenomenon. Hourly scans of FPs along φ = 200 • started before snow removal on 3 February 2017 at 10:00 and continued until the end of C1. (θ) as functions of nadir angle θ, for p={H,V}, and the three consecutive measurement scenarios "removed snow", "fresh dry snow", and "compressed snow". It shows that T H,200 • B (θ) increases when the FP areas change from "removed snow" to "fresh dry snow", and finally to the "compressed snow" scenario. This increasing T H,200 • B (θ) with increasing ρ S agrees with the LS-MEMLS simulations previously reported in [23]. It is known that the effect of changing dry Snow density ρ S on T V B (θ) at vertical polarization is limited [23]. It is shown in [57] that near the Brewster angle, brightness temperatures at vertical polarization are least influenced by the snowpack. Strictly speaking, the definition of a Brewster angle (θ Brewster = arctan(n 2 /n 1 ) where n 2 and n 1 are the refractive indices of the regions containing the incident and the transmitted wave) is only applicable to a double layer system (one interface). However, the cumulative Brewster effect of multiple dielectric interfaces leads to a Brewster-like angular behavior of emission at vertical polarization, meaning that emissivity at vertical polarization is maximal at a given observation angle. Accordingly, the difference between T V,200 • B (θ) for the "fresh dry snow" and the "compressed snow" scenarios is much smaller than the difference between "removed snow" and "fresh dry snow" scenarios. Accordingly, after the "removed snow" scenario T V,200 • B (θ) responds clearly to the formation of "fresh dry snow" with ∼ 20 K increase as is apparent in Fig. 12(c).
is ∼ 2.4 K larger for the "fresh dry snow" scenario with lower ρ S than for the "compressed snow" scenario with correspondingly higher density. This is an experimental proof of the theoretical anticipation that for p=V and θ larger than the Brewster angle, T p B (θ) decreases with increasing ρ S of dry snow [23]. This effect is more distinguishable for snow cover over unfrozen ground.

IX. SPATIAL HETEROGENEITIES
Emission model simulations such as performed with MEMLS [4] indicate that regardless of snow cover conditions, T H B (θ) should consistently decrease with increasing θ. However, when inspecting the time-series of T p B (θ) in each campaign it is evident that the arrangement of T p B (θ) for different nadir angles θ is sometimes mixed and thus appears to contradict theory. In some cases, this can be due to RFI contamination of measurements. However, even after rigorous RFI filtering, such examples exist in the brightness temperature time-series. Here, we use photographic records of the site to establish our hypothesis that such anomalous signatures can be due to the combined effect of 1) L-band RM scanning system and 2) existing spatial heterogeneities over the site. We first explain that, as illustrated in Fig. 2, in our L-band RM setups, elevation scanning across different θ means observing physically different FP areas. There can be conditions where different FP areas exhibit significantly different S-G SPs and thus influencing angular dependence of T p B (θ). Fig. 13 In addition to the multiangle L-band Brightness temperatures T p,φ B (θ) measured at polarization p={H,V}, in-situ soil, snow, and meteorological measurements were collected whose summary was provided. The site schematics indicate that the measurements were designed to explore spatial heterogeneities over the site by scanning along various azimuth angles (φ) and at multiple nadir angles (θ). Furthermore, in-situ soil sensors, deployed along multiple transects, characterized the subnivean ground and its spatial heterogeneities. L-band radiometry was performed hourly to capture the effects of both short-and long-term events, such as precipitation and ground permittivity variations, respectively, on the upwelling microwave emission of the snowcovered ground.
