Environmental science is increasingly reliant on remotely sensed observations of the Earth's surface and atmosphere. Observations from polar-orbiting satellites have long supported investigations on land cover change, ecosystem productivity, hydrology, climate, the impacts of disturbance, and more and are critical for extrapolating (upscaling) ground-based measurements to larger areas. However, the limited temporal frequency at which polar-orbiting satellites observe the Earth limits our understanding of rapidly evolving ecosystem processes, especially in areas with frequent cloud cover. Geostationary satellites have observed the Earth's surface and atmosphere at high temporal frequency for decades, and their imagers now have spectral resolutions in the visible and near-infrared regions that are comparable to commonly used polar-orbiting sensors like the Moderate Resolution Imaging Spectroradiometer (MODIS), Visible Infrared Imaging Radiometer Suite (VIIRS), or Landsat. These advances extend applications of geostationary Earth observations from weather monitoring to multiple disciplines in ecology and environmental science. We review a number of existing applications that use data from geostationary platforms and present upcoming opportunities for observing key ecosystem properties using high-frequency observations from the Advanced Baseline Imagers (ABI) on the Geostationary Operational Environmental Satellites (GOES), which routinely observe the Western Hemisphere every 5–15 min. Many of the existing applications in environmental science from ABI are focused on estimating land surface temperature, solar radiation, evapotranspiration, and biomass burning emissions along with detecting rapid drought development and wildfire. Ongoing work in estimating vegetation properties and phenology from other geostationary platforms demonstrates the potential to expand ABI observations to estimate vegetation greenness, moisture, and productivity at a high temporal frequency across the Western Hemisphere. Finally, we present emerging opportunities to address the relatively coarse resolution of ABI observations through multisensor fusion to resolve landscape heterogeneity and to leverage observations from ABI to study the carbon cycle and ecosystem function at unprecedented temporal frequency.
Modern environmental science would be unrecognizable without satellite remote sensing, which has revolutionized our field since its advent over a half-century ago (Kerr and Ostrovsky, 2003). The platforms by which we observe the Earth system are increasingly diverse and now include miniaturized satellites (CubeSats), sensors like the ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) traveling on board the International Space Station (Hulley et al., 2017), and even lidar systems (Coyle et al., 2015; Qi et al., 2019), yet most environmental science applications employ polar-orbiting satellites, namely Landsat and the Moderate Resolution Imaging Spectroradiometer (MODIS). The Landsat and MODIS programs have radically improved the ability of ecologists to track vegetation change and its impacts on habitat, biogeochemical cycling, and other ecosystem services (De Araujo Barbosa et al., 2015). Like all remote sensing platforms, polar-orbiting satellites must make compromises with respect to spectral resolution, spatial scale, and temporal scale that limit their ability to measure all things, all the time. Notably, the 1 to 2 d cadence of MODIS and Visible Infrared Imaging Radiometer Suite (VIIRS) and 3 to 5 d cadence of the combined Landsat- and Sentinel-2-class sensors may be insufficient for tracking ecological phenomena that occur at shorter timescales, including the timing of rapid environmental change (White et al., 2009) and the diurnal behavior of land surface function, such as sub-hourly variations in ecosystem carbon and water fluxes (Chudnovsky et al., 2004; Grant et al., 2000).
As a part of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), the Satellite Application Facility for Land Surface Analysis (LSA SAF) has leveraged high frequency observations from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on board the Meteosat Second Generation (MSG) geostationary satellites to provide operational products relevant for studying vegetation, wildfires, the surface radiation budget, and the carbon and water cycle (Trigo et al., 2011) at sub-daily timescales. These opportunities are also available in the Western Hemisphere. Focusing on the Advanced Baseline Imager (ABI), a joint effort by the National Oceanic and Atmospheric Administration (NOAA) and the National Aeronautics and Space Association (NASA) on board the Geostationary Operational Environmental Satellites (GOES), we argue that the GOES constellation – commonly used as weather satellites – represents an underexplored opportunity for environmental science in situations where spatial resolution can be compromised in favor of more frequent imagery, especially now that ABI's spectral sensitivity has approached that of Landsat and MODIS (Table 1; Fig. 1). Given the constellation of geostationary satellites around the world, extending environmental science applications to ABI and generating relevant data products is crucial for achieving near-global coverage of satellite environmental data available at the timescale of minutes. Developing algorithms that can be applied to data from multiple geostationary satellites will be an important component for achieving near-global coverage.
A comparison of the spectral sensitivity of the Advanced Baseline Imager (ABI) with the Landsat Operational Land Imager (OLI) and MODIS Terra (McCorkel et al., 2020).
Here, we detail a number of applications by which GOES and other
geostationary satellites have enhanced or could enhance our understanding of
ecological phenomena that occur at timescales as short as minutes. We keep
our focus on the GOES constellation, but we also discuss research and
applications from various satellites in the global constellation of
geostationary satellites to provide the larger context for emerging
applications from GOES ABI. We outline the technical steps necessary to make
imagery from ABI more useful for environmental science, with an eye toward
near real-time monitoring of environmental phenomena across the globe. We
also note complementarity between GOES and other geostationary platforms,
including Japan's Himawari-
Instrument characteristics for the GOES-R Advanced Baseline Imager,
Advanced Himawari Imager on Himawari-
Geostationary remote sensing began with the launch of six Applications Technology Satellites (ATS), starting in 1966 (Suomi and Parent, 1968). The subsequent successful launches of the Synchronous Meteorological Satellites (SMSs) were the precursor to the GOES mission, which began in 1974 and continues to the present (Menzel, 2020). By 1979, the global constellation of geostationary satellites included the European Space Agency's Meteosat, the Japanese Geostationary Meteorological Satellite (GMS), and two GOES (Menzel, 2020). In total, 17 GOES have been successfully sent to space as of 2018, two of which – GOES-16, positioned at 75.2 W (currently GOES-East), and GOES-17, positioned at 137.2 W (currently GOES-West) – arise from the GOES-R series that include additional visible and near-IR channels that are commensurate with channels observed by Landsat and MODIS (Table 1; Fig. 1). The GOES-R series has an expected operational lifetime to 2036, which promises multiple years of continuous data availability. This will be followed by the Geostationary and Extended Orbits program, which is planned for operation in 2030–2050 and is anticipated to include “GOES-R-class” imagers (Sullivan et al., 2020).
