Evaluation of CYGNSS Observations for Snow Properties, a Case Study in Tibetan Plateau, China

Abstract

Snow plays an important role in the water cycle and global climate change, and the accurate monitoring of changes in snow depth is an important task. However, monitoring snow properties is still challenging and unclear, particularly in the Tibetan Plateau, which has rough land and uneven terrain. The traditional monitoring methods have some limitations in monitoring snow depth changes, and the Global Navigation Satellite System-Reflectometry (GNSS-R) provides a new opportunity for snow monitoring. This paper employed data from the Cyclone Global Navigation Satellite System (CYGNSS) to discover the effect of snow properties. Firstly, the observations of CYGNSS were used to find the sensitive to snow properties, and the relationships between signal to noise ratio (SNR), leading edge slope (LES), surface reflectivity (SR), and snow depth were studied and analyzed, respectively. It is found that the correlation between the first two parameters and snow depth is poor, while SR can indicate the changes in snow depth, and is proposed as an indicator of SR change, namely, surface reflectivity–difference ratio factor (SR–DR factor). Furthermore, the long-time series data in the Tibetan Plateau (2018–2019) are used to analyze its effects on the time series of the SR–DR factor, while the influences of the soil freeze/thaw (F/T) process and soil moisture are excluded during the analysis. The results indicate that the SR–DR factor can be a good indicator and discriminator for snow depth. Our work shows that space-borne GNSS-R has the potential for the monitoring of snow properties.

1. Introduction

One of the fundamental components of the water cycle and global energy is the change in snow, which is of great importance for regional climate change, water use, and natural disaster monitoring [1]. The Tibetan Plateau is a sensitive area, and an ecologically fragile zone at risk of climate change, due to its harsh climate and environment, fragile ecology, and frequent snow disasters. It is the region with the highest average elevation in the world. The Tibetan Plateaus is rich in glaciers, snow, and underground ice resources [2], including the headwaters of many rivers in China, and snowmelt water is an important supplementary source of rivers. In the last few years, due to the climate change, the melting of snow and ice on the Tibetan Plateau has accelerated, and the accumulated total amount of snow is also decreasing. The snow on the Tibetan Plateau has an influence not only on the climate of the Chinese mainland, but also on the water circulation and climate systems of the East Asian region. Therefore, the research on snowpack in this area is of great hydrological and climatic importance [3,4]. The Tibetan Plateau is one of the main snow distribution areas in China, and is also an important pastoral area in China. The development of local agriculture and animal husbandry is also closely related to the change in snow cover. As a result, the study of snow cover on the Tibetan Plateau is of great significance for the sustainable development of the regional ecological environment, and agriculture and animal husbandry [5,6,7].

The Tibetan Plateau is a rough land, and has uneven terrain with an average altitude of more than 4000 m. The ground observation stations are rare with the uneven spatial distribution, and the observation time is discontinuous, which cannot meet the needs of research on large-scale snow distribution characteristics [8].

Satellite observations are an effective way to monitor snow cover. Space-borne optical passive sensors provide useful information for monitoring snow cover, but cannot work in high altitude and cloudy areas, due to the limitation of cloud cover [9,10]. Passive microwave remote sensing data are one of the main means of monitoring snow changes, as this method can penetrate through the porous soil layer, cloud, fog, and so on. It can penetrate a certain depth of surface to obtain information on the surface physical parameters. Therefore, microwave remote sensing is the best technology to obtain regional snow depth and snow water equivalent monitoring [11,12,13].

Global Navigation Satellite System-Reflectometry (GNSS-R), using the reflected signals from navigation satellites, was used for remote sensing in recent years [14,15,16]. This method mainly works in the L-band, with strong penetrability, so it is essentially a microwave remote sensing technology. It has the advantages of low cost, low power consumption, and high spatial and temporal resolution, due to the continuous use of navigation satellite signals as a signal source, which does not require the development of a special transmitter. It is a useful supplement to the traditional polar orbit satellites [17].

