Comparison of a Semiempirical Algorithm and an Artificial Neural Network for Soil Moisture Retrieval Using CYGNSS Reflectometry Data

Various remote sensing satellites can be used for extracting soil moisture (SM), each characterized by unique spatial and temporal resolutions. Missions such as Soil Moisture Active Passive (SMAP) have provided fresh insights into the storage of near-surface soil moisture through L-band radiometry, achieving a spatial resolution of 30–50 km and the full Earth coverage in 2-3 days. The demonstrated sensitivity of the L-band electromagnetic signal to the water content of observed targets and its significant penetration depth underscores the potential of Global Navigation Satellite System-Reflectometry (GNSS-R) techniques in diverse land applications. An illustrative example of this advancing application is evident in missions like the NASA's Cyclone GNSS (CyGNSS), originally designed to detect wind speed at sea in tropical cyclones measuring the Earth surface reflections of GNSS signals of opportunity.Within this context, the capability to retrieve soil moisture through the exploitation of GNSS-R reflections by Artificial Neural Networks has been confirmed in the literature (e.g., see [1] and [2]). In this paper, a sophisticated Artificial Neural Network (ANN) algorithm is used to explore the impact of additional auxiliary data able to account for other factors affecting the GNSS-R signal. They include topography, Above Ground Biomass (AGB), land use, roughness, soil texture, soil porosity, and dynamic variables like Vegetation Water Content (VWC) and Vegetation Optical Depth (VOD) from Soil Moisture Active Passive (SMAP). It also considers data such as Land Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI) and Solar Induced Fluorescence (SIF) from Moderate Resolution Imaging Spectroradiometer (MODIS). Moreover, the effect of using latitude/longitude as input on the performances of the algorithm is assessed. The study also aims at evaluating the impact of different stratification approaches, setting up different ANN’s in different geographical and landcover-based stratifications. To assess how these variables contribute to improving the accuracy of soil moisture retrieval, the datasets are collocated in space and time and resampled onto the EASE grid v2.0 projection at 25km resolution. The algorithm is subsequently trained and validated using target soil moisture values derived from SMAP L3 global daily products and in-situ measurements from the International Soil Moisture Network (ISMN). The work has been carried out in the framework of the ESA Scout 2 HydroGNSS mission development, expected to be launched at the end of 2024. Reference[1]       E. Santi et al., “Combining Cygnss and Machine Learning for Soil Moisture and Forest Biomass Retrieval in View of the ESA Scout Hydrognss Mission,” Sep. 2022, doi: 10.1109/IGARSS46834.2022.9884738.[2]       O. Eroglu, M. Kurum, D. Boyd, and A. C. Gurbuz, “High Spatio-Temporal Resolution CYGNSS Soil Moisture Estimates Using Artificial Neural Networks,” Remote Sensing 2019, Sep. 2019, doi: 10.3390/RS11192272.

In recent decades, GNSS Reflectometry (GNSS-R) has gained significant popularity for hydrological applications, including monitoring soil moisture (SM) and vegetation status. This innovative technique utilizes freely available L-band GPS signals reflected off the Earth's surface, providing high sensitivity to surface changes and an unprecedented spatio-temporal sampling rate.Most previous GNSS-R studies have focused on monitoring a single hydrological variable, such as SM or vegetation, often using deep learning algorithms. In this research, we assess the effectiveness of NASA’s GNSS-R mission (CYGNSS) observations for the simultaneous estimation of Vegetation Optical Depth (VOD) and SM. VOD measures the attenuation of microwave radiation as it passes through vegetation, providing insight into canopy water content and biomass, and reflecting the vegetation's density and moisture levels. We compare VOD and SM estimates from two independent methods across the contiguous United States (CONUS): a deep learning-based artificial neural network (ANN) and the L-MEB (L-band Microwave Emission of the Biosphere) model, a semi-empirical approach. Our SM and VOD estimates are validated against SMAP SM and SMOS VOD, respectively.Our findings indicate strong potential for CYGNSS observations in simultaneously monitoring VOD and SM, with the ANN outperforming the semi-empirical model. The spatial correlation between SMAP SM and CYGNSS SM estimates is approximately 0.6 for the semi-empirical model and 0.91 for the ANN. Both models capture daily SM variations well, with correlation coefficients of 0.74 for semi-empirical SM and 0.8 for ANN-predicted SM over the years 2020 and 2021.Similar to SM estimations, the ANN VOD predictions outperform those from the semi-empirical model. The ANN shows a high correlation (around 0.9) and a root mean square error (RMSE) of approximately 0.025 between its daily VOD estimates and SMOS VOD, whereas the daily VOD values from the semi-empirical model poorly track changes in SMOS VOD. Overall, our work underscores the effectiveness of CYGNSS observations for providing simultaneous L-band VOD and SM, as well as the superiority of the ANN over the semi-empirical model.Estimating soil moisture and vegetation optical depth simultaneously not only reduces the number of ancillary datasets, such as Vegetation Water Content (VWC), used to correct for vegetation attenuation in GNSS-R observations when estimating soil moisture separately, but also helps reduce potential errors that can arise from relying on these datasets. Additionally, the L-VOD derived from CYGNSS is particularly valuable for two reasons. First, it fills gaps in VOD data from existing L-band missions like SMAP and SMOS, ensuring continuity even if those missions encounter operational challenges. This highlights the importance of monitoring L-VOD through GNSS-R missions and emphasizes the value of our work. Second, VOD serves as a key indicator of water content in aboveground woody biomass, directly supporting the objectives of the upcoming GNSS-R mission, HydroGNSS.