The ABI is the primary instrument on board GOES-
In late April 2018, an issue with the GOES 17 ABI cooling system was detected due to malfunctioning of the loop heat pipe which transfers heat from the ABI detectors and helps maintain adequate temperatures for proper functioning (Yu et al., 2019). This resulted in the loss of infrared data during some nighttime hours, around 13:30 UTC, during parts of the year due to the Sun heating the seven ABI detectors faster than they can be cooled, resulting in infrared emissions from the overheated detectors (NOAA and NASA, 2020; NOAA, 2020). This nighttime data loss can also fluctuate seasonally, depending on how much solar radiation the instrument absorbs (NOAA, 2020). A data quality flag in the metadata of the Level 1b Radiances and Level 2 Cloud and Moisture Imagery can identify the faulty nighttime data (NOAA and NASA, 2020). This malfunction will result in the loss of some nighttime data for data products that utilize the infrared bands and are relevant for observing the full diurnal cycle.
The ABI collects top-of-atmosphere (TOA) data from a given location at a constant view zenith angle (VZA) and varying solar zenith angles (SZA) throughout the day. While the increased sampling of SZA creates opportunities in surface bidirectional reflectance factor (BRF) modeling (Ma et al., 2020), the large VZA at off-nadir locations can present challenges for studying the land surface, including coarsened resolution, potentially degraded locational accuracy, and more complex atmospheric compensation due to longer slant paths. Below, we review the existing efforts to address these challenges and make ABI imagery more suitable for studying the land surface.
The geolocation accuracy of ABI on GOES-16 and GOES-17 has been tracked and
improved throughout its provisional and operational stages using the Image
Navigation and Registration (INR) Performance Assessment Tool Set (IPATS; Tan et al., 2018, 2019, 2020). IPATS quantifies the navigation error, i.e., the
difference between the location of a pixel in ABI imagery and a reference
location (Tan et al., 2020). Some of the largest navigation errors,
calculated from correlations between subsets of ABI and Landsat 8 imagery
concentrated along the coast of North and South America, were 10–13
Geostationary satellites observe most of the hemisphere at an angle relative
to the zenith, which introduces a challenge due to parallax, i.e., the effect of observing an object from a large VZA. Parallax can result in uncertainties
in land surface observations in mountainous terrain and can introduce errors
in the mapped location of clouds (Bieliński, 2020). These errors vary by
VZA and the height of the feature and are largest for high VZA and high
feature altitude relative to the surface (Bieliński, 2020; Zakšek et
al., 2013). For example, the parallax shift at 49
Correcting for atmospheric attenuation of radiation to derive surface reflectance from TOA reflectance is a crucial prerequisite for studying surface processes from satellite platforms. Current efforts to estimate surface reflectance from the ABI, the Advanced Himawari Imager (AHI), and the Geostationary Ocean Color Imager (GOCI) on board the Communication, Ocean, and Meteorological Satellite (COMS) include generating lookup tables from the Second Simulation of the Satellite Signal in the Solar Spectrum (6S) radiative transfer model (He et al., 2019; Tian et al., 2010; Vermote et al., 1997; Yeom and Kim, 2015; Yeom et al., 2018, 2020). Optimal estimation methods that estimate surface BRF from SEVIRI have been extended to estimate surface broadband albedo and surface reflectance from the ABI and the AHI on Himawari-8 (Govaerts et al., 2010; He et al., 2019, 2012; Wagner et al., 2010). These algorithms estimate surface reflectance and broadband surface albedo by minimizing the difference between TOA BRF estimated through radiative transfer modeling and measured by the satellite (He et al., 2019, 2012). Unlike the surface reflectance algorithm currently used for SEVIRI, the algorithm for the ABI and the AHI takes the diurnal variation in aerosol optical depth into account (He et al., 2019). Originally developed for atmospheric correction of MODIS imagery, the Multi-Angle Implementation of Atmospheric Correction (MAIAC) has also been adapted to provide provisional daytime surface reflectance every 10 min for bands 1–6 of the AHI with plans to extend the algorithm to ABI (Li et al., 2019b). The surface reflectance from the AHI showed less variation compared to surface reflectance from MODIS and the differences in surface reflectance between the AHI and MODIS were smaller for the red, near-infrared (NIR), and shortwave infrared (SWIR) bands compared to the blue and green bands (Li et al., 2019b). ABI channels 1, 3, 5, and 6 are accurate to within 2 %, but channel 2 has a bias error of up to 5 % (McCorkel et al., 2020). With geolocation, parallax, atmospheric correction, and sensor accuracy taken into account, imagery from the ABI can provide sub-hourly estimates of various land surface variables.
Atmospheric attenuation due to water vapor requires atmospheric correction
of thermal data collected from satellites and also limits surface
temperature retrieval to thermal bands that have the lowest atmospheric
absorption (Sun and Pinker, 2003). A single-channel approach requires the
use of one thermal channel within an atmospheric window at around 10
Similar to surface reflectances, the directional thermal radiation recorded by sensors on satellites can also be impacted by the VZA of the sensor (Diak and Whipple, 1995). Products that utilize diurnal thermal data from GOES have used the difference in surface radiometric temperature during the morning hours, which has shown to be less sensitive to changes in VZA compared to absolute surface radiometric temperature (Anderson et al., 1997; Diak and Whipple, 1995). Other methods that address the impacts of varying VZA include adding zenith angle correction terms to split-window algorithms in order to address the large path lengths at high VZA (Sun and Pinker, 2003; Yu et al., 2009).
