Reviews and syntheses: Recent advances in microwave remote sensing in support of terrestrial carbon cycle science in Arctic–boreal regions

Spaceborne microwave remote sensing (300 MHz–100 GHz) provides a valuable method for characterizing environmental changes, especially in Arctic–boreal regions (ABRs) where ground observations are generally spatially and temporally scarce. Although direct measurements of carbon fluxes are not feasible, spaceborne microwave radiometers and radar can monitor various important surface and near-surface variables that affect terrestrial carbon cycle processes such as respiratory carbon dioxide (CO₂) fluxes; photosynthetic CO₂ uptake; and processes related to net methane (CH₄) exchange including CH₄ production, transport and consumption. Examples of such controls include soil moisture and temperature, surface freeze–thaw cycles, vegetation water storage, snowpack properties and land cover. Microwave remote sensing also provides a means for independent aboveground biomass estimates that can be used to estimate aboveground carbon stocks. The microwave data record spans multiple decades going back to the 1970s with frequent (daily to weekly) global coverage independent of atmospheric conditions and solar illumination. Collectively, these advantages hold substantial untapped potential to monitor and better understand carbon cycle processes across ABRs. Given rapid climate warming across ABRs and the associated carbon cycle feedbacks to the global climate system, this review argues for the importance of rapid integration of microwave information into ABR terrestrial carbon cycle science.

1 Introduction

Northern regions host two important terrestrial biomes, the boreal and the Arctic (hereafter called Arctic–boreal regions, ABRs). Belowground carbon (C) stocks in ABRs comprise 30 %–40 % of the planetary terrestrial carbon and, as such, understanding changes in carbon cycle processes in this vast region has global importance (Pan et al., 2011; Tarnocai et al., 2009). Arctic–boreal regions store substantial quantities of belowground C due to their inherently slow decomposition rates, largely attributable to cold temperatures (Ravn et al., 2020). A large portion of ABRs is underlain by permafrost (perennially frozen ground), which contains approximately half of the world's belowground C stocks (1672 Pg C in the top 3 m of soil; Tarnocai et al., 2009; van Huissteden and Dolman, 2012). In contrast, C stocks in aboveground biomass in ABRs are relatively trivial. For example, although the boreal biome constitutes the second-largest terrestrial biome with a third of the world's forested area, it contains only approximately 15 % of the global forest aboveground biomass (FAO, 2001; Pan et al., 2011, 2013; Carreiras et al., 2017) but 32 % of the global forest carbon stocks (Pan et al., 2011).

Above- and belowground C pools in ABRs are vulnerable to climate change (Grosse et al., 2011). Arctic–boreal regions are warming at a disproportional rate compared to the rest of the planet with potential feedbacks to the global climate system (Box et al., 2019; Derksen et al., 2019; IPCC, 2019; Rantanen et al., 2022). In conjunction with warming temperatures, ABRs are experiencing altered precipitation regimes (Callaghan et al., 2011; Bokhorst et al., 2016; Dolant et al., 2018); permafrost thawing and deepening of the hydrologically and biogeochemically active layer (Liljedahl et al., 2016; Miner et al., 2022); spatially variable vegetation responses associated with “greening” and “browning” (i.e., increasing and decreasing productivity, respectively; Sulla-Menashe et al., 2018; Myers-Smith et al., 2020); earlier spring thaw; later freeze-up and lengthening of the growing season (Kimball et al., 2004a; Euskirchen et al., 2006; Kim et al., 2012); modifications of land cover (Wang et al., 2020); and intensifying disturbance regimes such as fire, drought, pervasive insect infestation and disease (Peng et al., 2011; Yi et al., 2013; Foster et al., 2022). Although ongoing warming has the potential to enhance photosynthesis and plant growth across ABRs, which would also increase aboveground C storage (Sturm et al., 2005; McMahon et al., 2010; Myers-Smith et al., 2020), the vegetation response to climate change is variable and complex. Furthermore, warmer air and soil temperatures enhance soil organic matter decomposition and the subsequent release of carbon dioxide (CO₂) via respiration (Schädel et al., 2016). Large uncertainties remain in terrestrial biosphere models used to estimate CO₂ and CH₄ fluxes in ABRs (Tei and Sugimoto, 2020; Fisher et al., 2018), including the amount of CO₂ and CH₄ released during winter (Natali et al., 2019; Zona et al., 2015). If increases in ecosystem respiration exceed those of photosynthetic CO₂ uptake from enhanced plant growth, ABRs may shift from a weak net CO₂ sink to a net CO₂ source, thereby generating a potentially non-negligible, positive feedback to the global climate system. This potential change underlines the importance of understanding changes in ABR C pools and fluxes (Hayes et al., 2011; Schuur et al., 2015; Gauthier et al., 2015; Virkkala et al., 2021).

The vastness and remoteness of ABRs make in situ observations challenging and costly. For example, although FLUXNET, the global initiative of tower-based eddy covariance flux observations (Baldocchi et al., 2001; Pastorello et al., 2020), is the most broadly used reference for C-flux measurements at the ecosystem scale, ABRs are notoriously underrepresented in this network (Fig. 1; Baldocchi et al., 2001; Pastorello et al., 2020; Pallandt et al., 2022). Satellite remote sensing and terrestrial biosphere models show promise for monitoring land surface–atmosphere interactions across ABRs (Fisher et al., 2018; Lees et al., 2018). Although direct measurement of C fluxes is not yet possible through remote sensing, it is possible to use spaceborne sensors to monitor variables and ecosystem structural parameters that exert control over various ecosystem processes (J. Du et al., 2019). In the last decade, radiometers sensitive to wavelengths in the visible and infrared portions of the electromagnetic spectrum have been widely used to support C-cycle science in ABRs (Turner et al., 2004; Mao et al., 2016; Lees et al., 2018; Xiao et al., 2019). Visible and infrared remote sensing observations have been used in C-cycle science to map spectral vegetation indices as proxies for vegetation abundance and productivity (e.g., normalized difference vegetation index, NDVI; Tucker, 1979; Lees et al., 2018; S. Du et al., 2019), forest disturbance (e.g., fire, tree mortality; Kim et al., 2012), land cover (Kimball et al., 2009), vegetation structure (e.g., light detection and ranging, lidar; Xiao et al., 2019), snow cover extent (Hori et al., 2017), land surface temperature (Sitch et al., 2007; Xiao et al., 2019), albedo (Xiao et al., 2019), solar-induced chlorophyll fluorescence (Wohlfahrt et al., 2018; Magney et al., 2019), and atmospheric CO₂ concentration (Buchwitz et al., 2007; Tu et al., 2020; Lorente et al., 2021). However, spaceborne visible and infrared radiometers are affected by signal degradation from atmospheric effects, have minimal signal penetration depth in vegetation and depend on solar illumination meaning they are restricted to daytime, clear-sky observations (Kim et al., 2012).