In 2004 and 2014, the United Kingdom Disaster Monitoring Constellation (UK-DMC) and Techdemosat-1 (TDS-1) test satellites were launched for GNSS-R remote sensing research [18]. In December 2016, the Cyclone Global Navigation Satellite System (CYGNSS) was launched by the National Aeronautics and Space Administration (NASA). The main scientific goal of the system is to measure the ocean surface wind field information during tropical storms and hurricanes [19]. It also provides an important opportunity for the study of land surface parameters, such as soil moisture, vegetation, flood inundation, and wetland monitoring [20,21,22,23]. To the best of our knowledge, there is no reported research on snow cover using CYGNSS data. In this study, the CYGNSS data in the Tibetan Plateau from 1 January 2018 to 31 December 2019 are selected for snow cover analysis. This provides a new method for expanding the research field of satellite-borne CYGNSS, and provides a unique remote sensing method for the study of snow characteristics on the Tibetan Plateau. With the advantages of CYGNSS, using it in the remote sensing of the surface snow depth will improve the temporal resolution of snow depth data, and reduce the effects of background brightness temperature and radio frequency interference (RFI).

The data description and processing method are presented in Section 2. The analysis using CYGNSS surface reflectivity (SR) and the surface reflectivity–difference ratio (SR–DR factor) are carried out in Section 3. Finally, the conclusions are given in Section 4.

2. Method

2.1. Data Description

The Tibetan Plateau is the highest plateau in the world, located between 26°N to 39°47′N latitude and 73°19′E to 104°47′E longitude. The topography is complex, and it has large altitude differences in various regions (Figure 1). The average annual temperature decreases from 20 °C in the southeast to below −6 °C in the northwest. In this study, we employed five public datasets to investigate the potential of CYGNSS for the study of snow cover on the Tibetan Plateau: (1) CYGNSS Level-1 (L1) data [24], (2) the Terra and Aqua combined Moderate Resolution Imaging Spectroradiometer (MODIS) Land Cover Climate Modeling Grid (CMG) (MCD12C1) version 6 data (0.05 °C) [25], (3) long-time series dataset of snow depth in China (1979–2019) [26], (4) Soil Moisture Active and Passive (SMAP) L3 radiometer global daily 36 km EASE-grid soil moisture, version 7 [27], (5) European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) soil temperature data [28].

2.1.1. CYGNSS Data

CYGNSS was successfully launched in December 2016. Eight small satellites carrying GNSS reflection signal receivers make up its constellation. Each CYGNSS satellite is equipped with four delay-Doppler map instruments (DDMI), and each DDMI includes two left-hand circularly polarized (LHCP) antenna and one right-hand circularly polarized (RHCP) antenna. The LHCP antenna receives GNSS signals from specular and scattering points on the Earth’s surface. The RHCP antenna receives GNSS signals directly, to calculate its position and velocity. Each CYGNSS satellite can measure and observe reflections in four directions simultaneously, which means that the CYGNSS system can obtain data from 32 reflection points [29]. The delay-Doppler mapping (DDM) obtained by the receiver is a function of the surface medium, antenna gain, distance, statistical properties, and scattering geometry. The CYGNSS data used in this study were from level 1, version 2.1.

At present, most studies consider that the receiver primarily collects the coherent scattered energy in the first Fresnel region. When only the coherent energy is considered, we can calculate the DDM based on the Friis transmission formula and the Fresnel reflection coefficient of an equivalent smooth surface [30].

$$Pcoh=\frac{P_T{G}_T{\lambda }^2{G}_R}{{\left(4\pi \right)}^2{\left(R_R+R_T\right)}^2}\Gamma \left(s,\theta i,\epsilon \right)$$

(1)

• $P_T$: The transmitted power of GNSS satellite;
• $G_T$: The GNSS satellite antenna gain at specular point direction;
• $G_R$: The receiver antenna gain;
• $P_T{G}_T$: The GNSS equivalent isotopically radiated power (EIRP);
• $\lambda$: The wavelength;
• $R_R$ and $R_T$: The distance between the receiver and the specular point and the distance between the transmitter and the specular point, respectively.

According to Equation (1), the reflectivity can be expressed as:

$$\Gamma \left(s,\theta i,\epsilon \right)=\frac{{\left(4\pi \right)}^{2}Pcoh{\left(R_R+R_T\right)}^2}{{\lambda }^2{G}_R{G}_T{P}_T}$$

(2)

where $s$ is the surface rms height, which can characterize the surface roughness, ${\theta }_i$ is the incidence angle and the dielectric constant of the geophysical parameters. Surface reflectivity (SR) is usually a small positive value, which is converted to decibel for visualization and data analysis according to Equation (3).

