A temporal polarization ratio algorithm for calibration-free retrieval of soil moisture at L-band
Introduction
Surface soil moisture (SM) is an essential hydrologic state variable in the Earth system as it controls evapotranspiration (Wetzel and Chang, 1987; Vivoni et al., 2008; Jung et al., 2010), affects space-time patterns of mesoscale overland rainfall processes (Guillod et al., 2015; Lin et al., 2017a, Lin et al., 2017b), and thus modulates the global water and energy cycle (Seneviratne et al., 2010; Babaeian et al., 2019). It is shown that surface SM, all by itself, contains adequate information about the land surface net water flux and hence variation of total subsurface water storage (Crow et al., 2017; Sadeghi et al., 2019, Sadeghi et al., 2020).
Today, L-band (1–2 GHz) microwave radiometry offers an all-sky capability for frequent observations of SM on a global scale (Njoku and Li, 1999) and has been the cornerstone of two satellite missions including the ESA's Soil Moisture and Ocean Salinity (SMOS, 2009) (Kerr et al., 2010) and NASA's Soil Moisture Active Passive (SMAP, 2015) (Entekhabi et al., 2014). Both missions have provided new opportunities to measure global SM at the topmost ~5 cm depth with a 3-day maximum revisiting time and have significantly extended our understanding of global water, energy, and carbon cycle (McColl et al., 2017a; Feldman et al., 2018b; Wigneron et al., 2020; Tian et al., 2018).
At L-band, the surface emission, from the top ~5 cm of soil layer and the aboveground vegetation, is often obtained from a zeroth-order approximation of the radiative transfer equation, known as the τ–ω model (Mo et al., 1982; Tsang et al., 1985; Ulaby et al., 2014). This model consists of three components including direct upward soil emission, direct upward vegetation emission, and reflected downward vegetation emission. These emission signals are parameterized by the rough surface reflectivity r, vegetation single scattering albedo ω, and vegetation transmissivity γ.
The single scattering albedo, defined as the ratio of scattering to extinction coefficient (Ulaby et al., 1983; Wigneron et al., 1993), accounts for the lumped contribution of scattering in the total vegetation attenuation and is often called the “effective scattering albedo” (Kurum et al., 2012). The albedo parameter is commonly prespecified in the inversion algorithms through experimental (Wigneron et al., 2004) or numerical (Ferrazzoli et al., 2002) calibrations using ground-based (Wigneron et al., 2007; Saleh et al., 2007; Yan et al., 2015; Fernandez-Moran et al., 2017b) and/or spaceborne (De Lannoy et al., 2013, De Lannoy et al., 2014) observations. In practice, for the SMAP and SMOS official products, the global values are determined empirically as a function of land cover/vegetation types (Wigneron et al., 2017). Nevertheless, comparisons between zeroth- and high-order radiative transfer models demonstrate that the commonly used effective scattering albedo in τ-ω model is generally smaller than its theoretical values because of significant scattering by the canopy trunks and branches (Kurum, 2013; Schwank et al., 2018). More recently, attempts have been made to estimate ω directly in the inversion process (Vittucci et al., 2017), assuming that it remains invariant over a window of time (Konings et al., 2016, Konings et al., 2017; Feldman et al., 2018a; Karthikeyan et al., 2019) or space (Gao et al., 2020b). However, due to the ill-posed nature of the inversion, setting ω as an additional unknown parameter may lead to extra uncertainties in the retrievals.