Geostationary satellites can now measure a number of common vegetation indices used for ecological applications and make measurements that support derived products, including land surface temperature, as noted, incident solar radiation, and more. We describe these measurements with an eye toward explaining the benefits and challenges of using geostationary platforms such as the ABI for ecology and environmental science.
Geostationary satellites are equipped to estimate incident solar radiation at the Earth's surface (Diak, 2017; Pinker et al., 2002), critical for the surface energy balance, photosynthesis, and solar power applications. Earlier efforts to do so include a simple physical model by Gautier et al. (1980), which estimated incident solar radiation during clear and cloudy conditions using the reflectance from the visible band of the Visible Infrared Spin-Scan Radiometer (VISSR) on GOES-2. The model included Rayleigh scattering and water vapor absorption of downwelling and reflected shortwave radiation. Cloud albedo and absorption were estimated from a linear relationship with the satellite-measured cloud reflectance, and estimates of incident solar radiation were subsequently improved by including ozone absorption of shortwave radiation in the atmosphere (Diak and Gautier, 1983; Diak, 2017). Continued improvements in both the physical model and the spatiotemporal resolution of the GOES imagery have resulted in higher accuracy of hourly and daily insolation estimates when compared to pyranometer measurements (Diak, 2017; Otkin et al., 2005). More recent models of the transmittance of direct and diffuse shortwave radiation through aerosol extinction by different aerosol components, gaseous absorption, and Rayleigh scattering have provided estimates of the surface shortwave radiation flux and its diffuse fraction from SEVIRI observations at 15 min resolution (Carrer et al., 2019) and could also be applied to ABI observations. Applying algorithms for estimating incident solar radiation to data from multiple geostationary satellites can lead to near-global coverage and be beneficial for near-global estimates of evapotranspiration and gross primary productivity that are driven, in part, by solar radiation.
Incident solar radiation in the wavelengths of photosynthetically active radiation (PAR; 400–700 nm) can also be estimated using the visible bands of geostationary satellites (Janjai and Wattan, 2011). Specifying a range of SZA, VZA, cloud types, aerosol types, cloud extinction coefficient, and atmospheric visibility, lookup tables generated from simulations of Moderate Resolution Atmospheric Transmission (MODTRAN) have been used to estimate downwelling PAR from at-sensor radiance (Zhang et al., 2014; Zheng et al., 2008). These methods have been extended to multiple geostationary satellites, including GOES-11 and GOES-12 and MODIS surface reflectance data, to generate global incident PAR estimates (Zhang et al., 2014).
Although these methods account for elevation, validation efforts have shown that satellite-derived PAR in high-altitude areas can be biased when compared to ground measurements, possibly due to the inaccurate specification of atmospheric profiles governing water vapor corrections (Zhang et al., 2014). Furthermore, PAR estimated from satellites has been reported to underestimate PAR measured on the ground when the model assumed urban aerosol absorption over areas where maritime aerosols were more dominant (Janjai and Wattan, 2011). Despite these limitations, frequent estimates of PAR and incident solar radiation from geostationary satellites may be uniquely suited to drive the land surface models that are operating at increasingly fine spatial and temporal resolutions, providing a natural link for using geostationary satellite observations to improve our understanding of the carbon, water, and energy cycles (Williams et al., 2009).
Terrestrial photosynthesis is particularly responsive to diffuse PAR, which penetrates plant canopies more efficiently than direct PAR (Emmel et al., 2020; Gu et al., 2003). The diffuse fraction of incoming PAR is well-described as a linear function of transmissivity, or a clearness index, through the atmosphere within certain inflection points (Erbs et al., 1982; Oliphant and Stoy, 2018; Weiss and Norman, 1985). Estimates of cloud height, optical depth, and particle size, along with aerosols from GOES, can be used to further partition incoming PAR into direct and diffuse beam fractions as currently provided by EUMETSAT at 15 min intervals (Carrer et al., 2019). Such observations could ultimately prove useful for analyses of the diurnal pattern of carbon cycling (see Sect. 6.1), given that the variability in terrestrial carbon cycling is often most sensitive to the variability in PAR at timescales from minutes to days (Stoy et al., 2005).
The normalized difference vegetation index (NDVI) – the normalized difference between the reflectance in red and near-infrared wavelengths – is strongly correlated to chlorophyll content, green biomass, leaf area index (LAI), and the fraction of incoming PAR absorbed by leaves (fAPAR; Gamon et al., 1995; Jordan, 1969; Rouse et al., 1974; Running et al., 1986; Tucker, 1979; Tucker et al., 1985) and, therefore, is a critical variable for monitoring the land surface. The ABI also has the ability to measure the Enhanced Vegetation Index (EVI) which is beneficial in areas (and periods) of dense vegetation cover where, unlike NDVI, EVI does not saturate and in open canopy areas because of a correction factor applied for canopy background (Huete et al., 2002; Zhou et al., 2014). The near-infrared reflectance of vegetation (NIRv) is strongly correlated to the amount of incoming PAR absorbed by, plants and therefore, photosynthesis at half-hourly to annual timescales and has shown stronger relationships with photosynthesis compared to NDVI (Badgley et al., 2017, 2019; Baldocchi et al., 2020; Wu et al., 2020) and can also, in principle, be measured by GOES (Table 1).