Figure 1 Permafrost extent (Brown et al., 2002) and distribution of eddy covariance sites where ecosystem fluxes are monitored continuously (Baldocchi et al., 2001; Pallandt et al., 2022; dot size represents the number of years of available data). The Arctic biome is delineated following the Conservation of Arctic Flora and Fauna working group of the Arctic Council, and the boreal biome is delineated following Potapov et al. (2008). Permafrost extent is estimated in percent areal coverage: continuous (> 90 %–100 % areal extent), discontinuous (> 50 %–90 %), sporadic (10 %–50 %) and isolated patches (< 10 %).

Spaceborne microwave remote sensing can be used in synergy with visible and infrared radiometers to maximize the benefits of a wider frequency span in the electromagnetic spectrum (Sitch et al., 2007; Kimball et al., 2009; Arslan et al., 2011; Kim et al., 2012; Xiao et al., 2019). Microwaves encompass electromagnetic radiation within the frequency range of 0.3–300 GHz, which corresponds to wavelengths from 1 mm to 1 m. Microwave remote sensing provides several advantages such as its relative insensitivity to atmospheric attenuation, cloud cover and solar illumination at frequencies below 100 GHz, which is essential during winter months in regions undergoing polar night (Sitch et al., 2007; J. Du et al., 2019). The spatial resolution of microwave imagery ranges from coarse (several-kilometer scale) to fine (meter scale) with varying temporal resolutions that can reach more than one observation per day. Microwave remote sensing is particularly suitable for ABRs since the signal penetration depth range allows for the retrieval of (1) volumetric information such as certain snow properties including density and microstructure (e.g., Nagler and Rott, 2000; Takala et al., 2011; Lievens et al., 2019) and vegetation optical depth (VOD), which relates to aboveground biomass and vegetation liquid water content (Konings et al., 2017, 2019; Mialon et al., 2020), and (2) near-surface information on variables such as soil moisture (Kerr et al., 2012; Colliander et al., 2017) and the freeze–thaw state (Kim et al., 2012; Roy et al., 2015; Rautiainen et al., 2016; Derksen et al., 2017; Prince et al., 2019) because of its sensitivity to liquid water.

This paper aims to introduce to the C-cycle science community the potential of spaceborne microwave remote sensing to help overcome some of the challenges specifically posed by ABRs for terrestrial C-cycle science and monitoring. We focus on the main vertical C fluxes between the land surface and the atmosphere, specifically, respiratory CO₂ fluxes; photosynthetic CO₂ uptake; and processes related to net methane (CH₄) fluxes governed by CH₄ production, transport and consumption. After summarizing the principles of microwave remote sensing (Sect. 2), we review how spaceborne microwave remote sensing can be exploited to help monitor key variables important for CO₂ and CH₄ fluxes in ABRs. These include soil moisture (Sect. 3.1) and temperature (Sect. 3.2), the surface freeze–thaw status (Sect. 3.3), aboveground biomass (Sect. 3.4), the vegetation water status (Sect. 3.5), land cover (Sect. 3.6), and snow cover (Sect. 3.7). For each variable, we will first explain how the variable at least partially governs C exchanges, outline the potential of microwave remote sensing to monitor the key variable and then introduce presently available microwave remote sensing products related to each variable.

2 Principles of microwave remote sensing

Microwave remote sensing can be used passively, by measuring the natural microwave emission from the planetary surface (using radiometers; receiving only), or actively, by measuring the backscattering of a previously emitted signal (using radar instruments; emitting and receiving).

Radiometers measure the natural microwave emission coming from the planetary surface. This emission is quantified as the brightness temperature (T_B; Fig. 2b), which corresponds to the temperature of a blackbody delivering the same luminance as the studied surface. Brightness temperature is related to the physical temperature of a surface ($T_{\mathrm{B}}=e\cdot T$) through emissivity (e; unitless). Emissivity is an inherent material property ranging from 0, for a perfectly non-emitting material, to 1, for a purely emitting material (blackbody). The microwave electric field generally displays a preferred orientation, called the microwave polarization, and is often related to the geometric structure of the source or target (Ulaby et al., 1982, 1983). The spatial resolution of passive microwave radiometers is generally coarser than many visible and infrared radiometers owing to the longer wavelengths and practical restrictions of the receiving antenna dimensions on spacecraft (Jenson, 2006). Spaceborne passive microwave radiometers typically have spatial resolutions in the 10–100 km range. Although passive microwave spatial resolutions are often too coarse to capture small-scale landscape heterogeneity (meter to kilometer scale), it is not an obstacle for regional or global applications of terrestrial biosphere models commonly run at spatial resolutions of several kilometers. The temporal resolution of most microwave instruments is generally less than 2–3 d, and in polar regions several overpasses a day may be achieved. The fine temporal resolution might be more critical for terrestrial biosphere models than good spatial resolution since computational power limits the spatial resolution of such models (Washington et al., 2009; Schär et al., 2020).

