*2.2. Land Surfaces*

The use of GNSS-R for land-surface applications requires additional considerations because the dielectric properties of this scattering medium have significantly more variability than that of the ocean surface. Here, we provide some of the latest findings of the CYGNSS mission, including key-results, discussions, and on-going research activities. At present, a significant number of researchers within the CYGNSS Science Team are working on land surface applications, which remain less explored than ocean products.

Session 3: Land Processes I

The first Land Processes Session consists of a series of presentations and discussions focusing on the use of CYGNSS data and observations related to (a) surface SMC estimation, (b) creation of dynamic inland water body masks, and (c) advancements in forward GNSS-R-related electromagnetic scattering and forward models. GNSS-R-like observations have been shown to be sensitive to changes in surface SMC [27–31] as well as the presence (or absence) of inland water bodies [32,33]. More recently, a 3 km CYGNSS Signal-to-Noise SNR-based surface SMC estimation approach was developed in [30], with results comparable to those by the NASA SMAP mission [34] and unbiased Root-Mean-Squared (ubRMSE) on the order of 0.045 m3/m3. Given this background, the following briefly summarizes the key finding from each of the session's presentations.

Forward scattering properties of Earth-reflected GNSS signals were evaluated over land surfaces. The CYGNSS End-to-End Simulator (E2ES) was updated. A GNSS-R model capable of evaluating both the incoherent and the coherent scattering terms was developed based on the Huygens–Kirchhoff principle [35]. Results demonstrated the impact of higher order Fresnel zones on the spatial resolution of GNSS-R over heterogeneous areas, showing "ringing" fluctuations in the reflected power near high-contrast boundaries.

An inland water body detection method was also presented during this session. The method [36] demonstrates the utility of ~2 years CYGNSS Level-1 Delay-Doppler Maps (DDM) to create [1,3] km water masks (Figure 6). Inland water surfaces at L-band are presumed smooth enough such that specular scattering is the dominant features in the observed DDM. Therefore, by deploying a DMM coherence persistence and detection method [30], mapping inlands becomes possible. The derived inland water body masks were qualitatively and quantitively compared to, and validated against, historical (30 years) water body occurrence maps by Pekel et al. [37]. False-positive areas were identified and removed via comparison to the Pekel occurrence maps, and examination of high-quality Digital Elevation Maps (DEM). Areas with persistent coherence, unknown water bodies, and exceptionally flat surfaces were also removed. The current analysis was based on ~2 years of CYGNSS observations. Additional investigation to identify false-positive and improve the overall detection method is required.

Russo et al. [38] outlined an entropy-based CYGNSS coherence detector—via eigenvalue decomposition of the DDMs—which, overall showed good agreement with wetland maps derived from Advanced Land Observing Satellite-2 (ALOS-2). However, the current method does not discriminate between open water and flooded vegetation and merits further investigation. Additionally, Russo et al. outlined a newly initiated research project with the ultimate goal of fusing CYGNSS-derived wetland with future wetland products from the NASA-ISRO Synthetic Aperture Radar (NISAR) mission. The primary motivation of this work was to develop the GNSS-R/SAR framework for future CYGNSS/NASA-ISRO (NISAR) activities.

Santi et al. [39,40] presented a series of Artificial Neural Network (ANN) implementations to use CYGNSS SNR observations to estimate various geophysical properties. Specifically, a multi-layer ANN model was trained to estimate AGB and tree height using CYGNSS SNR observations which uses training and reference data from the Geocarbon Pan-tropical forest maps by Avitabile et al. and tree height maps derived by the Geoscience

Laser Altimeter System (GLAS) instrument aboard the Ice, Cloud, and land Elevation Satellite (ICESat-1)

**Figure 6.** The 1−km land water body mask derived from CYGNSS L1 coherence-detection by M. Al-Khaldi et al. The map was generated using 1 year of CYGNSS data.

Recent theoretical developments in GNSS-R and scattering models over rough surfaces were also presented in this session [41]. Results and model validation efforts over the San Luis Valley, CO, showed promising matchups with the CYGNSS L1 Bistatic Radar Cross Section (BRCS). However, these studies also stressed the importance of accurately capturing and modeling multi-scale surface roughness and topographic effects.

This session also included several topics about retrieval and characterization of SMC and vegetation properties. A machine learning approach using random forest regression was presented to estimate SMC at a global scale [42]. The model was trained and generated using the reflectivity and relevant geophysical data layers, such as Normalized Difference Vegetation Index (NVDI) and DEM maps, along with SMC from the International Soil Moisture Network (ISMN). Results were validated over the Contiguous United States (CONUS) with consistent spatial patterns and magnitudes as those observed by the SMAP mission.

Similarly, by deriving a semi-analytical expression of vegetation transmissivity, and soil reflectivity, it was demonstrated how SMAP and CYGNSS observations can be concurrently leveraged to estimate either the Vegetation Optical Depth (VOD) and the SMC [43] (Figure 7). Additionally, a newly initiated study by Pu et al. sought to examine the effects of assimilation SMAP and CYGNSS SMC in near-surface weather forecasting models. This ongoing study showed that CYGNSS-derived SMC assimilation is on par with SMAP-based forecasts. However, further investigation is required to better quantify the added value by CYGNSS for SMC assimilation. Prior work by Pu et al. [44,45], however, demonstrates that strongly coupled land-surface assimilation frameworks which assimilate in situ, or SMAP, soil moisture can provide additional short-range and near-surface weather forecasting. The effects of CYGNSS SMC assimilation will be reported in future studies.