Table 1. The Cyclone Global Navigation Satellite System (CYGNSS) level 1 data variables.

Name	Comment
ddm_timestamp_utc	DDM sample time
sp_lat	Specular point latitude, in degrees north
sp_lon	Specular point longitude, in degrees east
sp_inc_angle	The specular point incidence angle, in degrees
sp_rx_gain	The receive antenna gain in the direction of the specular point, in dBi
gps_eirp	The effective isotropic radiated power (EIRP) of the L1 C/A code signal within ± 1 MHz of the L1 carrier radiated by space vehicle, sv_num, in the direction of the specular point, in Watts
rx_to_sp_range	The distance between the CYGNSS spacecraft and the specular point, in meters
tx_to_sp_range	The distance between the GPS spacecraft and the specular point, in meters
power_analog	17 × 11 array of DDM bin analog power, Watts

shows the main parameters for the L1 data used in this study.

$$dB=10lo{g}_{10}\left(\Gamma \left(s,\theta i,\epsilon \right)\right)$$

(3)

It is worth noting that the treatment of the coherent and incoherent scattering is an open issue when dealing with the CYGNSS data [31]. When using CYGNSS data for soil moisture retrieval, the coherent energy comes primarily from specular reflections of water inland within the GNSS-R footprint, which leads to the increase in DDM peak value. The peak energy of the coherent part is many times greater than that in the non-coherent portion. The satellite-based GNSS-R receiver mainly receives coherent scattered signals from the first Fresnel zone around the specular reflection point [32]. In fact, the signals received by GNSS-R receivers, in most cases, contain signals from both coherent and incoherent scattered fields, due to variations in surface roughness. Most of the currently available studies on satellite-based GNSS-R remote sensing of surface soil moisture simplify the scattering mechanism of the L-band on the actual surface, assuming that only coherent scattering occurs on the terrestrial surface, and incoherent scattering is ignored [31]. We removed data with SNRdB less than 2 dB and CYGNSS antenna gain of less than 0 dB. The elevation angle of the specular points is less than 65°.

2.1.2. IGBP Land Cover Classification

The land surface scattering process is complex. In addition to the influence of surface roughness, the surface scattering signals received directly by GNSS-R receivers are also impacted by other factors such as vegetation layer and topography. There are various types of land cover on the Tibetan plateau, and different vegetation covers have different effects on the received signals. In order to reduce their influence on the received signal power, the consequent analysis is performed in relatively uniform land cover categories. In this way, we can ensure that the features in each zone are comparatively unified, so that the differences in the CYGNSS observations caused by various types of surface coverage can be considered.

The MODIS Land Cover Climate Modeling Grid Product (MCD12C1) provided are the sub-pixel proportions of each land cover class in each 0.05° pixel and the aggregated quality assessment information for the IGBP scheme [25]. Here, we have employed the IGBP of 2020 to obtain the land cover information of the Tibetan Plateau region (Figure 2a). The Tibetan Plateau region is dominated by four categories: high-vegetation-covered area; moderate-vegetation-covered area, low-vegetation-covered area, and barren or desert area (Figure 2b). In this paper, we conducted the analysis of snowpack characteristics within the same land cover type, as far as was possible.

2.1.3. Long Time Series Dataset of Snow Depth in China

The long-term series of daily snow depth dataset in China (1979–2019), released by the National Tibetan Plateau Data Center (NTPDC), provides daily snow depth distribution data for China from 1 January 1979 to 31 December 2019. The raw data used to invert this snow depth dataset are from the scanning multichannel microwave radiometer (SMMR) (1979–1987), the special sensor microwave imager (SSM/I) (1987–2007), and the special sensor microwave imager/sounder (SSMI/S) (2008–2019) daily passive microwave bright temperature data (EASE-grid), processed by the National Snow and Ice Data Center (NSIDC). It has a spatial resolution of 25 km, which can be used for climate analysis, hydrological modeling, and water management on a large scale and in long time series [33,34,35]. During the processing of the data, the brightness temperatures of different sensors are first cross-calibrated, and the observations affected by the snow depth observation errors of ground stations, the surface water, and the liquid water content in the snow layer are removed before the inversion of the snow depth.