LAI from SEVIRI is produced on a daily and 10 d basis through the LSA SAF program (Trigo et al., 2011). High temporal estimates of LAI from ABI will have widespread utility in the Western Hemisphere by providing an important variable needed for modeling seasonal vegetation dynamics and energy, water, and carbon fluxes (Anderson et al., 2011; Guan et al., 2014; Robinson et al., 2018). An ABI LAI product can provide harmony in temporal scales and data sources needed to estimate the fractional vegetation cover needed for a two-source energy balance model used for estimating evapotranspiration from GOES thermal data (further discussion in Sect. 5.1; Anderson et al., 2011). Similarly, the ABI LAI product can provide a data source for plant respiration modeling (see further discussion in Sect. 6.1; Robinson et al., 2018).
The increased temporal frequency of measurements available from geostationary satellites compared to polar-orbiting satellites provides more opportunities for measuring NDVI, EVI, LAI, and NIRv in areas with frequent cloud cover (Miura et al., 2019). However, the geostationary position captures reflected radiation at varying SZA throughout the day, and these novel Sun–sensor geometries, not previously captured by polar orbiting satellites, can cause diurnal variation in vegetation indices calculated from TOA reflectance (Tran et al., 2020). EVI shows less SZA-induced diurnal variation compared to NDVI and is less impacted by the midday hot spot effect during times of the year (spring and autumn equinox) when the SZA and AHI VZA are aligned (Tran et al., 2020). NDVI measurements can be normalized to a reference Sun–target–sensor geometry by estimating bidirectional reflectance distribution functions (BRDFs) to address the impacts of varying Sun–sensor geometry (Fensholt et al., 2006; Seong et al., 2020; Tian et al., 2010; Yeom and Kim, 2015; Yeom et al., 2018). Given the SZA sensitivity of NDVI measurements, Wheeler and Dietze (2019) demonstrated a Bayesian model to estimate a daily midday NDVI value from diurnal NDVI calculations using ABI observations.
Liquid water absorption influences reflectance by plants in the atmospheric
windows of shortwave infrared (SWIR) wavelengths (1.3–2.5
The ABI, along with the Advanced Meteorological Imager (AMI) on
GEO-KOMPSAT-2A, the AHI, and SEVIRI all offer bands in the SWIR regions,
with ABI band 5 placed in the refined interval of 1.55–1.75
The ABI features three longwave infrared bands with spatial resolutions of about 2 km for measuring land surface temperature (LST) – the skin temperature of the uppermost layer of the land surface – including correction for atmospheric moisture (Yu et al., 2012). Since the emission from the land surface diverges from a blackbody, the knowledge of land surface emissivity is a crucial component for the retrieval of LST from at-sensor radiance (Li et al., 2013; Sun and Pinker, 2003). Global emissivity data may be gathered by consulting compiled tables of emissivities for various land covers along with land cover classifications of the landscape (Li et al., 2013). Land surface emissivity can also be estimated through its relationship with the NDVI. This method only applies to vegetation and soil and requires knowledge about the fractional cover of vegetation in a pixel (Li et al., 2013). Furthermore, land surface emissivities can be estimated by using surface temperature–emissivity separation techniques applied to multispectral thermal satellite observations (Li et al., 2013). Data sources for land surface emissivity in LST retrievals from geostationary satellites can include spectral libraries given a specific type of environmental surface (Peres and DaCamara, 2005), the MODIS operational land surface emissivity product (MOD11), or the Combined ASTER and MODIS Emissivity over Land (CAMEL) product (Pinker et al., 2019).
The hourly ABI LST product uses the difference between the brightness
temperatures of ABI bands 14 (11.2
For the generation of a consistent, long-term record of LST, a single channel
approach has also been proposed for LST retrieval from GOES 12 channel 4
(10.7
Diurnal observations from GOES provide multiple estimates of directional surface radiometric temperature and downwelling solar radiation each day to estimate water and energy fluxes from the soil and canopy (Diak and Stewart, 1989). One approach for estimating evapotranspiration (ET) that is well suited for geostationary implementation is the Atmosphere–Land Exchange Inverse (ALEXI) model (Anderson et al., 1997, 2007a; Mecikalski et al., 1999), which estimates the bulk surface energy balance (net radiation, sensible heat flux, latent heat flux, and soil heat flux) and the nominal partitioning of these fluxes between the soil and canopy. ALEXI is a time-integrated model based on the two-source (vegetation and soil) energy balance (TSEB) approach of Norman et al. (1995). ALEXI models the growth and sensible heating of the atmospheric boundary layer based on the morning rise in surface radiometric temperature that can be measured by GOES, estimating time-integrated latent heat flux (or ET, in units of mass flux) as a residual to the overall energy balance (Anderson et al., 1997). The model performs best when the insolation inputs are also derived from geostationary satellite data (Sect. 4.1), giving optimal spatial and temporal correspondence between net radiation forcings and surface temperature response signals. In comparison with other sources of insolation data, geostationary-based insolation could significantly improve ET retrievals, particularly in areas of frequent cloud cover where reanalysis estimates may not accurately capture the timing and spatial extents of clouds (Anderson et al., 2019; Wonsook et al., 2020).
ALEXI-based ET estimates are produced routinely at 4 km resolution for the United States (Anderson et al., 2020). Also, daily 2 km resolution ALEXI-based ET estimates have been generated from ABI observations as part of the GOES ET and Drought (GET-D) product system (Fang et al., 2019). A surface energy balance approach has also been used to estimate 30 min ET from albedo and downwelling radiation from MSG SEVIRI over the areas covered by MSG and 3 h ET in the Haihe River Basin in China from hourly LST observations from MTSAT (Multifunctional Transport Satellites), a Japanese geostationary satellite (Ghilain et al., 2011; Zhao et al., 2019). By measuring LST, geostationary satellites can estimate sensible heat flux and, therefore, also the Bowen ratio, which can give insight into atmospheric boundary layer heat and moisture transport, as well as plant water stress (Diak and Whipple, 1995). Applications of ALEXI-based ET and energy fluxes for drought monitoring and modeling carbon fluxes are discussed below.