Radars emit electromagnetic waves to calculate the backscattering coefficient (σ; Fig. 2b) of a target area from the power ratio between the emitted and returned pulse (Ulaby et al., 1982, 1986). The return signal of a radar carries three main pieces of information: the magnitude of σ; the phase of the electromagnetic wave returning; and a delay between the emitted and received signal, which is related to the distance between the radar and the target area. Scatterometers focus particularly on the magnitude of the backscattered signal by a medium to extract backscattering coefficient (Figa-Saldaña et al., 2002). Synthetic-aperture radar (SAR) measures backscattering coefficients using a radar technique that can achieve spatial resolutions at a meter scale by combining scenes of the target area from multiple points of view (Tomiyasu, 1978; Bamler, 2000). Synthetic-aperture radars achieve comparable resolutions to visible and infrared radiometers (i.e., below 10 m of spatial resolution). However, radar transmitters are much more energy-consuming than radiometers because of the power requirement for emitting microwaves. Because of high energy demands, most SAR transmitters operate during a small fraction (1 %–30 %; Grasso et al., 2021; Leanza et al., 2019; Dubock et al., 2001) of their orbit around Earth. The lower the transmission time, the greater the duration between measurements at a given point, often several days (Marghany, 2019). This is currently being overcome via deployment of several satellites equipped with SAR instruments flying in tandem (Sentinel-1; RADARSAT Constellation Mission).

Unlike visible and infrared remote sensing that are limited to information from the surface of any target, the longer wavelengths of microwaves allow for microwaves to penetrate vegetation, ground and snowpack, allowing for subsurface measurements (Ulaby et al., 1986); the longer the microwave, the deeper the penetration depth. However, deeper penetration means that the received signal is a combination of different contributing components, i.e., usually vegetation, soil, snow and atmosphere (Kerr et al., 2012; Roy et al., 2012, 2014), challenging the interpretation of the signal. Consequently, microwave remote sensing is often used together with radiative transfer models to decouple and extract the information on terrestrial surface conditions from microwave observations. These models generally calculate the scattering, reflection and attenuation of the electromagnetic waves of the different components of the surface (Fig. 2) (Mo et al., 1982; El-Rayes and Ulaby, 1987; Huang et al., 2017; Picard et al., 2018). By considering the contribution of each component of the surface (i.e., soil, vegetation and/or snow), radiative transfer models allow for disentangling microwave signals to retrieve key surface state variables of interest (Wigneron et al., 2007).

Figure 2 Microwave-signal decoupling through electromagnetic wave interaction with matter. (a) Mo et al. (1982) model for passive microwaves with T_B representing the brightness temperature, e being the soil emissivity, γ being the vegetation attenuation, ω being the vegetation single-scattering albedo, and T being the thermodynamic temperature of soil and vegetation (Mo et al., 1982). (b) Water cloud model for active microwaves with σ representing the backscattering coefficient, γ being vegetation attenuation, A being a vegetation empirical parameter and θ being the radar incident angle (Attema and Ulaby, 1978). The σ_veg term is an approximation for the first-order vegetation scattering processes. Snow increases the complexity of the interaction between the microwave and the ground level because it is acting as an additional semi-opaque layer comparable to the vegetation layer. Snow and atmosphere are not accounted for in the equations presented.

Multiple microwave radiometers have been launched in recent decades covering most of the microwave portion of the electromagnetic spectrum: L-band (1–2 GHz), C-band (4–8 GHz), X-band (8–12 GHz), K-band (18–26.5 GHz), K_a-band (26.5–40 GHz) and W-band (75–110 GHz) (Table A1 in Appendix A). There is a nearly continuous publicly available radiometric dataset from 1978 to the present covering the microwave bands from the C-band to the W-band (Fig. 3a). Additionally, in the last decade, L-band data have become accessible through the launch of the Soil Moisture Ocean Salinity mission (SMOS; Kerr et al., 2010), the Aquarius mission (Brucker et al., 2004) and the NASA (National Aeronautics and Space Administration) Soil Moisture Active Passive mission (SMAP; Entekhabi et al., 2010). An exhaustive list of radars would be too long to present here; instead we focus on a selection of recent radar missions (Fig. 3b). These spaceborne radars were selected for the purpose of the review to represent a range of frequencies and available temporal coverage.

Figure 3 (a) Temporal coverage of spaceborne passive microwave instruments (radiometers between 1–100 GHz, sun-synchronous nearly polar orbits; more details in Table A1). (b) Selected number of spaceborne radar missions (active sensors) covering the L-band (1–2 GHz), C-band (4–8 GHz), X-band (8–12 GHz) and K_u-band (12–18 GHz). More details are provided in Table A2.

3 Microwave remote sensing for retrieval of surface variables

Although direct measurements of vertical C fluxes are not feasible through microwave remote sensing, it is possible to use spaceborne radiometers and radar to monitor key variables important for different C-cycle processes (Fig. 4). Carbon dioxide and CH₄ fluxes, two of the most potent greenhouse gases, are continuously exchanged between Earth's surface and the atmosphere. Carbon dioxide is absorbed through photosynthesis by vegetation and is released by plants and soil through respiration (i.e., autotrophic respiration, R_a, and heterotrophic respiration, R_h, respectively) (Chapin et al., 2006). In ecosystems, CH₄ is produced by methanogens under anaerobic conditions and consumed during oxidation by methanotrophs under aerobic conditions (Lai, 2009). Methane transport through the soil column and to the atmosphere occurs through diffusion, ebullition and plant-mediated transport (Lai, 2009).

Figure 4 (a) Carbon dioxide and methane fluxes between the land surface and the atmosphere including gross primary production (GPP), heterotrophic respiration (R_h), autotrophic respiration (R_a) and net methane fluxes ($F_{{\mathrm{CH}}_{\mathrm{4}}}$) (units: gC m⁻² yr⁻¹). The fluxes are color-coded (green: GPP and R_a; red: R_h; purple: $F_{{\mathrm{CH}}_{\mathrm{4}}}$) to match relevant key variables important for CO₂ and CH₄ fluxes and derivable with microwave remote sensing. (b) Relationships of net primary productivity and net ecosystem productivity (NEP: negative net ecosystem exchange, NEE) and their component fluxes with ecosystem respiration (ER = R_a+R_h).