Using this dataset, we calculated the snow cover change in the Tibetan Plateau between January 2018 and December 2019, and the corresponding snow depth changes for four different land cover types. The snow accumulation trend under different land cover types is similar. From October every year, the Tibetan Plateau gradually enters the snowfall period, reaching the peak of snowpack in January and February of the following year. The snow gradually melts after April as the temperature rises. Figure 3 shows the snow depth map of the Tibetan Plateau for one month in summer and winter using this dataset.

By comparing the values with the daily reflectivity data of CYGNSS in the same region, the correlation between snow depth and reflectivity can be inferred, and the potential of CYGNSS in snow feature research can be verified.

2.1.4. SMAP L3 Radiometer Global Daily 36 km EASE-Grid Soil Moisture, Version 7

Previous studies show that GNSS signals reflected from the surface are sensitive to changes in soil moisture, and the soil moisture also changes the SR [22]. This soil moisture product provides a composite of daily estimates of global land surface conditions retrieved by the Soil Moisture Active Passive (SMAP) passive microwave radiometer. The data are on a regular latitude/longitude grid, with predictable disparities in space and time. We used the soil moisture data to plot a time series of mean snow depth with SMAP mean soil moisture with different land cover types on the Tibetan Plateau. For more detail, please see the following. By analyzing the variations in soil moisture over the same period, the effect of SR changes due to soil moisture can be excluded; thus, we believe that the variations of SR are due to the changes of snow coverage.

2.1.5. ERA5 Soil Temperature Data

The dataset is the soil temperature at level 1 (in the middle of layer 1). The European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) has a four-layer indication for the soil, where layer 1 ranges from 0 cm to 7 cm. Soil temperature is measured in the middle of each layer. When a freeze–thaw transition occurs in soil, the water in the soil changes from solid ice to liquid water. The dielectric constants of water and ice differ significantly, so the reflectivity changes accordingly. By analyzing the changes in snow depth and SR during the period when the surface temperature does not change from positive to negative (or from negative to positive), we assessed whether the snow depth can be monitored by CYGNSS.

2.2. Calculation of the Surface Reflectivity

The incidence angle is an important factor affecting the reflectivity of GNSS constellation; the incidence angle of CYGNSS ranges from 0 to 70°. To improve data quality, the incidence angle greater than 60° is eliminated and the data in the direction of the lowest point are used for normalization. Based on the data of CYGNSS L1 from 1 January 2018 to 31 December 2019 in the Tibetan plateau, we calculated and converted all the values to the dB scale, according to Equations (1)–(3). In this paper, we assume that the energy of CYGNSS data is coherent, and incoherent scattering is not considered in the analysis. The effective sampling range of CYGNSS is between approximately 38°N to 38°S, while the latitude range of the Tibetan Plateau is about 26°N to 40°N, so some areas in the north lack data [29]. Figure 4 shows the surface reflectivity of CYGNSS specular points in the Tibetan Plateau on 1 March 2018. Comparing with the satellite image in Figure 1, we can see that the SR is higher in the area with water and snow. In order to exclude the influence of water bodies on the results, these data should be removed from the analysis.

In order to be consistent with the time series analysis, the raw SR of the specular points was transformed into gridded data. Considering the study area and the spatial resolution of snow data (Section 2.1.3), a grid was generated with a resolution of 0.25 along the geodetic latitude and longitude. The data values for each grid cell were defined as the average value of the specular points falling into that grid.

To test the reaction of SR to the changes in surface parameters on the Tibetan Plateau, the SR at the CYGNSS specular reflection points for January 2019 and July 2019 are given in Figure 5a,b. As seen in the figure, the reflectivity increases in the whole region in July compared to January. To exclude differences in CYGNSS observations due to different ground cover types, the variation in SR in response to surface snow depth is analyzed under the same ground cover types.