Drought indicators based on remotely sensed thermal observations can improve the effectiveness of drought early warning systems due to their high spatial resolution and the tendency for large decreases in ET to precede visible reductions in vegetation biomass during early stages of drought development (Anderson et al., 2013a; Otkin et al., 2015). The Evaporative Stress Index (ESI; Anderson et al., 2013a) is a drought indicator based on standardized anomalies in the actual-to-reference ET ratio, where actual ET is retrieved with ALEXI using the morning LST rise signal, typically obtained from GOES. ESI has demonstrated the ability to provide early signals of developing vegetation stress (Anderson et al., 2007b, 2013a, 2016; Otkin et al., 2015, 2018a).
A recent prominent application of the ESI has been in the detection of flash droughts (Otkin et al., 2013). Flash drought conditions are characterized by a period of rapid drought intensification and typically include warm air temperature and low cloud cover anomalies, with dew point suppressions and high winds that can increase ET and hasten the removal of water from ecosystems (Gerken et al., 2018; Otkin et al., 2014, 2016, 2018b). ESI has proven to be an effective indicator of moisture stress in vegetation and the onset of flash drought conditions (Otkin et al., 2014, 2016, 2018b). For example, rapid temporal changes in the ESI toward increasing vegetation stress appeared several weeks earlier than the point at which the U.S. Drought Monitor (USDM) classified regions of the central United States to be experiencing moderate to exceptional drought in 2003 and 2012 (Otkin et al., 2014). The ESI was also able to capture the onset of vegetation stress and the subsequent vegetation recovery during the flash drought and flash recovery sequence of 2015 in the south central United States (Otkin et al., 2019).
Drought indices based on precipitation and atmospheric demand highlight areas with the potential for vegetation stress, but these stresses may not materialize due, for example, to beneficial rainfall, management (e.g., irrigation), or plant root access to groundwater. ESI uses LST to diagnose actual stress materializing on the ground and, therefore, has been used as a moisture stress indicator for estimating drought impacts on crop yields (Anderson et al., 2016; Mladenova et al., 2017). ESI is routinely generated at 4 km resolution over the CONUS, and 5 km globally, or can be downscaled to sub-field or stand scales (30 m) using higher resolution thermal data from Landsat (Yang et al., 2018, 2020). The ability of ESI to detect drought stress earlier than USDM and other indices is shown in Fig. 2 (adapted from Anderson et al., 2013b).
Comparison of drought evolution between the U.S. Drought Monitor classification, the Evaporative Stress Index (ESI), and the Vegetation Drought Response Index (VegDRI). The figure is adapted from Anderson et al. (2013b) and distributed under a CC BY-NC-ND-3.0 license.
The Automated Biomass Burning Algorithm (ABBA) was developed from the 4 and 11
While GOES-R thermal observations can provide biomass burning emissions at a fine temporal scale, the coarse spatial resolution of GOES-R presents a challenge in detecting small sub-pixel fires and emissions. Differences between medium (20 m) and coarse resolution (500 m) imagery can result in substantial differences in total detected burned area, estimated emissions, and the length of the fire season (Ramo et al., 2021). Small fires can make up a great portion of total burned area and emissions, and they can result in a lengthening of the fire season in regions where anthropogenic fires are prevalent (Ramo et al., 2021). Similar to other coarse-spatial-scale emissions datasets, emissions from GOES-R should be considered conservative in areas with substantial undetected small fires (Ramo et al., 2021). Similar to Ramo et al. (2021), studies comparing biomass burning emissions from GOES-R with emissions from finer spatial resolution satellite imagery should reveal the magnitude of differences and trade-offs between high temporal and spatial resolution in estimating emissions.
Plant canopies have unique and observable events that occur annually as a part of their phenology. The phenology of photosynthesis and plant growth is sensitive to temperature, precipitation, and photoperiod (Bauerle et al., 2012; Fu et al., 2017; Piao et al., 2019; P. C. Stoy et al., 2014), and shifts in the phenology of carbon uptake and plant growing season in response to changing climate have important implications for ecosystems (Bradley et al., 1999; Xu et al., 2020). These shifts often occur on timescales that cause uncertainty from polar-orbiting satellites, especially when cloudy conditions are present during spring in the temperate zone (Richardson et al., 2013) and dry-to-wet (and wet-to-dry) seasonal transitions in tropical forests (Ganguly et al., 2010). Research on land surface phenology to date has used a combination of satellite remote sensing and near-surface remote sensing via webcams to detect seasonal transitions in vegetation greenness and photosynthesis, such as the start, peak, and end of the growing season (Dannenberg et al., 2020; Gamon et al., 2016; Seyednasrollah et al., 2019; Wong et al., 2019; Zhang et al., 2003). These observations have varying spatial and temporal resolutions, depending on the method and instrumentation used (Brown et al., 2016; Filippa et al., 2018; Liu et al., 2017).
Geostationary satellites such as GOES have unique capabilities that could further enhance plant phenology research. Compared to polar-orbiting satellites, the large number of diurnal observations from geostationary satellites capture greater variation in sun-angle geometries. This increased variability allows for better BRDF adjustments and improved investigations about the impact of the SZA on the vegetation indices used for extracting phenological transitions (Ma et al., 2020). Time series of LAI, NDVI, and the two-band Enhanced Vegetation Index (EVI2) from SEVIRI, AHI, and GOCI show increased observations during cloudy conditions and improved estimates of phenological cycles and transitions (Guan et al., 2014; Miura et al., 2019; Yan et al., 2016; Yeom and Kim, 2015; Yeom et al., 2018). Figures of time series from various papers cited above demonstrate the value of geostationary imagers in capturing greenness trajectories as a complement to polar-orbiting satellites. NDVI from BRDF-adjusted reflectance from GOCI has demonstrated an improved ability to monitor the growth of rice paddies in North Korea and South Korea compared to MODIS NDVI from BRDF-adjusted reflectance, especially during the monsoon season which has frequent cloud cover that limits the ability of polar-orbiting sensors to observe the surface (Yeom et al., 2015, 2018). In the Congo Basin, the multiple annual phenological cycles of greenness in evergreen broadleaf forests were better captured by the increased observations from SEVIRI compared to MODIS (Yan et al., 2016). In Japan, the greenness trajectories from an EVI2 time series revealed differences in the length of seasonal transitions (timing between start of spring to end of spring) between AHI and MODIS (Yan et al., 2019).