A new SMAP soil moisture and vegetation optical depth product (SMAP-IB): Algorithm, assessment and inter-comparison

Abstract. Vegetation attenuates the microwave emission from the land surface. The strength of this attenuation is quantified in models in terms of the parameter vegetation optical depth (VOD) and is influenced by the vegetation mass, structure, water content, and observation wavelength. Earth observation satellite sensors operating in the microwave frequencies are used for global VOD retrievals, enabling the monitoring of vegetation at large scales. VOD has been used to determine above-ground biomass, monitor phenology, or estimate vegetation water status. VOD can be also used for constraining land surface models or modelling wildfires at large scales. Several VOD products exist, differing by frequency/wavelength, sensor, and retrieval algorithm. Numerous studies present correlations or empirical functions between different VOD datasets and vegetation variables such as the normalized difference vegetation index, leaf area index, gross primary production, biomass, vegetation height, or vegetation water content. However, an assessment of the joint impact of land cover, vegetation biomass, leaf area, and moisture status on the VOD signal is challenging and has not yet been done. This study aims to interpret the VOD signal as a multi-variate function of several descriptive vegetation variables. The results will help to select VOD at the most suitable wavelength for specific applications and can guide the development of appropriate observation operators to integrate VOD with large-scale land surface models. Here we use VOD from the Land Parameter Retrieval Model (LPRM) in the Ku, X, and C bands from the harmonized Vegetation Optical Depth Climate Archive (VODCA) dataset and L-band VOD derived from Soil Moisture and Ocean Salinity (SMOS) and Soil Moisture Active Passive (SMAP) sensors. The leaf area index, live-fuel moisture content, above-ground biomass, and land cover are able to explain up to 93 % and 95 % of the variance (Nash–Sutcliffe model efficiency coefficient) in 8-daily and monthly VOD within a multi-variable random forest regression. Thereby, the regression reproduces spatial patterns of L-band VOD and spatial and temporal patterns of Ku-, X-, and C-band VOD. Analyses of accumulated local effects demonstrate that Ku-, X-, and C-band VOD are mostly sensitive to the leaf area index, and L-band VOD is most sensitive to above-ground biomass. However, for all VODs the global relationships with vegetation properties are non-monotonic and complex and differ with land cover type. This indicates that the use of simple global regressions to estimate single vegetation properties (e.g. above-ground biomass) from VOD is over-simplistic.

Impact of vegetation water content information on soil moisture retrievals in agricultural regions: An analysis based on the SMAPVEX16-MicroWEX dataset

Improved estimation of vegetation water content and its impact on L-band soil moisture retrieval over cropland

Data from the Cyclone Global Navigation Satellite System (CYGNSS) mission augmented with a physical surface scattering model were analyzed to develop a semiempirical model, which consists of three main modeling components for soil moisture, vegetation, and surface roughness. CYGNSS data collected from March 2017 to March 2020 were collocated with the soil moisture data from the Soil Moisture Active Passive (SMAP) mission and climatology vegetation water content (VWC) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) data. The matchup data were binned as a function of soil moisture, VWC, and incidence angle. The CYGNSS data were calibrated using a coherent reflection equation to obtain an effective reflectivity. The response of CYGNSS effective reflectivity to soil moisture changes is consistent with the change of the Fresnel reflectivity based on Mironov’s soil dielectric constant model used by the SMAP and Soil Moisture Ocean Salinity (SMOS) missions for soil moisture retrieval. The CYGNSS effective reflectivity decreases approximately linearly (in dB) with respect to the NDVI-VWC. The estimated values of vegetation attenuation parameter ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$b$ </tex-math></inline-formula> ) agree with values published in the literature and are corroborated with the estimated values of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$b$ </tex-math></inline-formula> using the SMAP dual-polarized channel algorithm based on land cover types. A CYGNSS surface scattering map has been derived and reveals a mixed contribution of coherent and incoherent scattering effects and the effects of topography. The semiempirical model, leveraging two of the key modeling functions used by microwave radiometry, will pave the way for a synergistic use of reflectometry and radiometry data for multiparameter retrieval and development of consistent soil moisture products.