The slanted vegetation transmissivity γ is directly linked to the vegetation optical depth (VOD), denoted by τ, which is determined differently in the family of single channel algorithms (SCA) (Jackson, 1993; Bindlish et al., 2015) and the dual channel algorithms (DCA) (Njoku and Entekhabi, 1996; Njoku et al., 2003; Shi et al., 2006). In the SCA, currently used to produce the official SMAP products, climatology of the normalized difference vegetation index (NDVI) (Chan et al., 2013) from the Moderate Resolution Imaging Spectroradiometer (MODIS) (Justice et al., 1998) is used to estimate the vegetation water content (VWC) (Jackson et al., 1999). The VOD is then linked to VWC through a set of empirical equations (Jackson and O'Neill, 1989; Jackson and Schmugge, 1991), which paves the way for direct estimation of SM from the τ-ω model. However, DCA attempts to directly retrieve VOD and SM simultaneously using some ancillary a priori data. For example, in the SMOS L-band Microwave Emission of the Biosphere (L-MEB) model (Kerr et al., 2006), the leaf area index (LAI) (Chen and Black, 1992) is utilized to obtain an a priori estimate of the VOD values (Kerr et al., 2012). A quadratic regularization term is then used to regress the retrievals toward this a priori estimate, in a Bayesian setting, for reducing uncertainties in the retrievals of both SM and VOD (Pardé et al., 2004). More recently, a box-constrained DCA inversion paradigm was also introduced indicating that constraining the SM and VOD retrievals to their physical/climatological range of variability not only reduces the uncertainties but also yields higher-resolution retrievals (Ebtehaj and Bras, 2019; Gao et al., 2020a). Although a few new algorithms such as SMOS-IC (Fernandez-Moran et al., 2017a) and multi-temporal DCA (Konings et al., 2016) are able to retrieve both VOD and SM independent of any a priori information, it is still challenging for classical DCA and SCA to properly account for VOD in the inversion process.
Soil moisture is usually obtained from the soil dielectric constant (Mironov et al., 2009) that is related to the smooth surface reflectivity through the Fresnel equations. However, the incoherent surface emission is explained by the rough surface reflectivity, which needs to be linked to the smooth surface reflectivity. This linkage has been explained by physically-based numerical models (Li et al., 2000; Huang et al., 2010; Huang and Tsang, 2012; Tsang et al., 2012), analytical models (Wu et al., 2001; Chen et al., 2003), and semi-empirical models (Choudhury et al., 1979; Wang and Choudhury, 1981; Wigneron et al., 2001). The two former approaches usually require parameterization of surface emissivity via Monte Carlo simulations under various surface characteristics (Shi et al., 2002; Chen et al., 2009; Tsang et al., 2013; Cui et al., 2016) and have not yet been widely used for spaceborne retrievals due to the lack of estimates of geometric roughness parameters at the scale of satellite footprint (Loew, 2008). However, the semi-empirical schemes, such as the Q/H model (Lawrence et al., 2013; Peng et al., 2017; Karthikeyan et al., 2017), are commonly deployed for the L-band SM retrieval owing to their relatively high degree of accuracy, simple structure, and low computational cost.
Despite significant efforts made, there are still discrepancies among the prescribed or estimated values of the vegetation parameters on a global scale (De Lannoy et al., 2014; Konings et al., 2016, Konings et al., 2017) and proper parameterization of surface roughness, over the satellite footprint, is not yet well-understood (Neelam et al., 2020). Consequently, the associated uncertainties will propagate into the SM retrievals (Theis and Blanchard, 1988; Grant et al., 2008). For example, the usage of 13-year VOD climatology from MODIS NDVI observations in SMAP SCA can misrepresent real-time dynamics of VOD and leads to significant errors in the SCA SM retrievals (Dong et al., 2018). Moreover, because ω and surface roughness parameter are commonly obtained as a function of global land cover maps (Wigneron et al., 2007; O'Neill et al., 2017), any uncertainty in land cover classification (Fritz et al., 2011; Congalton et al., 2014) can add additional errors to the SM retrievals.
Besides the vegetation parameters and surface roughness, the quality of SM retrievals is moderately affected by the biases in the effective soil temperature (De Rosnay et al., 2006). An accurate estimation of the effective soil temperature often requires high-quality measurements of soil temperature and moisture profiles (Choudhury et al., 1982). Several simplified schemes have been proposed for estimating the effective temperature at a satellite footprint scale (Choudhury et al., 1982; Wigneron et al., 2001, Wigneron et al., 2008; Holmes et al., 2006; Shaoning et al., 2014). An inter-comparison of these schemes (Lv et al., 2016) based on the soil temperature reanalysis from Modern-Era Retrospective analysis for Research and Applications (MERRA) (Gelaro et al., 2017) indicates an up to 5 K uncertainty in the resulted effective soil temperature data sets. This uncertainty can lead to a non-negligible bias in the satellite SM retrievals.