GOES-R can also provide increased observations for the remote sensing of dryland phenology, which can include multiple growing cycles per year, and where phenological transitions can be triggered by pulses of rainfall and present an ongoing challenge for the remote sensing of land surface phenology (Smith et al., 2019). However, since drylands feature heterogeneous vegetation, studies will need to investigate whether increases in temporal resolution with coarse spatial resolution are enough to offer an improvement in dryland phenology (Smith et al., 2019). Leveraging diurnal observations from GOES to estimate greenness trajectories and phenological transitions (Hashimoto et al., 2021; Wheeler and Dietze, 2021) across the Western Hemisphere, coupled with the ability to extract these transitions from AHI, SEVIRI, and GOCI, can result in a near-global improvement in estimating seasonal vegetation growth and decline. The capacity of GOES to track events in plant phenology is shown in Figs. 3 and 4 for pixels within and outside of the Kincade fire scar, with notable increases in vegetation greenness and moisture during March and April at the end of the typical rainy season in California's Mediterranean ecosystems.
Geostationary satellites have been used for nearly a decade for monitoring
the dynamics of ocean color. The ocean color signal can identify suspended
particulate matter and phytoplankton (Neukermans et al., 2009; Ruddick et
al., 2014), including harmful algal blooms (Choi et al., 2014; Noh et al.,
2018), and may be a sentinel for the impacts of climate change on marine
ecosystems (Dutkiewicz et al., 2019). Most research to date has involved the
Geostationary Ocean Color Imager (GOCI), which has transmitted eight images
per day since 2010 in six visible (412, 443, 490, 555, 660, and 680 nm) and
two infrared channels (745 and 865 nm), with 20 nm bandwidth at 500 m spatial resolution centered around the Korean Peninsula at 128.2
Estimates of surface–atmosphere carbon flux from polar-orbiting instruments like MODIS are usually produced on 8 d to annual time steps (Zhao et al., 2005). The impact of rapidly evolving meteorological conditions on terrestrial carbon uptake has gained recent attention, suggesting that more frequent observations will improve our understanding of the carbon cycle. Precipitation events and the resulting short-term changes in meteorological conditions on the order of days result in local anomalies in canopy photosynthesis and respiration that influence seasonal ecosystem exchange (Randazzo et al., 2020). Fluctuations in carbon uptake can result from upwind climate extremes through heat and moisture advection, revealing more complexity in how climate extremes impact ecosystem carbon fluxes (Schumacher et al., 2020). Smoke from large wildfires can result in a short-term decrease in incoming PAR but an increase in the diffuse fraction of incoming PAR which, under the right circumstances, can increase seasonal carbon uptake through changes in light-use efficiency (LUE; Hemes et al., 2020). The resulting daily anomalies in gross primary productivity (GPP) from sudden changes in limiting resources have been shown to disproportionately affect longer-term ecosystem carbon uptake (Kannenberg et al., 2020). Multiday positive anomalies in GPP are critical for explaining its interannual variation at ecosystem and global scales (Fu et al., 2019; Zscheischler et al., 2016). All of these recent findings point to the importance of more frequent observations of ecosystem carbon cycling to improve our understanding of global carbon cycling.
Now that geostationary satellite imagers, such as the ABI, measure similar
spectral bands to MODIS (Fig. 1; Table 1), they can be used to monitor the
carbon cycle in similar ways but at higher temporal frequency. Estimates of
GPP often rely on LUE models, which are rooted in a linear relationship
between absorbed PAR (APAR) and net primary production (Medlyn, 1998;
Monteith, 1972). An ideal LUE in the absence of environmental stresses is
specified and attenuated with the use of stress functions that describe the
relationship between LUE and environmental stressors (Anderson et al., 2000;
Mahadevan et al., 2008; Robinson et al., 2018; Running et al., 2004; Yuan et
al., 2007; Zhang et al., 2016; Zhao et al., 2005). The most widely used
environmental stressors include functions to describe temperature and
moisture stress on LUE. Multiple approaches for estimating GPP from space
exist based on the LUE approach, with differences arising from the
spatiotemporal resolution of the inputs, the meteorological data used,
incorporating the impacts of CO
Vegetation indices calculated from GOES-R observations will provide spatiotemporal harmony with other GOES-R inputs, such as downwelling shortwave radiation, in estimating APAR. Various remotely sensed vegetation indices have been used for both estimating fAPAR to estimate APAR and in formulations of environmental stresses on LUE. The MODIS GPP algorithm uses NDVI and the MODIS fAPAR/LAI product to estimate APAR. The Vegetation Photosynthesis and Respiration Model (VPRM; Mahadevan et al., 2008) uses a similar approach and estimates the gross ecosystem exchange (GEE; similar to GPP) using the Enhanced Vegetation Index (EVI) instead of NDVI. The land surface water index (LSWI), the normalized difference between satellite-derived reflectance in near-infrared and shortwave infrared, is used to adjust LUE in response to water stress and leaf phenology (Mahadevan et al., 2008).
