Global Navigation Satellite System Reflectometry (GNSS-R) is a microwave remote sensing technique that takes advantage from signals of opportunity emitted by the GNSS constellations (e.g., GPS) to measure the properties of the Earth surface to reflect the signal toward the receiving antenna, thus implementing a sort of bistatic radar configuration for observing the Earth. The application of this technique for measuring ocean wind and waves from space is consolidated, whilst the applications over land are getting interest from the scientific community. This project should demonstrate: i) to what extent what we know from models and past ground/airborne experiments on GNSS-R applications for soil moisture and vegetation biomass is still applicable from space, ii) which additional phenomena one should try to model, iii) which final bio-geophysical products one can deliver and how (with recommendations for future systems). It will exploit previous data sets collected from ground-based and airborne receivers and GNSS-R data delivered by the UK TechnoDemoSAT-1 satellite carrying on board a GNSS-R receiver and the NASA CYGNSS constellation of 8 small satellites. Spaceborne data will be compared to reference data from in-situ sensors and other earth observation products. The physical and electromagnetic phenomena that determine the signal will be investigated through model simulations, to understand specific limitation of spaceborne systems (e.g., related to SNR). The simulator developed in the past will be updated introducing those physical mechanisms peculiar of a spaceborne sensor (e.g., topography). Simulator outputs will help developing methods to retrieve the target parameters (soil moisture and vegetation biomass) from real data. A proper combination of GNSS-R data with virtual (i.e., a priori statistical information) or real (e.g., from other sensors/platforms) measurements will be exploited to overcome the ill-posed retrieval problem.
The main novelty of the proposed research is the development of the physically based approach for retrieving soil moisture and biomass from satellite GNSS-R data. The method will exploit machine learning algorithms trained by a data base of target surface parameters (soil moisture, vegetation cover, topography, etc.) and related GNSS-R observables (DDM¿s or peak reflectivity). Such a training set has the advantage of covering a wider range of target parameters with respect to empirical databases, to investigate new system configurations in view of future payload concepts and can be eventually complemented by empirical data to improve algorithm robustness. The previous experience of the proposing team in the electromagnetic modelling of the earth surface scattering and the availability of a simulator of the GNSS-R DDM¿s (SAVERS) provide a suitable heritage to make the implementation of the approach feasible. We remind that there are in the literature implementations of soil moisture retrieval algorithms, generally based on empirical approaches, whilst the retrieval of biomass from satellite data is at a very early stage. Secondly, the new method will try to integrate observations from different sensors, like Sentinel-1 and CYGNSS to implement a sort of multistatic system. Scattering mechanisms in monostatic and specular configurations are different, so that a combined exploitation of those data can bring more information into the inverse retrieval problem. This idea could be implemented in a new small satellite mission by a suitable choice of the orbit to enable almost coincident measurements of the backscattering coefficient by a radar (like Sentinel-1) and reflectivity measurements by a GNSS-R based platform. In summary, the research aims not only to generate accurate products from existing GNSS-R missions (especially CYGNSS) but also to conceive new system configurations to be implemented in future missions.
The preliminary analysis of sensitivity of GNSS-R signal to land parameters will exploit the physical relations between the GNSS-R signals (SNR, reflectance) to biomass and soil moisture and perturbations from topography and incidence geometry. Data colocation with other sensors will enable to explore the indirect relations with other related variables e.g. NDVI and LST. The analysis will be carried out in regions that offer a good network of soil moisture measurements by in-situ stations (e.g., sites of the International Soil Moisture Network, ISMN, where data can be freely downloaded, or from the Merguellil catchment in central Tunisia thanks to a cooperation with CESBIO in France), which will be complemented by collocated soil moisture measurements from the SMOS and SMAP satellite missions. Large scale comparisons are also foreseen using satellite-based products, such as biomass maps available at global scale. Some auxiliary information will also be gathered, such as precipitation conditions, vegetation conditions, and land cover (e.g., using the NDVI from the MODerate Imaging Spectroradiometer). All these activities will be based on freely available data (including CYGNSS and TDS-1) and are therefore feasible; moreover, part of this preliminary work has been already done in previous projects so that the needed data sets are available with limited efforts.
Modeling the reflected GNSS signal observed from space over land is without a shadow of a doubt a very new and challenging task. The proposing team has been deeply involved in accurate modeling of the coherent and incoherent land surface scattering exploiting asymptotic models for soil and radiative transfer equation solution for vegetation, including multiple scattering. The already developed software tool (SAVERS) will be updated to model more accurately scattering mechanisms and/or satellite-related effects determining the signal peculiarity revealed by the sensitivity analysis task. The SAVERS simulator already considers the topography, whose effect on the satellite signal is well established. It will be further updated and used to understand the contribution of coherent and incoherent signal at spaceborne height, and the effect of target inhomogeneity on both of them. In fact, this is an open issue at scientific level. The signal exhibits fluctuations that are due both to interference effects (e.g., interferences of the different Fresnel¿s zones as they pass through different portions of an inhomogeneous surface according to the satellite motion) as well as random fluctuations due to the random nature of the surface. Understanding and modelling this variability is important to set up a suitable strategy in the receiver, in particular setting the coherent integration time to increase the SNR and the incoherent integration time to reduce the fluctuation standard deviation. Moreover, the knowledge of the fluctuation properties (i.e., the error model) is important for the design of the estimation algorithm retrieving the target parameters.