In this paper, to cope with the above-mentioned uncertainties, a new approach is proposed for calibration-free retrieval of SM at L-band, called the “Temporal Polarization Ratio Algorithm” (TPRA). The term “calibration-free” here means that the TPRA does not depend on any empirical values for vegetation or surface roughness parameters, which are inevitable in the conventional algorithms. The main advantage of the new approach is that its SM retrievals are independent of vegetation parameters, surface roughness, and thus land cover maps. Assuming that the soil roughness and vegetation parameters are time-invariant over a sufficiently short window of time, TPRA uses ratios of temporal differences of polarized emissivity values, which are only a function of temporal changes of surface reflectivity. As a result, TPRA is not only independent of the effective scattering albedo, VOD, and surface roughness but also remains relatively robust to systematic emissivity biases arising from the effective soil temperature errors. Nonetheless, this new algorithm is unable to retrieve SM when the observed emissivity does not change significantly over time.
The organization of the paper is as follows: Section 2 describes TPRA following a brief review of the τ–ω radiative transfer model and lays out the algorithmic steps to implement TPRA for satellite observations. In section 3, we first examine the performance of TPRA through a series of controlled numerical experiments and then implement it using level-III SMAP brightness temperatures over Australia. The results are compared with SMAP official products and validated against in-situ measurements from the OzNet monitoring network. In section 4, we conclude and discuss potential future research directions.
Section snippets
The τ–ω model
The τ–ω model (Mo et al., 1982) describes the upwelling surface emission ep, at horizontal and vertical polarization p ∈ {H, V}, by three terms including direct upward soil emission (1 − rp)γ, direct upward vegetation emission (1 − ω)(1 − γ) and reflected downward vegetation emission (1 − ω)(1 − γ)rpγ:where the polarized emissivity is ep = Tbp/Ts, Tbp is the polarized brightness temperature and Ts denotes the effective soil temperature assuming that the soil and canopy
Algorithm validation
In this section, we first examine the statistical properties of polarization ratios α and β. Then we conduct a series of numerical experiments to understand and quantify the advantages of TPRA over DCA under random errors in the parameterization of soil roughness and vegetation radiometric properties as well as systematic biases in emissivity. Through numerical experiments, we also elaborate on the effects of temporal changes of VOD and emissivity on the accuracy of TPRA. In addition, we
Conclusions and discussion
In this paper, a new algorithm, called Temporal Polarization Ratio Algorithm (TPRA), is proposed to retrieve SM from microwave emissivity observations. This method relies on polarized emissivity values at two points in time. The main assumptions are: (1) land surface emissivity changes sufficiently (>0.01) and (2) surface roughness, scattering albedo, and VOD do not change markedly within the time interval between the two observation points. In this setting, the proposed method is capable of
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The supporting grants from the Future Investigators in NASA Earth and Space Science (FINESST, 80NSSC19K1333) through Dr. A. Leidner and the NASA's Terrestrial Hydrology Program (THP, 80NSSC18K1528) through Dr. J. Entin are highly acknowledged. The first author also appreciates early discussion with Prof. Dara Entekhabi at Massachusetts Institute of Technology and Dr. Xiwu Zhan at NOAA NESDIS Center for Satellite Applications and Research.
References (113)
- et al.
Evaluating soil moisture retrievals from esa’s smos and nasa’s smap brightness temperature datasets
Remote Sens. Environ.
(2017) - et al.
Uncertainty quantification of geos-5 l-band radiative transfer model parameters using bayesian inference and smos observations
Remote Sens. Environ.
(2014) - et al.
A physically constrained inversion for high-resolution passive microwave retrieval of soil moisture and vegetation water content in l-band
Remote Sens. Environ.
(2019) - et al.
Characterization of higher-order scattering from vegetation with smap measurements
Remote Sens. Environ.
(2018) - et al.
A new calibration of the effective scattering albedo and soil roughness parameters in the smos sm retrieval algorithm
Int. J. Appl. Earth Obs. Geoinf.
(2017) - et al.
Microwave retrievals of soil moisture and vegetation optical depth with improved resolution using a combined constrained inversion algorithm: application for smap satellite
Remote Sens. Environ.
(2020) - et al.
Vegetation effects on the microwave emission of soils
Remote Sens. Environ.
(1991) - et al.
Vegetation water content mapping using landsat data derived normalized difference water index for corn and soybeans
Remote Sens. Environ.
(2004) - et al.
Four decades of microwave satellite soil moisture observations: part 1. a review of retrieval algorithms
Adv. Water Resour.
(2017) - et al.
Simultaneous retrieval of global scale vegetation optical depth, surface roughness, and soil moisture using x-band amsr-e observations
Remote Sens. Environ.