3.1 Soil moisture

In ABRs, water availability and air and soil temperatures are considered important environmental constraints on photosynthesis and thus C sequestration (Lieffers and Rothwell, 1987; Angert et al., 2005; Jones et al., 2017). For example, both heat stress and drought can limit tree growth across latitudes in boreal forest stands (Walker and Johnstone, 2014; Sniderhan et al., 2021). Near-surface soil moisture also affects methanogenesis and methanotrophy, which depends strongly on oxygen availability (Lai, 2009). Environmental controls on soil respiration, i.e., the sum of belowground R_a and R_h, include soil moisture and temperature, which in ABRs are both influenced by permafrost and active (seasonally thawed) layer dynamics (Huntzinger et al., 2020). Hence, in a changing climate where soils across ABRs are generally expected to be drying (e.g., Gauthier et al., 2015; Andresen et al., 2020), reliably quantifying soil moisture dynamics and thus water availability will become essential for a predictive understanding of C-cycle dynamics.

The sensitivity of the microwave signal to soil moisture content has been widely demonstrated previously, but several challenges remain including accounting for vegetation attenuation and scattering related to the surface roughness (Das and Paul, 2015; Colliander et al., 2022), as well as organic soil model parametrization (Mironov and Savin, 2015; Bircher et al., 2016). The sensitivity to soil moisture generally increases with lower microwave frequencies, making the L-band the most sensitive to soil moisture content. The depth over which soil moisture content can be estimated from microwave remote sensing is limited at best to the top 5 cm of the soil profile (roots are most dense in the top 20 cm). Measurement depth depends on the frequency used (longer wavelengths realize greater depths; Adams et al., 2015), the soil moisture content, and the vegetation type and density (Wigneron et al., 2007). These measurement depths do not capture the water availability in the full rooting zone, which will be most relevant to predicting the photosynthetic CO₂ update. However, it remains possible to estimate root zone soil moisture content from near-surface soil moisture content using pedotransfer equations, meaning these surface measurements can be quite useful (Stefan et al., 2021; Dimitrov et al., 2022).

Extensive methodological research to develop soil moisture retrievals from microwave remote sensing has been motivated for agricultural applications (Engman, 1991; Wigneron et al., 2007; Lakhankar et al., 2009). Soil moisture retrieval was the main motivation behind several passive L-band satellite missions, including SMAP and SMOS. The resulting expertise and soil moisture products present a promising data stream to inform terrestrial biosphere models (see Sect. 4). Table 1 presents several available soil moisture products. The European Space Agency (ESA) Climate Change Initiative (CCI) project produces a soil moisture product merging active and passive microwave sensors, exploiting the microwave instruments temporal coverage since 1978 (Gruber et al., 2019).

Table 1 Presently available soil moisture products from spaceborne microwave remote sensing.

^a WindSat/Coriolis surface soil moisture (LPRM) L3 1 day 25 km × 25 km daytime V001. ^b SMOS L1 and L2 Science data. ^c Aquarius L3 Weekly Polar-Gridded Sea Surface Salinity, Version 5 (AQ3_SSS). ^d AMSR-E/AMSR2 Unified L2B Half-Orbit 25 km EASE-Grid Surface Soil Moisture, Version 1 (AU_Land). ^e SMAP L4 Global 3-hourly 9 km EASE-Grid Surface and Root Zone Soil Moisture Geophysical Data, Version 6 (SPL4SMGP). ^f ERS-2 SCATTEROMETER Surface Soil Moisture Time Series and Orbit product in High and Nominal Resolution. ^g ASCAT Soil Moisture at 25 km Swath Grid in NRT – Metop. ^h RADARSAT-2 Surface Soil Moisture. ⁱ 1 km SSM (surface soil moisture) Version 1 product (SSM1km). ^j ESA CCI SM.

3.2 Soil temperature

Rates of photosynthesis and ecosystem respiration (R_eco; Fig. 4) are generally strongly controlled by soil temperature (Angert et al., 2005; Jones et al., 2017; Stocker et al., 2018). Similarly, laboratory and field observations have shown that R_h-related CO₂ fluxes are non-negligible below the freezing point and increase with warming soil temperatures (Fahnestock et al., 1998, 1999; Welker et al., 2000; Mikan et al., 2002; Panikov et al., 2006; Natali et al., 2019). Despite lower fluxes, winter CO₂ emissions from soil respiration may constitute an important contribution to annual net ecosystem productivity (NEP), especially in ABRs with very short growing seasons (Elberling et al., 2007; Webb et al., 2016; Natali et al., 2019).

Microwave T_B depends on thermodynamic temperature ($T_{\mathrm{B}}=e\cdot T$); thus it is possible to retrieve land surface temperature (LST) from radiometric data (Duan et al., 2020). Retrieval of LST from microwave remote sensing is technically more challenging and results in lower precision than similar products from thermal infrared remote sensing, 1–5 K for microwave LST vs. 0.2–2 K for thermal infrared (Jiménez-Muñoz and Sobrino, 2006; Jones et al., 2010; Osińska-Skotak, 2007; Krishnan et al., 2020; Zhang and Cheng, 2020). Microwave emissivity is also more sensitive to environmental changes (e.g., liquid water content) than thermal infrared emissivity. Although infrared LST is more precise than microwave LST, there are instances where microwave LST may be important. Specifically, microwave LST can support gap filling when thermal infrared LST is unavailable because of extended periods of cloud cover or when fine temporal resolution is required. Microwave can be used to sense soil temperature up to a few centimeters deep because of its soil penetration depth. It is an important advantage over thermal infrared remote sensing, which can only detect LST and is subject to significant atmospheric contamination (Jones et al., 2007). Some studies have also shown the potential to use microwave observations to retrieve soil temperature under the snow (Kohn and Royer, 2010; Marchand et al., 2018).

Although several studies have obtained LST from microwave data, no operational microwave LST product is operationally available. Microwave LST algorithms typically achieve a precision of 2–5 K using data from AMSR-E/AMSR2 (Jones et al., 2010; Zhang and Cheng, 2020) and SMM/I and SMMR (Pulliainen et al., 1997; Basist et al., 1998; Fily et al., 2003; Mialon et al., 2007; Royer and Poirier, 2010). Other studies have used machine learning approaches, obtaining precision around 1–3 K with AMSR2 data (Aires et al., 2001; Mao et al., 2018).