P-band radiometry for enhanced vegetation optical depth (VOD) and soil moisture retrieval in dense crop canopies

Understanding and reducing the uncertainties in the inversion of the first-order radiative transfer models at the L-band are important for the improved spaceborne retrievals of soil moisture (SM) and vegetation optical depth (VOD) over dense canopy. This article quantifies and compares the sensitivity of dual-channel inversion of the two-stream (2S) and $\tau $ – $\omega $ models and proposes a new inversion approach for simultaneous retrievals of SM, VOD, and vegetation-scattering albedo ( $\omega $ ) from a single satellite overpass. In particular, the inversion algorithm incorporates the information of the nearby spatial observations, assuming that the values of VOD and $\omega $ remain locally invariant, and constrains its solutions to high-resolution a priori physical/climatological knowledge of the retrieval variables. The results demonstrate that the uncertainty in the inversion of 2S model is slightly higher than the $\tau $ – $\omega $ model under noisy observations and remains homoscedastic for SM and $\omega $ , while grows heteroscedastically for higher VOD values due to the shape of the cost function. The results are validated using the SMAP data, the dense Mesonet SM network, the in situ measurements from the International SM Network (ISMN), and the derived VOD from the Moderate Resolution Imaging Spectroradiometer (MODIS)-normalized difference vegetation index (NDVI) over the state of Oklahoma in the United States. It is shown that the new approach can recover simultaneously high-resolution features of SM, VOD, and $\omega $ only from a single Soil Moisture Active Passive (SMAP) overpass, where the unbiased root-mean-squared error (ubRMSE) of SM and VOD is reduced by 30% and 70%, respectively, when compared with an unconstrained time-windowed inversion approach.

The potential of moderate-resolution imaging spectroradiometer (MODIS) time-series in monitoring crop condition was examined in the Weishan Irrigation Zone along a downstream reach of the Yellow River, China. First, MODIS-derived spectral indices, including the normalized difference water index (NDWI) and normalized difference vegetation index (NDVI), were employed to retrieve the crop physiological parameters (such as vegetation water content and dry matter) during the winter wheat growing season. NDWI was found to be less susceptible than NDVI to the directional effects induced by the changes of illumination and view angles. NDWI1640 and NDWI2130, which were based on the shortwave infrared (SWIR) 1640 nm and 2130 nm bands, showed great potential for vegetation water content (VWC) estimation and performed better than NDVI and NDWI1240. Both NDWI1640 and NDWI2130 had a significant positive correlation with dry matter content before the blooming stage. A significant negative correlation between NDWI2130 and dry matter content was found after the blooming stage. Second, NDWI was also employed to assess the canopy water deficit during the same period as a direct indicator of crop water content. The strong correlation between NDWI and soil moisture after the jointing stage showed that NDWI can also be utilized as an indicator of soil moisture changes. An integrated analysis of the temporal variation of NDWI and the observed VWC demonstrated that the wheat was under water stress to some extent during the later part of the growing period in 2006, and this was supported by a field experiment.

Reappraisal of SMAP inversion algorithms for soil moisture and vegetation optical depth

Microwave retrievals of soil moisture and vegetation optical depth with improved resolution using a combined constrained inversion algorithm: Application for SMAP satellite

Vegetation attenuates the microwave emission from the land surface. The strength of this attenuation is quantified in models in terms of the parameter Vegetation Optical Depth (VOD), and is influenced by the vegetation mass, structure, water content, and observation wavelength. Earth observation satellites operating in the microwave frequencies are used for global VOD retrievals, enabling the monitoring of vegetation status at large scales. VOD has been used to determine above-ground biomass, monitor phenology or estimate vegetation water status. VOD can be also used for constraining land surface models or modelling wildfires at large scale. Several VOD products exist differing by frequency/wavelength, sensor, and retrieval algorithm. Numerous studies present correlations or empirical functions between different VOD datasets and vegetation variables such as normalised difference vegetation index, leaf area index, gross primary production, biomass, vegetation height or vegetation water content. However, an assessment of the joint impact of land cover, vegetation biomass, leaf area, and moisture status on the VOD signal is challenging and has not yet been done. This study aims to interpret the VOD signal as a multi-variate function of several descriptive vegetation variables. The results will help to select certain VOD wavelengths for specific applications and can guide the development of appropriate observation operators to integrate VOD with large-scale land surface models. Here we use VOD from the Land Parameter Retrieval Model (LPRM) of Ku-, X- and C-bands of the harmonised VODCA dataset and level 3 L-band derived from SMOS and SMAP sensors. Within a multivariable regression random forest model for simulating these VOD signals, leaf area index, live-fuel moisture content, above-ground biomass, and land cover are able to explain up to 0.95 of the variance (coefficient of determination). Thereby, the variance in L-band VOD is reproduced spatially and for Ku-, X- and C-band VOD spatially as well as temporally. Analyses of accumulated local effects demonstrate that Ku-, X- and C-band VOD is mostly sensitive to leaf area index and L-band VOD to above-ground biomass. However, for all VODs the global relationships with vegetation properties are non-monotonic and complex and differ with land cover type. This indicates that the use of simple global regressions to estimate single vegetation properties (e.g. above-ground biomass) from VOD is over-simplistic.