3. Results and Analysis

3.1. Comparison of Surface Reflectivity and Parameters on the Tibetan Plateau

The CYGNSS daily SR and snow depth were analyzed and compared under different land cover types. The snow depth in the Tibetan Plateau region was calculated using the long-term series of daily snow depth dataset in China (1979–2019). Figure 6 shows the time series of CYGNSS SR and snow depth, and we assess the ability of CYGNSS data to monitor changes in snow cover depth by comparing them.

The SR varies with the surface dielectric constant, so there are differences in SR for different ground objects. The snow on the Tibetan Plateau is mainly produced from October to April each year, during which the Tibetan region enters a period of high snow frequency, and SR increases as the snow depth increases. A portion of the SMAP data was lost between June 2019 and July 2019, due to data sampling issues with the satellite. Figure 6 shows the variation of CYGNSS reflectivity (blue left axis), which is consistent with the oscillation of snow depth (red right axis) during the months with snow accumulation. In addition, the part of the anomalous oscillation is mainly caused by the attenuation and volume scattering of vegetation. The change trend of SR and snow depth is closest in the high-vegetation-covered area. In the low-vegetation-covered area and barren/desert area, the CYGNSS time series fluctuates more by the noise, but shows an overall increasing trend. The analysis results show that the CYGNSS SR has a certain correlation with the variation in snow depth, which can be used for snow monitoring on the Tibetan Plateau.

Soil moisture is an important factor affecting SR. To exclude its influence on the results, we plotted the time series of soil moisture using the SMAP surface soil moisture data and examined it with the variation of snow depth over time, in order to investigate the correlation between them. It can be seen from Figure 7 that the interannual variation of soil moisture is closely related to the variation in snow cover, and the soil moisture content fluctuates greatly when the snow cover melts. The low-vegetation-covered area has the largest change in soil moisture. In addition, when the snow cover on Tibetan Plateau is obvious (November 2018–February 2019), the value of soil moisture under different land cover types fluctuates around 0.1 cm³/cm³, and does not change significantly. Therefore, we learn that it is not the change in soil moisture that causes the oscillation of SR.

To further verify the effect of snow depth on the SR, we also used the ERA5 soil temperature data to obtain the monthly average soil temperature over this four month period (Figure 8). We found that there is no inversion from negative to positive values of surface temperature in most areas of the Tibetan Plateau during this period. The northern region shows the relatively large temperature variation. However, as this region is beyond the detection zone of CYGNSS, we exclude the effect of surface temperature variations.

Table 2. The mean values of snow depth, SR, and soil temperature for the four different land cover types over a four month period.

Time	Snow Depth (cm)	Surface Reflectivity (dB)	Soil Temperature (°C)
(a) Mixed–Forest
2018.11	0.30817	–27.36039	–0.43541
2018.12	0.75975	–26.51822	–0.72392
2019.01	1.15806	–25.96435	–1.34392
2019.02	0.99447	–27.32706	–0.82510
(b) Open–Shrubland (Desert)
2018.11	1.98258	–35.25758	–3.73655
2018.12	3.66645	–33.51377	–7.68061
2019.01	4.54840	–32.83397	–9.16413
2019.02	4.04246	–34.33123	–7.49004
(c) Grassland
2018.11	3.38779	–34.03826	–2.11510
2018.12	6.08088	–33.24549	–4.36104
2019.01	6.35053	–33.78475	–5.42212
2019.02	5.02427	–34.37498	–4.49595
(d) Barren/Desert
2018.11	1.86681	–31.47885	–1.29014
2018.12	2.97565	–30.19818	–7.56332
2019.01	3.60599	–31.14914	–8.23865
2019.02	3.38220	–30.62580	–4.55624

shows the mean values of snow depth, SR, and soil temperature for the four different land cover types over a four month period. It can be seen that the trends of snow depth and SR over time are approximately the same when the soil temperature is negative without more significant changes. For other commonly used evaluation metrics, such as leading edge slope (LES) and signal to noise ratio (SNR), we provide a brief discussion in Appendix A.