3.3 Surface freeze–thaw state

During the short growing season in ABRs, photosynthetic CO₂ uptake exceeds the respiratory CO₂ losses; thus ABRs generally act as net growing season CO₂ sinks (Ciais et al., 2019; Virkkala et al., 2021). However, in winter, plants become dormant and photosynthesis and R_a cease (gross primary productivity, GPP = 0; R_a≈0) owing to the cold temperatures and highly reduced photoperiod (Kimball et al., 2004b; Rafat et al., 2021). Although R_h may continue in frozen soil, it decreases substantially (Natali et al., 2019). Modeling annual NEP in areas undergoing seasonal surface freeze–thaw cycles requires the ability to estimate the length and timing of the growing season (Seiler et al., 2022; El-Amine et al., 2022). The growing season length has a direct impact on annual GPP and thus NEP, but there is also a strong relationship between surface freeze–thaw timing and photosynthetic CO₂ uptake (Frolking et al., 1996; McDonald et al., 2004; Kim et al., 2012; Fu et al., 2017; Pierrat et al., 2021). Changes in the timing of spring thaw can create a shift in growing conditions when photosynthesis is initiated (Jarvis and Linder, 2000; Tanja et al., 2003; Kimball et al., 2004a; Piao et al., 2008; Kim et al., 2012; El-Amine et al., 2022). The timing of the start of the growing season has been shown to be more important to annual GPP than the timing of the end because of the superior light and water availability during the spring period (Tanja et al., 2003; El-Amine et al., 2022). The surface freeze–thaw state is a useful proxy for the timing and thus duration of photosynthetic activity (Harrison et al., 2020) and can potentially be used to track CH₄ emissions in ABRs (Tenkanen et al., 2021).

Satellite detection of the surface freeze–thaw state is based on the dielectric contrast between water and ice at microwave frequencies. Therefore, soil emissivity is highly sensitive to phase state changes in its liquid content. When water freezes, ε_soil drops drastically as liquid water changes to ice because of the crystalline structure of frozen water. The rapid decrease in ε_soil in freezing soils translates into a much higher microwave emission and backscattering from the surface. This allows for surface freeze–thaw state retrieval from passive and active microwave measurements using temporal change detection algorithms (Mortin et al., 2012; Rautiainen et al., 2012; Roy et al., 2015; Chen et al., 2019) and threshold-based methods (Kim et al., 2012, 2017; Derksen et al., 2017). Furthermore, for oblique incidence angles, horizontal polarization is more affected than its vertical counterpart during the surface freeze–thaw transition, which favors the use of a polarization ratio as an effective tool for determining the surface freeze–thaw state (Rautiainen et al., 2016; Roy et al., 2017a, b).

The use of passive and active microwave remote sensing for surface freeze–thaw state detection has been widely studied, improving quickly and steadily in the last decade and resulting in various publicly available products (Table 2). As for soil moisture studies, lower microwave frequencies such as the L-band, have been the most exploited because of the prominent water phase dielectric contrast and reduced attenuation from aboveground vegetation and snow (Rautiainen et al., 2016). Surface freeze–thaw state products offer the opportunity to constrain the vegetation growing season timing and photosynthetic CO₂ uptake in terrestrial biosphere models.

Table 2 Available surface freeze–thaw state products from spaceborne microwave remote sensing.

^a Sensors: SMMR, SSM/I, SSMIS, AMSR-E and AMSR2. ^b Aquarius L3 Weekly Polar-Gridded Landscape Freeze/Thaw Data, Version 5 (AQ3_FT). ^c SMOS Level 3 Freeze and Thaw (F/T). ^d SMAP Enhanced L3 Radiometer Global and Northern Hemisphere Daily 9 km EASE-Grid Freeze/Thaw State, Version 3 (SPL3FTP_E). ^e ASCAT Surface Soil Moisture/Freeze-Thaw V2 product.

3.4 Aboveground biomass

Although not a direct control on C-cycle processes, aboveground biomass (AGB) can be used to estimate aboveground C stocks. The annual net gain or loss of C by vegetation (NPP) leads to a proportional increase or decrease in AGB, respectively (Turner et al., 2004; Gough, 2011). Therefore, C storage in aboveground vegetation over a defined period can be inferred from AGB estimates from either terrestrial, aerial or spaceborne techniques (Das et al., 2021). However, using AGB only provides average C exchanges between two data acquisitions and provides no understanding of the underlying ecophysiological and biogeochemical processes. It should be noted that AGB is a minor component of the C stocks in ABRs, where most of the C is stored belowground (Houghton, 2005; Pappas et al., 2020; Walker et al., 2020). Information on AGB allows for initialization, parameterization and evaluation of terrestrial biosphere models and helps to understand, for example, the impact of discrete disturbances such as wildfire and insect outbreaks (Chirici et al., 2016).

Microwave remote sensing can provide information on AGB since microwave wavelengths typically penetrate and interact with moderately dense vegetation cover, depending on the microwave length (Liu et al., 2011a, b). Vegetation attenuation of microwaves is characterized by the vegetation optical depth (VOD; $\mathit{\gamma }=e^{-\mathrm{VOD}/\cos \mathit{\theta }}$), which is proportional to vegetation density (i.e., biomass) and water content. Vegetation optical depth is frequency-dependent and affected by the geometry (e.g., canopy structure) of the vegetation (Ulaby et al., 1990). The impact of vegetation geometry on attenuation is challenging to parametrize because of the complexity of forest canopies; therefore it is typically accounted for through vegetation type-dependent, empirically determined parameters introduced in the relationship between VOD, vegetation water content and biomass (Jackson and Schmugge, 1991; Konings et al., 2019). Microwave VOD is better suited to directly monitoring vegetation compared to spectral vegetation indices obtained using visible and infrared wavelengths such as NDVI that saturate at relatively low AGB of around 50–80 Mg ha⁻¹ (Rodríguez-Fernández et al., 2018; Mialon et al., 2020; Turner et al., 2004). In contrast, VOD was shown to saturate AGB of up to 350 Mg ha⁻¹ (Vittucci et al., 2019). It has also been shown that there is a significant difference in what phenological aspects of the growing season are captured by vegetation indices such as NDVI vs. microwave VOD (Milaon et al., 2020). VOD seems to correspond better to key physiological processes such as sap flow and vegetation water storage than vegetation indices such as NDVI, which better capture dynamics of canopy phenology (Lawrence et al., 2014; Cui et al., 2015; Tian et al., 2016; Holtzman et al., 2021; Wigneron et al., 2021).