Normalized difference vegetation index (NDVI) calculated using TOA
reflectance factor from ABI on GOES-16 and Nadir Bidirectional Reflectance
Distribution Function-Adjusted Reflectance (NBAR) from MODIS on the Terra
and Aqua satellites for pixels inside and outside the Kincade fire (23 October 2019) perimeter in northern California for the period March 2019 to March 2020. The GOES Clear Sky Mask algorithm was applied to observations from GOES. MODIS NBAR with good quality flags were used. All observations are daytime, with a solar zenith angle of
NDII calculated using the TOA reflectance factor from ABI on GOES-16 and NBAR from MODIS on the Terra and Aqua satellites for pixels inside and outside the Kincade fire (23 October 2019) perimeter in northern California for the period March 2019 to March 2020. The GOES Clear Sky Mask algorithm was applied to observations from GOES. MODIS NBAR with good quality flags were used. All observations are daytime, with a solar zenith angle of
Implementing a model to estimate carbon uptake from ABI presents opportunities to improve LUE-based models by using emerging variables, as opposed to the commonly used air temperature and vapor pressure deficit, to represent environmental stressors on LUE such as soil moisture, diffuse radiation, LST, and the evaporative fraction (Anderson et al., 2000; Li et al., 2021; Yuan et al., 2007; Zhang et al., 2016). Ecosystem GPP increases with increases in diffuse radiation if light does not limit photosynthesis because diffuse radiation penetrates plant canopies more readily, resulting in a more even distribution of light among shaded and sunlit leaves (Hemes et al., 2020; Mercado et al., 2009). Incorporating the diffuse component of incoming PAR has been noted as a priority for improving LUE models (McCallum et al., 2009; Yuan et al., 2014). Recent attempts at incorporating diffuse radiation into LUE models as a stress on GPP have demonstrated an enhancement of LUE during overcast skies (Zhang et al., 2016). Fog events can be identified using GOES and increase both diffuse fraction of radiation and ecosystem water use efficiency (Baguskas et al., 2021). Partitioning estimates of PAR from ABI into diffuse and direct components, along with the ABI LST product and ALEXI-based estimates of sensible and latent heat flux, can provide harmony between inputs for environmental stressors on LUE, PAR, and vegetation indices by not having to rely on estimates from sources other than ABI. Furthermore, estimates of LAI from GOES can contribute to respiration modeling data needs for estimating net primary production (NPP) (Robinson et al., 2018).
Estimating GPP with ABI data can also benefit from various LUE model formulations that incorporate the coupling between carbon uptake and water loss through stomatal processes and water use efficiency (Anderson et al., 2000; Zhang et al., 2016). Estimates of APAR from ABI can be used in existing carbon–water coupled models through the ALEXI framework, which can estimate hourly fluxes which respond to diurnal variations in solar radiation and vapor pressure deficit (VPD) along with capturing the effect of soil moisture stress on LUE (Anderson et al., 2000). The ALEXI framework presents an opportunity to use both ABI thermal data, as previously mentioned in Sect. 5.1, and ABI reflectance data to estimate carbon, water, and energy fluxes simultaneously (Anderson et al., 2008). Diurnal estimates of carbon uptake and water loss from ABI will enhance our understanding of ecosystem water use efficiency and its response to environmental variation across geographic space. Furthermore, there are various opportunities to study the response of carbon fluxes estimated from ABI data to soil moisture variability with soil water content measured at various ground locations. This could be an important investigation considering the noted limitations of satellite GPP estimates in their ability to capture soil moisture effects on carbon uptake (Stocker et al., 2019). Observations from geostationary satellites could also, in principle, be used to assimilate instantaneous carbon flux observations from eddy covariance towers into ecosystem models (LeBauer et al., 2011).
TOA NDVI calculated using the TOA reflectance factor from GOES-16 imagery for
TOA NDVI, near-infrared reflectance of vegetation (NIRv), Enhanced Vegetation Index (EVI), and NDII calculated using the TOA reflectance factors from GOES-16 imagery for DOY 235 (22 August 2020) at 16:00 UTC, DOY 246 (2 September 2019) at 16:00 UTC, and their difference around the advisory track for Hurricane Laura (red line) in southern Louisiana (26 August 2020; DOY 239). Gray pixels represent cloudy pixels from the GOES Clear Sky Mask algorithm or vegetation index values below 0 for NIRv and EVI.
Remotely sensed data have made it possible to map regional burn severity from wildfires, while time series of measures of greenness and primary productivity enable monitoring of vegetation disturbance and recovery (Bolton et al., 2017; Goetz et al., 2006; Meng et al., 2018). Mapping damage from hurricanes and the resulting impact to the carbon cycle has also been possible, using remotely sensed data, through the use of various vegetation indices related to greenness and moisture (Chambers et al., 2007; Wang et al., 2010). The difference in NDVI from pre- and post-fire imagery from GOES-16 shows the potential to detect variations in burn severity and recovery from wildfires, as demonstrated for the notable Kincade fire in 2019 in California (Fig. 5). High-frequency estimates of NDVI (Fig. 3) and NDII (Fig. 4) from GOES-16 can track vegetation disturbance and recovery from wildfire in a similar manner to MODIS, although the magnitude of the vegetation indices differs between the two due to the atmospheric corrections and BRDF adjustments of the MODIS TOA reflectance (Figs. 3 and 4). Variations in SZA and VZA can lead to differences in NDVI values between GOES, MODIS, and other sensors (Fensholt et al., 2006; Pinker and Ewing, 1985), with MODIS having the advantage of a low SZA during its overflight. These differences in viewing geometry need to be resolved to reconcile differences between sensors. Imagery from ABI also have the potential to estimate the regional decline in vegetation indices after hurricanes (Fig. 6), which can potentially be converted into the decline in GPP using methods described in Sect. 6.1. A recent review has also highlighted the use of geostationary imagers in monitoring landslides and flooding (Higuchi, 2021).
