Satellite Remote Sensing of Savannas: Current Status and Emerging Opportunities

Abstract

Savannas cover a wide climatic gradient across large portions of the Earth’s land surface and are an important component of the terrestrial biosphere. Savannas have been undergoing changes that alter the composition and structure of their vegetation such as the encroachment of woody vegetation and increasing land-use intensity. Monitoring the spatial and temporal dynamics of savanna ecosystem structure (e.g., partitioning woody and herbaceous vegetation) and function (e.g., aboveground biomass) is of high importance. Major challenges include misclassification of savannas as forests at the mesic end of their range, disentangling the contribution of woody and herbaceous vegetation to aboveground biomass, and quantifying and mapping fuel loads. Here, we review current (2010–present) research in the application of satellite remote sensing in savannas at regional and global scales. We identify emerging opportunities in satellite remote sensing that can help overcome existing challenges. We provide recommendations on how these opportunities can be leveraged, specifically (1) the development of a conceptual framework that leads to a consistent definition of savannas in remote sensing; (2) improving mapping of savannas to include ecologically relevant information such as soil properties and fire activity; (3) exploiting high-resolution imagery provided by nanosatellites to better understand the role of landscape structure in ecosystem functioning; and (4) using novel approaches from artificial intelligence and machine learning in combination with multisource satellite observations, e.g., multi-/hyperspectral, synthetic aperture radar (SAR), and light detection and ranging (lidar), and data on plant traits to infer potentially new relationships between biotic and abiotic components of savannas that can be either proven or disproven with targeted field experiments.

1. Introduction

1.1. The Ecogeography of Savannas

The savanna biome is characterized by the coexistence of woody (trees and shrubs) and herbaceous (grasses) plants (Figure 1). Savannas are biodiversity hotspots and habitat for several iconic species of megafauna and provide a home and livelihood for millions of people who live within or adjacent to them [1–3]. Savannas have a broad climatic scope as they receive mean annual rainfall amounts ranging from 300 to 2500 mm [4] and have mean annual temperatures between 14 and 30°C [5].

Figure 1 

Observing multiple drivers and processes in a savanna from space. Savannas have a distinct seasonality with alternating dry and wet seasons. They have a continuous layer of grass with sporadic woody cover. The grass layer provides fuel for extensive fires that coincide with the dry season, and there is widespread herbivory from grazing and browsing animals. Savannas are also subject to human activity such as livestock rearing by pastoral nomads, cropland expansion, and fuelwood harvesting. The illustrated satellites exemplify the main types of Earth observation systems that have been used to study savannas at large scales.

Savannas annually store between 36 and 42 tonnes of carbon per hectare [6], representing around a third of terrestrial net primary production, and comprise a critical regulating component of the land carbon sink [7, 8]. Much of this carbon is stored as aboveground biomass (AGB), which is the pool of photosynthesized carbon embodied within vegetation and is both an essential climate variable [9] and an important ecosystem service in savannas.

The climate of savannas is distinctly seasonal and alternates between a growing season during which most of the photosynthesis takes place [10] and a nongrowing season in which most of the herbaceous vegetation senesces and cures [11]. Tropical savannas have seasonal rainfall patterns with alternating wet and dry seasons. The dry season serves as the curing period for the grass layer, which accumulates and forms the fuel source for future fires [1]. However, unlike their tropical counterparts, temperate savannas do not have a distinctly alternating rainfall seasonality [1], and accumulation of fuel loads (i.e., combustible material) takes place in the winter when the temperatures are too low for grass growth [12].

Savannas contribute substantially to the seasonal variability of global greenhouse gas emissions from fires, and the biome represents more than 80 percent of the 4 million km² of land that is burned globally every year [13]. The carbon that is emitted from savanna fires is sequestered by postfire vegetation growth [14, 15], and recent evidence points to a limited long-term impact on global atmospheric carbon concentrations [16]. Several species of savanna plants are adapted to fires through evolutionary mechanisms such as thick barks [17] and require this disturbance in order to not be overtaken by other species. Fire allows mesic savannas to persist at the higher end of their rainfall range despite the climate being conducive to forests [18]. For example, at the border between tropical savannas and forests, dry C₄ grasses are replaced by moister leaf litter in the forest understory, thus forming a barrier to the spread of the fire [19]. The intensity and extent of fires can be amplified during periods of extreme drought that have been preceded by above-average rainfall [20] as well as by land management such as fire suppression.

Savannas have been undergoing changes that are altering the composition and structure of their vegetation such as increased rates of woody encroachment [21]. A combination of factors is responsible for these changes and includes CO₂ fertilization [22], land use [23], increasing aridity [24], more intense rainfall [25], and changes in livestock densities [26]. Although recent assessments using global ecosystem models have linked savannas to the interannual variability of the global carbon cycle [7, 8], they are still fairly overlooked [27] and eclipsed by forests [28] with which they are sometimes combined [19]. This primarily stems from ongoing ecological debates about the nature of savannas [29] but also from definitions used within remote sensing research that often overlap and are at times inconsistent with one another. Here, we review current use of satellite remote sensing in savannas, provide an overview of the challenges of terminology and large-scale mapping, summarize methods used to extract vegetation information from satellite data, and recommend several areas of emerging opportunities that will help reshape our understanding of savannas.