There is clear potential to improve C-cycle science in boreal forests using VOD-derived AGB (Rodríguez-Fernández et al., 2018). Several recent studies have also shown good correlations between VOD and GPP (Teubner et al., 2018, 2019). The L-band spaceborne radiometer record goes back to 2010 with SMOS, while higher-frequency VOD estimates extend back to the early 1980s. Table 3 presents several VOD products available. Santoro and Cartus (2018) counted 221 studies on SAR data applied to AGB retrieval from 1987 to 2017 using frequencies from 30 MHz up to 12 GHz. AGB investigations using active sensors can achieve a much finer spatial resolution than their passive counterparts, down to the 10 m scale. Also, recent promising advances in SAR interferometry (InSAR), polarimetric InSAR (PolInSAR) and SAR tomography (TomoSAR) techniques provide new opportunities for AGB estimates by surveying the 3D structure of vegetation (Neumann et al., 2012; Tebaldini et al., 2019). InSAR can be used to measure the vertical motion in peatlands, which is a direct indicator of the mass gain or loss by those ecosystems, which constitute a major global C pool (Alshammari et al., 2019; Zhou et al., 2019; Loisel et al., 2021). Some global biomass surveys have also exploited the multi-frequency synergy of data products in the microwave, visible and infrared wavelengths and lidar technologies (e.g., GlobBiomass: Santoro et al., 2018; Mialon et al., 2020). Although the low AGB of the Arctic tundra is challenging to monitor from microwave observations, studies have shown that it is possible to estimate the upper soil organic C (up to 30 cm from the soil–atmosphere interface) using active and passive microwave observations (Bartsch et al., 2016; Yi et al., 2022).

Table 3 Available products for vegetation optical depth from spaceborne microwave remote sensing.

^a Daily Global Land Parameters Derived from AMSR-E and AMSR2, Version 3 (NSIDC-0451). ^b L3 SMOS-IC v2. ^c SMAP-IB L-VOD. ^d ASCAT IB VOD.

3.5 Vegetation water storage

Water availability is considered an important environmental limitation on photosynthetic processes in ABRs (Ruiz-Pérez and Vico, 2020). In terrestrial biosphere models, a lack of water availability is an environmental stress reducing the photosynthetic capacity (i.e., NPP) (Mu et al., 2007). Water stored in vegetation is critical for stomatal regulation; therefore it is strongly correlated to vegetation growth (Köcher et al., 2013; Matheny et al., 2015). In addition, vegetation water storage can act as a buffer for the daily demands of transpiration (Matheny et al., 2015). However, soil moisture and/or precipitation are generally used to estimate water availability since vegetation water storage estimates are rarely available (Zhang et al., 2015; Stocker et al., 2018).

Since vegetation water storage strongly affects microwave VOD because of the high absorption of microwave by water (Konings et al., 2019), microwave attenuation holds potential for estimating vegetation water storage, which can be used to evaluate vegetation water stress (Holtzman et al., 2021). To the best of our knowledge, no large-scale vegetation water storage product yet exists, but efforts toward this goal are underway (Y. Liu et al., 2021). The microwave VOD sensitivity to both AGB and the vegetation water status complicates its interpretation, although the study of the temporal and spatial trends of VOD can allow for disentangling AGB vs. the vegetation water content (Dou et al., 2023). At short timescales (i.e., diurnal), biomass variation is small and VOD trends can largely be attributed to vegetation water status. At longer timescales (i.e., annual), VOD trends come mostly from biomass dynamics (Mialon et al., 2020). Another method to distinguish water storage and biomass-related VOD changes is to use periods with similar water stress levels for VOD comparison (Konings et al., 2019).

3.6 Land cover

Regional and global studies on C exchanges require information on land cover (Gasser et al., 2020). Repeated satellite-based image classification provides large-scale monitoring of land cover evolution (Wang et al., 2019). Land cover and wetland classifications based on SAR imagery have been developed with the same supervised or unsupervised classification algorithms used for classifying imagery obtained with visible and infrared remote sensing (van Zyl, 1989; Pierce et al., 1994; Dobson et al., 1995; Ranson and Sun, 2000; Bartsch et al., 2007; Whitcomb et al., 2009; Lönnqvist et al., 2010; Merchant et al., 2017, 2022). Recent studies have explored the use of machine learning for land cover classification based on SAR data (Merchant et al., 2019). SAR land cover classifications are enhanced when benefiting from multi-frequency instruments (Saatchi and Rignot, 1997) and multiple polarizations (Lee et al., 2001). The complementarity of SAR and visible/infrared imagery has already been exploited to reinforce spatial and temporal coverage and improve precision of land cover classifications (Töyrä et al., 2001; Ullmann et al., 2014; Merchant et al., 2019). SAR imagery has shown to be especially useful for delineating inundated areas (Bowling et al., 2003) or wet and moist tundra (Morrissey et al., 1996; Merchant et al., 2022), which has a strong impact on CH₄ emission (Watts et al., 2014). Microwave observations can also be used to monitor freshwater (FW) waterbody extent dynamics (Murfitt and Duguay, 2021). FW can act as important CH₄ emissions sources, especially during ice melt, but aquatic carbon cycle processes are very different than terrestrial carbon processes and were not within of the scope of this review (Matthews et al., 2020).

3.7 Snow cover

Unlike photosynthesis and R_a, R_h can continue through winter in the cold regions. The insulating properties of snow cover have an important indirect impact on C fluxes, keeping the ground warmer than the air during winter, thereby maintaining microbial activity and therefore R_h (Brooks et al., 1997; Brooks and Williams, 1999; Welker et al., 2000; Elberling et al., 2007; Ravn et al., 2020). Through modeling the land surface energy exchanges, the impact of snow on the soil thermal regime can be estimated (Melton et al., 2020). Several studies have demonstrated a correlation between C fluxes and snow depth due to the insulating properties of snow (Björkman et al., 2010; Rogers et al., 2010; Natali et al., 2019). Furthermore, snow density is the main factor controlling snow thermal conductivity (Sturm et al., 1997), which should consequently influence soil temperature, winter CO₂ and CH₄ fluxes.