Vegetation attenuates the microwave emission from the land surface. The strength of this attenuation is quantified in models in terms of the parameter Vegetation Optical Depth (VOD), and is influenced by the vegetation mass, structure, water content, and observation wavelength. Earth observation satellites operating in the microwave frequencies are used for global VOD retrievals, enabling the monitoring of vegetation status at large scales. VOD has been used to determine above-ground biomass, monitor phenology or estimate vegetation water status. VOD can be also used for constraining land surface models or modelling wildfires at large scale. Several VOD products exist differing by frequency/wavelength, sensor, and retrieval algorithm. Numerous studies present correlations or empirical functions between different VOD datasets and vegetation variables such as normalised difference vegetation index, leaf area index, gross primary production, biomass, vegetation height or vegetation water content. However, an assessment of the joint impact of land cover, vegetation biomass, leaf area, and moisture status on the VOD signal is challenging and has not yet been done. This study aims to interpret the VOD signal as a multi-variate function of several descriptive vegetation variables. The results will help to select certain VOD wavelengths for specific applications and can guide the development of appropriate observation operators to integrate VOD with large-scale land surface models. Here we use VOD from the Land Parameter Retrieval Model (LPRM) of Ku-, X- and C-bands of the harmonised VODCA dataset and level 3 L-band derived from SMOS and SMAP sensors. Within a multivariable regression random forest model for simulating these VOD signals, leaf area index, live-fuel moisture content, above-ground biomass, and land cover are able to explain up to 0.95 of the variance (coefficient of determination). Thereby, the variance in L-band VOD is reproduced spatially and for Ku-, X- and C-band VOD spatially as well as temporally. Analyses of accumulated local effects demonstrate that Ku-, X- and C-band VOD is mostly sensitive to leaf area index and L-band VOD to above-ground biomass. However, for all VODs the global relationships with vegetation properties are non-monotonic and complex and differ with land cover type. This indicates that the use of simple global regressions to estimate single vegetation properties (e.g. above-ground biomass) from VOD is over-simplistic.

Vegetation attenuates the microwave emission from the land surface. The strength of this attenuation is quantified in models in terms of the parameter Vegetation Optical Depth (VOD), and is influenced by the vegetation mass, structure, water content, and observation wavelength. Earth observation satellites operating in the microwave frequencies are used for global VOD retrievals, enabling the monitoring of vegetation status at large scales. VOD has been used to determine above-ground biomass, monitor phenology or estimate vegetation water status. VOD can be also used for constraining land surface models or modelling wildfires at large scale. Several VOD products exist differing by frequency/wavelength, sensor, and retrieval algorithm. Numerous studies present correlations or empirical functions between different VOD datasets and vegetation variables such as normalised difference vegetation index, leaf area index, gross primary production, biomass, vegetation height or vegetation water content. However, an assessment of the joint impact of land cover, vegetation biomass, leaf area, and moisture status on the VOD signal is challenging and has not yet been done. This study aims to interpret the VOD signal as a multi-variate function of several descriptive vegetation variables. The results will help to select certain VOD wavelengths for specific applications and can guide the development of appropriate observation operators to integrate VOD with large-scale land surface models. Here we use VOD from the Land Parameter Retrieval Model (LPRM) of Ku-, X- and C-bands of the harmonised VODCA dataset and level 3 L-band derived from SMOS and SMAP sensors. Within a multivariable regression random forest model for simulating these VOD signals, leaf area index, live-fuel moisture content, above-ground biomass, and land cover are able to explain up to 0.95 of the variance (coefficient of determination). Thereby, the variance in L-band VOD is reproduced spatially and for Ku-, X- and C-band VOD spatially as well as temporally. Analyses of accumulated local effects demonstrate that Ku-, X- and C-band VOD is mostly sensitive to leaf area index and L-band VOD to above-ground biomass. However, for all VODs the global relationships with vegetation properties are non-monotonic and complex and differ with land cover type. This indicates that the use of simple global regressions to estimate single vegetation properties (e.g. above-ground biomass) from VOD is over-simplistic.