Deployment of artificially created FootPrint (FP) areas was another key feature of the measurements at the Davos-Laret site. For example, the "charcoal-sand area" allowed for minimal surface scattering from the S-G interface and impedance matching to reduce the effect of the said interface on L-band emission. Furthermore, the "reflector area" (where a large metal-mesh was placed on the ground before snow accumulation) allowed for the exclusive measurement of L-band emission of snow, which is in direct relationship with the Snow liquid Water-content W S . Accordingly, Snow liquid Water-Column W C S , estimated from T p R (θ) over exclusively the "reflector area" and measured Snow height h S , was used as a "derived measurement" to study the response of T p,φ B (θ) to S-G SPs and meteorological events such as dry and wet precipitations.
Four key data analysis examples were provided to experimentally investigate the theoretically anticipated response of T p,φ B (θ) to S-G state variations. In some cases, experimental "long-term" proofs were missing. Accordingly, examples were provided and discussed where the effect of wet and dry precipitation on T p,φ B (θ) is evident. It was shown that T p,φ B (θ), especially at polarization p=H, temporarily increases in response to the lower density snow layer formed atop the snowpack as a result of intense dry snowfall that strengthens impedance match between air and the snowpack. Thanks to the long time-series of measurements stretched over several years, multiple instances of each of the studied phenomena can be found, hence the "long-term" proofs.
The overview of all measurements (Section V), the effect of spatial heterogeneities on the angular pattern of T p,φ B (θ) (Section IX), and the overview of all campaigns' RM data as a function of θ and in the framework of snow cover state (Section VI) are examples on how this dataset can be inspected, visualized, or used in future scientific analysis. Accordingly, the data presented are made available in [30].
A general finding that can be derived from the L-band T p B (θ) time-series measured during the four campaigns is the following: In contrast to passive microwave measurements at higher frequencies, it is rare at L-band that the emission of a seasonal snowpack behaves like a "black body" (T p B (θ) ≤ 273.15 K at any θ). In our observations, "black body"-like emission was most likely observed for the largest θ during the "isothermal" phases in C1, C2, and C3; for the "isothermal" phase in C4, this was not the case. It seems that "black body"-like emission at L-band is most likely when the snowpack is sufficiently moist to shield emission from the ground, but moisture in the top layer of the snowpack is not too high to dominate T p B (θ) via the high reflectivity of the snow-air interface.
In summary, in addition to a comprehensive RS dataset, made available in [30], this work provides novelties such as 1) using W C S as "derived measurements" for in-depth analysis of T p,φ B (θ), 2) artificial FPs such as the "reflector area" and the "charcoal-sand area", and 3) snowpack manipulations (compression and removal). Furthermore, the four consecutive campaigns offer several key take-away messages for future work and similar efforts. First, it is important to maintain a temporal resolution suitable for capturing the phenomena of interest. For example, intense snow precipitation changes SWE and Snow density ρ S ; however, its observed effect on T p,φ B (θ) depends on the temporal resolution of radiometry. Second, when performing scientific campaigns, video and photographic documentation of the environment is critically helpful in providing detailed insight into the changes in the environment and the snowpack. For instance, pictures easily highlight the relevance of snow surface topography and preferential meltwater flow paths. Third, it is important to establish long-term measurements to develop a large enough umbrella of knowledge covering multiple key phenomena, which influence scattering and microwave emission of the S-G system. These long-term measurements should consist of a baseline of measurements, which in our case was L-band radiometry over "natural area" and "reflector area" as well as auxiliary soil, snow, and meteorological measurements. Additional components and modifications can be introduced depending on the targeted scientific questions. For example, the creation of the "charcoal-sand area" and the introduction of additional in-situ sensor transects were further components introduced in Campaigns C2 and C4.
Finally, it is prudent to use the presented dataset in the future to understand and improve the performance of Snow density and Ground permittivity (ρ S , G ) retrieved from L-band RM data with the approach first proposed in [23], analyzed theoretically in [55], further developed and applied to close-range brightness temperatures in [28] and [56], and used for the first time with satellite-based T p B (θ) in [24]. It will allow to draw conclusions on the retrieval performances in various conditions such as frozen/unfrozen ground and in the presence of moisture (liquid water) in the snowpack.