Estimating snow accumulation in ABRs represents a challenge for microwave remote sensing. Microwaves are partially absorbed and scattered by snow cover, with an additional challenge in the boreal forest because of snow interception by the forest canopy (Li et al., 2019). Active and passive microwaves can retrieve information about the snowpack status based on microwave interaction with the snow microstructure (Picard et al., 2018). Because multiple snowpack characteristics influence microwaves in many ways (i.e., height, density, microstructure, layering, liquid water content), retrieving snowpack characteristics often leads to an underdetermined equation system (i.e., fewer equations than unknowns). During springtime and winter rain-on-snow (ROS) events, wet snow becomes a major limitation for microwave remote sensing. Liquid water absorbs microwave radiation, reducing microwave penetration depth in wet snow and preventing acquisition of information about either the snow or the underlying ground conditions. This is especially problematic for the retrieval of the surface freeze–thaw state during the spring thaw (Rautiainen et al., 2016). However, snowmelt is easily detected (Forster et al., 2001) and can serve as a proxy for the beginning of the growing season and the resulting initiation of C uptake (Pulliainen et al., 2017).

Estimating the snow water equivalent (SWE; i.e., the product of snow depth and density) is widely used to monitor bulk snow cover using microwaves (Chang et al., 1982; Pulliainen, 2006; Shi et al., 2016; Pulliainen et al., 2020; Saberi et al., 2020; Table 4). It should be noted that uncertainties are higher in the presence of vegetation because of the ensuing microwave attenuation (Mortimer et al., 2020) and for deep snow conditions (SWE > 150 mm: Larue et al., 2017).

Table 4 Available snow water equivalent (SWE) products from spaceborne microwave remote sensing.

^a GlobSnow v3.0 NH SWE; sensors: SMMR, SSM/I, SSMIS. ^b AMSR-E/AMSR2 Unified L3 Global Daily 25 km EASE-Grid Snow Water Equivalent, Version 1 (AU_DySno).

4 Assimilating microwave data in terrestrial biosphere models in Arctic–boreal regions

Microwave remote sensing has the potential to greatly improve predictions of C fluxes in terrestrial biosphere models. However, the use of key variables obtained from microwave remote sensing to inform terrestrial biosphere models is still limited (Lees et al., 2018). A recent effort assimilates a microwave remote sensing soil moisture data product into a simple C-cycle model to compute the SMAP L4 global daily 9 km EASE-Grid Carbon Net Ecosystem Exchange product (SPL4CMDL; Jones et al., 2017; Fig. 5). The SPL4CMDL product provides daily NEE, GPP, R_h, soil organic C and environmental constraints for eight plant functional types at 9 km spatial resolution. The product is publicly available with data starting in 2015. The L4C carbon exchange model estimates GPP using a light-use efficiency model, where R_a is defined as a fraction of GPP and R_h is estimated using a soil decomposition model with cascading soil organic C. The GEOS-5 atmospheric model is used for the meteorological inputs, and MODIS visible/infrared products are used for land cover and photosynthetically active radiation. There is ongoing work to improve L4C estimates by integrating surface freeze–thaw state information from microwave remote sensing. The SMAP L4C product is currently the only operational C-exchange model using satellite microwave information, which provides direct information for studying seasonal cycles and long-term trends in ecosystem C exchange. SMOS soil moisture data have similarly been used in terrestrial biosphere models but for a shorter period (2010–2015; Wu et al., 2020). Lastly, the terrestrial C-flux (TCF; Kimball et al., 2009) model used AMSR-E soil moisture and temperature information to provide environmental constraints on R_h in Arctic–boreal regions (Baldocchi et al., 2001; Pastorello et al., 2020). Although limited so far, those examples show the potential of integrating microwave products in terrestrial biosphere models.

Figure 5 Mean monthly net ecosystem exchange (NEE) from the SPL4CMDL for July and August (i.e., peak growing season) between 2015 and 2021. The Arctic biome is delimited following the Conservation of Arctic Flora and Fauna (CAFF) working group of the Arctic Council, and the boreal biome is delineated following Potapov et al. (2008). The yearly mean NEE for the Arctic (N_pixel = 282 288) and boreal regions (N_pixel = 187 003) is presented as boxplots for the 25th–75th percentiles (box) and the 1st–99th percentiles (bracket). SMAP L4C reference: Kimball et al. (2017).

5 Challenges and opportunities in microwave remote sensing for supporting carbon cycle science

Although microwave remote sensing data processing is reaching a certain maturity after decades of development, challenges remain in the advancement of microwave data and their assimilation into terrestrial biosphere models (Jones et al., 2017).

5.1 Disentangling the integrated microwave signal

Disentangling the integrated microwave signal originating from varying mixtures of soil, vegetation and snow remains challenging for all microwave remote sensing applications (Kerr et al., 2012; Roy et al., 2012, 2014). Recent advances have been made for both passive and active data to decouple the signal in boreal forests by exploiting multi-polarization, multi-angular and multi-frequency measurements (Larue et al., 2018; Cohen et al., 2019; Konings et al., 2019; Roy et al., 2020). Even in the Arctic biome, where AGB is relatively low, the plot-scale heterogeneity in vegetation composition and structure can pose a challenge since repeated field estimates of vegetation distribution are limited (Myers-Smith et al., 2011; J. Du et al., 2019). Furthermore, the low C fluxes of ABRs can be more challenging to estimate, and vegetation production trends can be lost in interannual variability fluctuations (i.e., low signal-to-noise ratio). The decoupling challenge can be exemplified clearly with the respective effects of soil and vegetation on surface freeze–thaw state detection since soil and vegetation might not freeze or thaw concurrently with implications for detection of surface freeze–thaw events (Roy et al., 2020). Microwave data processing must account for the high density of shallow waterbodies in ABRs, and recent efforts have sought to remove their effect on passive microwave observations in mixed pixels (Touati et al., 2019). These challenges highlight the importance of continued efforts to improve radiative transfer modeling in a way that will allow for the decoupling of the contribution of different components to the microwave signal.