3.2. Surface Reflectivity Difference Ration Factor

In Section 3.1, we evaluated the effects of snow depth on surface reflectivity, and find that it is more sensitive than the other GNSS-R observables, i.e., SNR and LES (Appendix A). Therefore, we provide an indicator in this section to illustrate its effect on the detection of snow properties. Here, we define this indicator as surface reflectivity–difference ratio factor (SR–DR factor) [36]. The factor SR–DR is defined as follows:

$$F\left(t\right)=\frac{\Gamma \left(t\right)-\Gamma min}{\Gamma max-\Gamma min}$$

(4)

where Γ(t) is the SR at time t, and ${\Gamma }_{max}$ and ${\Gamma }_{min}$ are the maximum and minimum SR in the time range, respectively. It should be noted that within the form of this equation, the snow water equivalent (SWE) can be achieved when the SWE is in quite good relationship with snow depth.

The CYGNSS data for 2019 was studied as a case study. Figure 9 shows the SR–DR factor for the Tibetan Plateau in January, February, March, and July 2019, which has the same resolution of 25 km as SR. Analyzing the factor for the Tibetan Plateau in 2019, we found that the SR–DR factor increases in summer compared to winter for the whole region. It also changes with snow depth in winter when there is snow cover on the Tibetan Plateau.

Figure 10 shows that the SR–DR factor for January 2019 is mainly concentrated between 0 and 0.4. In February and March 2019, SR–DR factor shows an increase, without a large change in soil moisture during this period. In July, it shows a higher value, mostly greater than 0.4. The snow depth is also lower at this time, due to the arrival of summer.

We selected a 1° × 1° size area with different land cover types for the analysis of the relationship between the SR–DR factor and snow depth as follows: (a) barren or desert area (79°–80°E, 33°–34°N); (b) low-vegetation-covered area (81.5°–82.5°E, 31.5°–32.5°N); and (c) moderate-vegetation-covered area (95°–96°E, 29°–30°N). We removed the high-vegetation-covered area because this part of the Tibetan Plateau is small and mostly concentrated in mountainous areas with complex topography. The topographic relief and vegetation roughness may have a great impact on the effective surface reflectivity, which will make the results inaccurate. The time with snow cover on the Tibetan Plateau is concentrated from November to March of the following year, so we choose CYGNSS data from November 2018 to March 2019, and the snow depth at the coordinates of the center point of the area is used to express the change in snow accumulation. Figure 11 presents the time series of SR–DR factor versus snow depth in the research area. From the simulation, we can see that the oscillation of the snow depth in the barren area is, in general, consistent with the daily CYGNSS SR–DR factor. The differences become progressively larger as the vegetation cover level increases.

4. Conclusions

Since the launch of GNSS-R satellites, many scholars have used them to study the parameters of wetland dynamics, sea surface wind speed, and soil freeze/thaw. This paper takes CYGNSS as an example to study its potential application in snow monitoring on the Tibetan Plateau. The CYGNSS data collected from January 2018 to December 2019 were processed and analyzed to calculate surface reflectivity and compare it with snow depth. Soil moisture from SMAP and temperature data from ERA5 were used to study the changes in both when the snow depth varied significantly.

With the help of typical GNSS-R observables (LES, SNR, and SR), we found that SR can be a good potential parameter to indicate the snow depth. Therefore, based on the SR, the finally developed indicator, which is the SR–DR factor, shows an apparent good relationship with snow depth. This factor can represent the snow water equivalent to some extent through the ratio form in the equation.

Compared with conventional remote sensing satellites, the sampled data from CYGNSS have high spatial and temporal resolution. This demonstrates the potential of satellite-based GNSS-R to monitor snow depth at a high spatial resolution.

In the analysis of GNSS-R data, it is generally believed that the signal received by the receiver is mainly composed of coherent scattering along the mirror direction, but, in fact, the incoherent scattering affects the results, and their relative contribution depends on the receiver height, surface roughness, and vegetation coverage. In this paper, only the case of coherent scattering is considered.

The CYGNSS was originally intended for sea surface wind field research, and many of its applications on land are exploratory studies. This paper qualitatively proves that satellite-based GNSS-R can be used for snow monitoring, but still lacks a quantitative evaluation to determine the specific value of snow depth as retrieved from SR. It can be further analyzed with other estimators in future research. In addition, the coverage of CYGNSS is mainly concentrated in the middle and low latitudes, so there is a lack of effective data in some areas of the northern Tibetan Plateau. In practical application, it may be necessary to combine the traditional remote sensing satellite data to achieve high-precision snow monitoring.