(2019)
Vegetation optical depth and scattering albedo retrieval using time series of dual-polarized l-band radiometer observations
Remote Sens. Environ.
L-band vegetation optical depth and effective scattering albedo estimation from smap
Remote Sens. Environ.
Quantifying scattering albedo in microwave emission of vegetated terrain
Remote Sens. Environ.
Effective tree scattering and opacity at l-band
Remote Sens. Environ.
Impact of surface heterogeneity on surface soil moisture retrievals from passive microwave data at the regional scale: The upper danube case
Remote Sens. Environ.
A reappraisal of global soil effective temperature schemes
Remote Sens. Environ.
Evaluation of multi-frequency bare soil microwave reflectivity models
Remote Sens. Environ.
Passive microwave remote sensing of soil moisture
J. Hydrol.
Global-scale surface roughness effects at l-band as estimated from smos observations
Remote Sens. Environ.
Reappraisal of the roughness effect parameterization schemes for l-band radiometry over bare soil
Remote Sens. Environ.
Estimates of surface soil moisture under grass covers using l-band radiometry
Remote Sens. Environ.
Investigating soil moisture–climate interactions in a changing climate: a review
Earth Sci. Rev.
Inter-comparison of microwave satellite soil moisture retrievals over the Murrumbidgee basin, southeast Australia
Remote Sens. Environ.
Passive microwave sensing of soil moisture content: the effects of soil bulk density and surface roughness
Remote Sens. Environ.
Ground, proximal, and satellite remote sensing of soil moisture
Rev. Geophys.
Global soil moisture from the aquarius/sac-d satellite: description and initial assessment
IEEE Geosci. Remote Sens. Lett.
Smap Level 3 Passive Soil Moisture Product Specification Document
Ancillary Data report for vegetation water content. SMAP Project Document
Defining leaf area index for non-flat leaves
Plant Cell Environ.
Emission of rough surfaces calculated by the integral equation method with comparison to three-dimensional moment method simulations
IEEE Trans. Geosci. Remote Sens.
A parameterized surface emission model at l-band for soil moisture retrieval
IEEE Geosci. Remote Sens. Lett.
Effect of surface roughness on the microwave emission from soils
J. Geophys. Res. Oceans
A parameterization of effective soil temperature for microwave emission
J. Geophys. Res. Oceans
Global land cover mapping: a review and uncertainty analysis
Remote Sens.
Upscaling sparse ground-based soil moisture observations for the validation of coarse-resolution satellite soil moisture products
Rev. Geophys.
Estimating annual water storage variations in medium-scale (2000–10 000 km 2) basins using microwave-based soil moisture retrievals
Hydrol. Earth Syst. Sci.
An algorithm for retrieving soil moisture using l-band h-polarized multiangular brightness temperature data
IEEE Geosci. Remote Sens. Lett.
Selection of soil attributes datasets for the smap mission. SMAP Science Document# D-53058, 1
Global calibration of the geos-5 l-band microwave radiative transfer model over nonfrozen land using smos observations
J. Hydrometeorol.
Parameterizations of the effective temperature for l-band radiometry. inter-comparison and long term validation with smosrex field experiment
Microwave dielectric behavior of wet soil-part ii: dielectric mixing models
IEEE Trans. Geosci. Remote Sens.
The error structure of the smap single and dual channel soil moisture retrievals
Geophys. Res. Lett.
Smap Handbook–Soil Moisture Active Passive: Mapping Soil Moisture and Freeze/Thaw from Space
Moisture pulse-reserve in the soil-plant continuum observed across biomes
Nat. Plants
Smos-ic: an alternative smos soil moisture and vegetation optical depth product
Remote Sens.
Simulating l-band emission of forests in view of future satellite applications
IEEE Trans. Geosci. Remote Sens.
Highlighting continued uncertainty in global land cover maps for the user community
Environ. Res. Lett.
A spatially constrained multichannel algorithm for inversion of a first-order microwave emission model at l-band
The modern-era retrospective analysis for research and applications, version 2 (merra-2)
J. Clim.
Calibration of the l-meb model over a coniferous and a deciduous forest
IEEE Trans. Geosci. Remote Sens.