5.2 Intra-pixel variability

Due to the coarse spatial resolution of passive microwave data products, intra-pixel variability and downscaling are important considerations. For example, a recent study from Prince et al. (2019) upholds that the time span of the surface freeze–thaw signal transition in passive microwave measurements might be related to the spatial variability of the soil state (frozen or thawed) through this transition (i.e., patchy frozen soil). Intra-pixel variability also includes the partial presence of snow during transition seasons, vegetation type mixing and snow depth distribution (Meloche et al., 2022). A recent study from Du et al. (2022) showed promising results in downscaling soil moisture to 3 m spatial resolution using machine learning with microwave spaceborne data. Although challenging for subregional studies, the typical passive microwave remote sensing spatial resolution is well suited for regional and global climate studies. Still, most of the terrestrial biosphere model performance evaluation is done using eddy covariance data which are the most trusted and widely used reference for C-flux measurements at large scales. However, the sparsity of the measurement network in ABRs reduces our capacity to represent ABR heterogeneity in the terrestrial biosphere model (Fig. 1; Fisher et al., 2018). However, the scale mismatch between measurements is important to keep in mind; tower-based measurements of C exchanges represent footprints of approximately 1 km or less, which is a much smaller scale than many spaceborne microwave remote sensing data products.

Table 5 Overview of selected future microwave satellite missions with a polar or nearly polar orbit.

NA – not available.

5.3 Potential and upcoming spaceborne microwave remote sensing missions

Upcoming spaceborne microwave remote sensing missions will provide new microwave sensors with additional frequencies and generally improved radiometric and spatial resolution (Table 5). These upcoming missions will increase remote sensing capabilities and extend the continuous microwave coverage over the next decade. The technical and scientific advances made through these missions should be seen as an additional motivation to actively identify ways to integrate microwave data products in C-cycle science. Furthermore, the novel approach of the opportunistic use of spaceborne reflectometry of the global navigation satellite system (GNSS) (W. Li et al., 2022; Yu et al., 2022) already showed promising results in evaluating soil moisture (Edokossi et al., 2020), the soil freeze–thaw state (Rautiainen et al., 2022) and snow water equivalent (Royer et al., 2021).

6 Conclusions

Microwave remote sensing is an efficient tool for monitoring soil, seasonal snow and vegetation properties affecting water and C-cycle processes and thus has great potential to enhance our understanding of rapidly changing Arctic–boreal regions. Understanding C-cycle processes in Arctic–boreal regions will greatly benefit from an increased use of microwave data, which can only result from expanded collaboration between the microwave remote sensing and C-cycle science communities. The potential applications of satellite-based microwave remote sensing in support of C-cycle science have not been fully realized. We propose four key aspects where increased use of microwave remote sensing could support advances in C-cycle science and monitoring: (1) improve radiative transfer model skills and capabilities and the understanding of microwave signals associated with different surface material; (2) improve assimilation approaches of microwave products in terrestrial biosphere models to overcome the challenges associated with remote cold regions where ground observations are spatially and temporally sparse; (3) develop and maintain long-term, spatially distributed, land-based measurement networks in Arctic–boreal regions to improve microwave-based products; and (4) keep a long-term perspective and coherency between space agencies to maintain the historical trends of microwave observations.

Appendix A

Table A1 Microwave radiometers in the 1–100 GHz frequency range on sun-synchronous nearly polar orbits. Instrument specifications from Das and Paul (2015) and specific references.

Table A2 Microwave radar missions.

Appendix B: Acronyms and abbreviations

ASAR	Advanced Synthetic Aperture Radar
ASCAT	Advanced SCATterometer
AGB	Aboveground biomass
AMSR	Advanced Microwave Scanning Radiometer
AMSR-E	AMSR for Earth Observing System
CARDAMOM	Carbon data model framework
CCI	Climate Change Initiative
CIMR	Copernicus Imaging Microwave Radiometer
CNES	Centre National d'Études Spatiales
CO2	Carbon dioxide
CH4	Methane
DLR	German Aerospace Center
e	Soil emissivity
ERS	European Remote Sensing satellites
ESA	European Space Agency
FAO	Food and Agriculture Organization of the United Nations
FW	Freshwater
GEOS-5	Goddard Earth Observing System, Version 5
GNSS	Global navigation satellite system
GPP	Gross primary production
H2O	Water
IFOV	Instantaneous field of view
InSAR	SAR interferometry
ISRO	Indian Space Research Organisation
LAI	Leaf area index
Lidar	Light detection and ranging
LST	Land surface temperature
MEaSUREs	Making Earth System Data Records for Use in Research Environments
MODIS	Moderate Resolution Imaging Spectroradiometer
MWRI	Micro-Wave Radiation Imager
NASA	National Aeronautics and Space Administration
NDVI	Normalized difference vegetation index
NEE	Net ecosystem exchange
NEP	Net ecosystem production
NISAR	NASA–ISRO SAR mission
NPP	Net primary production
NSCAT	NASA Scatterometer
PALSAR	Phased Array type L-band SAR
PolInSAR	Polarimetric InSAR
QuickSCAT	Quick Scatterometer
Ra	Autotrophic respiration
Reco	Ecosystem respiration
Rh	Heterotrophic respiration
REP	Red-edge position
ROS	Rain-on-snow events
ROSE-L	Radar Observing System for Europe – L-Band
SAR	Synthetic-aperture radar
SMAP	Soil Moisture Active Passive
SMAP L4C	SMAP Level-4 carbon product
SMMR	Scanning Multichannel Microwave Radiometer
SMOS	Soil Moisture and Ocean Salinity
SMOS-HR	SMOS High-Resolution

SWE	Snow water equivalent
SSM/I	Special Sensor Microwave/Imager
SSMIS	Special Sensor Microwave Imager/Sounder
TB	Brightness temperature
TCF	Terrestrial carbon flux
TomoSAR	SAR tomography
VOD	Vegetation optical depth
γ	Vegetation attenuation
η	Light conversion efficiency
ηmax	Light conversion efficiency maximum in optimal conditions
ε	Relative permittivity
θ	Incident angle
σ	Backscattering coefficient
τ	Optical depth
Ω	Vegetation single-scattering albedo