The use of CYGNSS high resolution ~3 km SMC maps was demonstrated by an experimental study supported by SERVIR [46]. This CYGNSS product, along with the ~3 km Land Information Systems (LIS) surface model, Integrated Multi-satellitE Retrievals for GPM (IMERG) rainfall, and Visible Infrared Imaging Radiometer Suite (VIRRS) vegetation information, was leveraged for locust monitoring applications in East Africa. SERVIR is a joint venture between NASA and the U.S. Agency for International Development. The SERVIR program is collaborating with regional entities in West and East Africa to evaluate the utility of Earth observations (EO), and their contribution to operational desert locust monitoring and tracking systems supported by the United Nations (UN) Food and Agriculture Organization (FAO).

**Figure 7.** Example of CYGNSS-derived VOD by Xu et al.

Session 4: Land Processes II

Several studies have demonstrated the importance of coherent scattering of the GPS reflected signal over land [23,36,47–49]. In general, for coherent returns to exist, the GPS signal has to be reflected over very large flat areas with the surface RMSE height lower than ~10 cm at L-Band as presented in [50]. Recent analysis suggests that coherent returns are therefore mostly associated with the presence of inland water bodies [51,52]. This session will provide updates on current research using the SNR to detect inland water bodies. For land applications, surface roughness is a critical parameter for scattering models as detailed in the session. This surface roughness can be estimated using lidar airborne measurements [53]. Reflectometry has been used in various studies for land applications especially for retrieving SMC [53–56]. In this session, an update on the existing and potential calibration and validation sites for soil moisture applications is provided.

The first three presentations demonstrated the potential of the use of CYGNSS over land surfaces to detect and monitor water extent over short time scales [57]. The first method analyzed the distribution of time series of SNR for a given pixel to define dry and wet areas. With this method, a dynamic water extend mask can therefore be derived. Several case studies showed good correlation with rainfall events and the Pekel mask considered as the truth. Sensitivity of CYGNSS to water extent changes over short time scales was demonstrated by analyzing the spatial SNR variations with the seasonality of the Pascagoula River. Better detection results were obtained with a higher sampling rate. Finally, an investigation on the impact of land characteristics, such as SMC, surface water, topography and Vegetation Water Content (VWC) on the coherence of CYGNSS reflected signals was presented. The coherence was quantified using the tracked carrier phase from CYGNSS raw IF data. It was concluded that the coherent scattering term is most often present over inland water bodies.

The next topic of this session was assessing the impact of surface roughness, which is a critical parameter for correctly assessing the scattering of GNSS signals over land surfaces. A theoretical model was developed to decompose the surface into topography elevation and slopes, small-scale surface roughness, and surface correlation length. This new parametrization of the surface was tested using several classical forward-scattering models, including Geometrical Optics, Physical Optics, and Numerical Maxwell Model 3D (NMM3D). The modeled DDMs were then compared to CYGNSS-derived DDMs, showing a good agreement. In May 2020, experimental activities were performed along two cal/val sites over the San Luis Valley to characterize the small-scale surface roughness using lidar. Lidar-derived DEMs were generated at two different spatial resolutions, ~10 cm and ~30 cm.

The last three presentations dealt with actual and potential SMC cal/val sites for CYGNSS. First, an update was provided about the status of the SOILSCAPE in situ SMC sensor networks, that were installed in two cal/val sites in the San Luis Valley. All sensors were shown to be working properly. Calibration of the SMC measurements is in process. New potential cal/val sites have been identified in the USA (Walnut Glutch, AZ, White Sands, NM) and in New Zealand. New Zealand cal/val sites are particularly of interest because the next generation GNSS-R receiver [58] is going to be installed on a regional Air New Zealand commercial aircraft to complement CYGNSS data, and to test this new receiver in preparation of a potential CYGNSS follow-on mission. An extensive analysis of the coverage provided by one Air New Zealand commercial aircraft was performed to identify the best cal/val site locations for SMC and wetlands studies. These locations were cross-compared with the CYGNSS coverage, showing several overlapping sites. This project will advance terrestrial and coastal retrievals, by generating long-term datasets with high spatial resolution and high spatiotemporal sampling. In addition, in support of this project, the deployment of the next-generation GNSS-R receiver on one Air New Zealand aircraft was simulated to get a better understanding of the differences of the reflected signal acquired on an aircraft as compared to a satellite. Flight paths, flight frequency, and GNSS-R coverage were analyzed. It was found to have promising coverage, except for some mountain regions in the South Island (Figure 8). Installing this new receiver on one aircraft will provide a large amount of information over both ocean and several land cover types. It was concluded that the extension to a fleet of regional aircrafts will generate an unprecedented GNSS-R scientifically valuable dataset.

**Figure 8.** Simulated number of GNSS-R measurements over 1-year, considering just 1-single Air New Zealand aircraft by Linnabary et al.