Cited by (10)
Improved estimation of vegetation water content and its impact on L-band soil moisture retrieval over cropland
2023, Journal of HydrologyCitation Excerpt :Finally, 10-day VOD climatology is produced from MODIS NDVI climatology composites based on the relationship established in the previous two steps. Since this (static) NDVI climatology cannot capture reported phenological shifts associated with global warming or the impact of inter-annual vegetation variations due to drought and fluvial conditions (Colliander et al., 2019; Fan et al., 2020; Gao et al., 2020a; Shen et al., 2022), the direct use of NDVI-derived VOD climatology in SM retrieval algorithms can lead to reduced SM accuracy (Judge et al., 2021). This raises two questions: (1) can dynamic VIs be used to obtain VWC and VOD estimates that significantly improve SM retrievals relative to the current SMAP approach of utilizing an NDVI climatology to estimate VWC and VOD?
Identification of shallow groundwater in arid lands using multi-sensor remote sensing data and machine learning algorithms
2022, Journal of HydrologyCitation Excerpt :Earlier attempts heavily relied on DEM-derived variables such as elevation, slope, flow direction, flow accumulation, and curvature (Naghibi et al., 2020; Pande et al., 2020), but less so on other relevant factors (e.g., soil moisture and vegetation intensity) that correlate well with SGW occurrences (Alshehri et al., 2020). Remote sensing–based datasets have been successfully used to map the spatial and temporal variations of soil moisture (Gao et al., 2020), vegetation cover (Costa et al., 2020), groundwater storage (Abdelmalik and Abdelmohsen, 2019; AbouAly et al., 2021; Das et al., 2018; Sahour et al., 2020a), landforms and land surface temperature (Duan et al., 2019) over extensive areas. In this study, we generate these products to identify SGW occurrences over the entire WD.
Error and uncertainty characterization of soil moisture and VOD retrievals obtained from L-band SMAP radiometer
2022, Remote Sensing of EnvironmentCitation Excerpt :The surface roughness effects are minimal compared to ω effects in the L-band TB observations (Fernandez-Moran et al., 2017b), and the effect of roughness depends on the type of vegetation and soil texture conditions (Parrens et al., 2017; Parrens et al., 2016). A strong influence of the ω parameter compared to the h parameter on SM retrievals is observed in some experimental studies (Fernandez-Moran et al., 2017b; Gao et al., 2020b). To circumvent these issues, parameter-free SM retrievals are in development (Gao et al., 2020b; Zeng et al., 2020).
A deep neural network based SMAP soil moisture product
2022, Remote Sensing of EnvironmentCitation Excerpt :Due to the importance of VOD in monitoring the aboveground biomass (Tian et al., 2018; Fan et al., 2019; Wigneron et al., 2020), DCA has received increasing attention in the community and is currently the baseline algorithm to generate official SMAP SM products (Chan et al., 2016; O'Neill et al., 2020c). Different variants of DCA have been proposed through the incorporation of a priori reanalysis data (Pardé et al., 2004; Wigneron et al., 2007; Fernandez-Moran et al., 2017; Li et al., 2022), multiple spatiotemporal observations (Konings et al., 2016, 2017; Ebtehaj and Bras, 2019; Gao et al., 2020a, 2020c, 2020b), and multi-angular and multi-frequency observations (Zhao et al., 2021). One exception is the land parameter retrieval model (LPRM) (Owe et al., 2001) that analytically resolves VOD using surface emissivity values to reduce the computational cost (Meesters et al., 2005).
A new SMAP soil moisture and vegetation optical depth product (SMAP-IB): Algorithm, assessment and inter-comparison
2022, Remote Sensing of EnvironmentCitation Excerpt :More recently, a modified DCA algorithm (MDCA) was also proposed (O'Neill et al., 2020) in which the cost function is augmented by an additional regularization term incorporating a priori information from NDVI to prevent noise amplification. However, a priori information should be used with caution as it: i) may bring uncertainty to the retrievals (Gao et al., 2020c; Wigneron et al., 2017); ii) makes the final product not independent of it by integrating its content in a hidden way (Fernandez-Moran et al., 2017b; Wigneron et al., 2021). As a consequence, the direct use of VOD climatology derived from MODIS (Moderate Resolution Imaging Spectroradiometer) NDVI to represent the L-VOD value (τ) in SMAP SCA could fail to represent the real-time dynamics of L-VOD and cause subsequently errors in the SCA SM retrievals (Gao et al., 2020c; Dong et al., 2018; Zwieback et al., 2018).