Observations of the diurnal behavior of land surface attributes can also be used to improve our understanding of the thermal properties of ecosystems. Large-scale changes to ecosystems, such as logging, the conversion of grasslands or forests to farmland, and wildland fires alter the mechanisms by which ecosystems regulate their energy acquisition and heat exchange and are thought to be related to the degree of complexity of the system (Kay and Schneider, 1992). The process of ecological succession modifies the structure of an ecosystem over time by gradually increasing its complexity (Odum, 1969). Ecosystems should, therefore, develop an increasing ability to dissipate incoming solar radiation as succession proceeds (Schneider and Kay, 1994) and, due to a higher order of complex structures, will improve cooling by transferring solar radiation into latent heat and metabolic energy (Norris et al., 2011). Effective energy dissipation results in a cooler surface (with less waste heat) for biological processes to occur. Diurnal patterns of LST can be used to quantify ecological complexity (Lin et al., 2009) and entropy production (Brunsell et al., 2011; Holdaway et al., 2010; P. Stoy et al., 2014; Wiesner et al., 2019) and may be useful for monitoring the success of ecological restoration projects that seek to re-establish ecological function (Aerts et al., 2004).
One of the most promising applications of geostationary satellites for land surface science may simply be their ability to provide temporally dense observations that can be combined (fused) with spatially dense observations from polar-orbiting or other satellite platforms. Such fusion products take advantage of the best features of different platforms, for example, by exploiting the finer spatial resolution of observations from Landsat, Sentinel-2, MODIS, VIIRS, or the recently harmonized Landsat Sentinel-2 surface reflectance and the temporal resolution of GOES observations. Based on the assumption that changes in a Landsat pixel are equivalent or proportional to the changes in the co-located MODIS pixel between two dates, fusion algorithms are able to predict the surface reflectance of a Landsat pixel between acquisition dates using the change observed in MODIS pixels between the reference and prediction date (Gao et al., 2015). Multisensor fusion, using polar-orbiting and geostationary satellites, has provided higher spatiotemporal resolution for estimates of forest disturbance, NDVI, phenology, LST, and ET and water management, yield estimation, and fire radiative power (Cammalleri et al., 2013, 2014; Gao et al., 2015; Hilker et al., 2009; Knauer et al., 2016; Li et al., 2019a; Semmens et al., 2016; Wu et al., 2015; Yang et al., 2018; Zhao and Duan, 2020). Given the importance of considering landscape heterogeneity in using satellite data to estimate various ecosystem variables and model processes such as LUE, carbon flux and storage, phenology, and ET (Ahl et al., 2004; Cammalleri et al., 2014; Leitão et al., 2018; Zeng et al., 2020), multisensor data fusion offers a way to address the relatively coarse spatial scale of imagery from geostationary satellites as we expand their use in studying ecosystem function and surface–atmosphere exchange.
The recently increased spectral and spatial resolution of imagers on board geostationary satellites creates new opportunities to use remotely sensed observations in environmental science, especially when combined with ongoing and forthcoming improvements to the spectral sensitivity of other geostationary imagers. The increased diurnal sampling from the ABI can find cloud-free observations when polar-orbiting satellites may be hindered by cloud cover to improve time series of vegetation greenness, canopy water content, LST, and energy–water–carbon fluxes. Temporally dense geostationary observations have been crucial in detecting vegetation moisture stress and wildfires. Extending the use of these observations to measure more land surface variables can capture sudden changes in ecosystem fluxes due to changing atmospheric conditions and extreme events. Furthermore, estimations of vegetation phenological cycles can be improved using geostationary satellites, especially by taking advantage of short periods of clear skies occurring during times of otherwise persistent cloud cover. Studying ecosystem disturbance events should also benefit through increased availability of observations pre- and post-disturbance to estimate burned area and vegetation recovery. Multisensor fusion between relatively coarse geostationary satellite observations and observations from platforms with finer spatial resolution offers a way to resolve within-pixel spatial variability in heterogeneous landscapes.
The ongoing efforts to improve the geolocation and radiometric quality of imagery and to provide higher level collections of data with surface reflectance, similar to the MODIS or Landsat program, will be a crucial component of using imagery from geostationary sensors. Furthermore, increased collaboration between the National Science Foundation, NASA, and NOAA, with increased funding opportunities to pursue research utilizing imagery from geostationary sensors for investigations in the science topics we have detailed here, will help realize the potential of these data. Leveraging the high-frequency observations from geostationary sensors for investigations in environmental science where they have been largely untapped will expand the scope of hyper-temporal remote sensing of the environment.
The GOES-16 ABI Level 2 Cloud and Moisture Imagery (ABI-L2-MCMIPC) used to
make the figures is available through
AMK and PCS conceived the study, and all the authors contributed to writing the paper. AMK and PCS wrote Sects. 1 and 2.1. AMK wrote Sects. 2.2–5.4, 6.1–6.2, 6.4, and 7. PCS and GD contributed to Sect. 4.1. JAO, MA, and PCS contributed to Sects. 4.4–5.2. PCS wrote Sect. 5.5. JM and CH contributed to Sects. 1, 2, 3, and 7. PCS and EMR wrote Sect. 6.3. JTD contributed to Sect. 5.4. All the authors provided suggestions and edits. AMK made all the edits. AMK prepared Figs. 3–6 and Table 1.
The authors declare that they have no conflict of interest.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank Timothy J. Schmit and Scott Lindstrom, for the critical insight into the GOES mission, for the assistance with the data acquisition, and for the valuable comments on the paper. We also thank David Wood and Susi Wiesner, for the valuable comments on the paper.
Paul C. Stoy acknowledges contributions from the U.S. National Science Foundation and the University of Wisconsin – Madison.
This research has been supported by the National Science Foundation (grant nos. DEB 1552976 and EF 1702029).
This paper was edited by Alexandra Konings and reviewed by two anonymous referees.