1.2. Overlapping Definitions of Savannas and the Challenges They Pose

Mapping the distribution of savannas is a longstanding challenge [30, 31] as there is little consensus on how to best to define them. For example, savannas that receive less than 700 mm of rainfall a year have been conflated with drylands, which include semiarid ecosystems. While there is some geographic overlap between the two (Figure 2), the term drylands denotes a climatic construct based on the availability, or lack thereof, of water. On the one hand, savannas comprise dryland ecosystems at the xeric end of their distribution, e.g., in the Sahel. On the other hand, some savanna ecosystems, such as the Cerrado and wet Miombo, receive between 1300 and 1700 mm of rainfall per year, which creates conditions that are favorable for a denser tree canopy than xeric savannas. Thus, at the mesic end of their distribution, savannas are sometimes included within the definition of forests. Characterizing some savannas, e.g., those in the tropics, as forests is problematic because their tree canopy cover, which is sparse relative to adjacent rainforests, leads to the perception that the forest had undergone some form of degradation [32] and requires restoration of the tree canopy density through tree planting programs.

Figure 2 

Overlap between global grasslands, savannas, forests, and drylands. The brown color shows the distribution of temperate, subtropical and tropical grasslands, savannas, and shrublands (referred to here collectively as “savannas”) in the Ecoregions 2017 dataset [33]. The distribution of forests (light blue color) is derived from the Global Ecological Zones (GEZ) dataset [34] and includes boreal, temperate, subtropical, and tropical forests. The dark blue color represents the apparent overlap between the Ecoregions 2017 and GEZ datasets, i.e., savannas and forests. The extent of drylands (dotted pattern) is based on the aridity index (AI) according to UNEP [35] where (i.e., including arid, semiarid, and dry subhumid ecosystems). The green dots represent examples of these ecosystems using imagery from Google Earth Pro: (a) Western shortgrass prairie, (b) Southeast US conifer savannas, (c) Eastern Anatolian steppe, (d) Daurian steppe, (e) Cape York Peninsula tropical savanna, (f) Central Zambezian wet Miombo, (g) Uruguayan savanna, and (h) Brazilian Cerrado.

Figure 3 exemplifies the diversity of both land-cover types and the nomenclatures used to map them in a part of Africa that is predominantly composed of savannas. Out of the five satellite-derived land-cover maps in Figure 3, only one (MOD12Q1) contains classes identified as savannas; others divide the region into a variety of different land-cover types, including forests. This inconsistency has led to a fundamental misunderstanding of savannas [27], which has, in part, caused the biome to be a major target of international initiatives [36–38] aimed at restoring “degraded landscapes” through tree planting for mitigating greenhouse gas emissions. Indeed, recent studies by Bastin et al. [39] and Bastin et al. [40] suggested that foresting the world’s sparsely vegetated areas, which include savannas, would help mitigate the effects of climate change by sequestering most of the carbon that is emitted from fossil fuel burning. These studies generated considerable debate [41–47] and criticism for ignoring, among other things, the fact that ecosystem functioning in savannas determines the spatial arrangement of trees and grasses allowing for their coexistence [48].

Figure 3 

Example of the diversity in the characterization of a savanna region according to continental-scale maps derived from satellite, climatic, and botanical stratification. The Great Lakes region provides an ideal setting to illustrate the differences in nomenclature. (a–d) Are based on PlanetScope satellite imagery at 3-meter spatial resolution and represent a gradient of woody canopy cover from dense (a) to sparse (d). (e) Represent five land cover classifications derived from satellite data: Global Land Cover SHARE (GLC-SHARE) [51], Moderate Resolution Imaging Spectroradiometer Land Cover Type (MCD12Q1) [52], European Space Agency Climate Change Initiative Land Cover (ESA CCI LC) [53], Copernicus Global Land Services (CGLS) [54], and GlobeLand30 [55]. Also shown is a map derived from botanical classification based on dominant species [56] and one derived from climatic stratification Koeppen-Geiger [57]. The overview figure in the lower left shows the delineation of subtropical and tropical grasslands, savannas, and shrublands in Africa (pink color) according to the Ecoregions 2017 dataset [33]. This delineation is also shown as a black line border in each of the seven maps in the main panel (e).

Most assessments that target the restoration of degraded forests employ end-user products derived from satellite data, such as tree- or land-cover maps. However, these products can contribute to the misconception that areas with sparse tree canopies are in need of reforestation [40]. Tree-cover maps are generally designed to focus on the percentage of a pixel that is covered by a closed canopy, and scattered trees in drier areas that do not form a canopy are largely excluded [49]. Similarly, the delineated classes in land-cover maps are entirely dependent on the land-cover classification system, and if that system does not include savannas as distinct ecosystems, it could support the narrative to prioritize them for reforestation [50]. This problem is illustrated in Figure 4, where the same savanna landscapes are classified differently depending on the land-cover product. There is little agreement in remote sensing literature on how to best define the tree-grass ecosystems we refer to here as savannas. This ambiguity leads to a patchwork of definitions as a result of different land-cover mapping initiatives adopting different nomenclatures. For example, the sparse savanna of the Serengeti is classified as grassland, herbaceous vegetation, shrubland, and tree-covered area (Figure 4). The denser savanna of the wet Miombo, by contrast, is classified as a forest, (closed) deciduous broad-leaf forest, mixed forest, and tree-covered area. Despite the differences in tree canopy cover and community composition, these savannas share the same basic ecological characteristics that define savannas such as tree-grass coexistence, prevalence of fire disturbance and herbivory, seasonal rainfall with similar mean annual amounts (600-1000 mm yr^-1, in this particular case), and an extended dry season lasting several months.

Figure 4 

Similarity and contrast in local-scale definitions of an African savanna along a canopy cover gradient. Close-up of a  m area within each map from Figure 3 shows inconsistencies in how tropical savannas along a woody canopy cover gradient are characterized in global products. The numbers within the parentheses under the title of each satellite-derived product indicate the native spatial resolution in which the product was provided. The (Planet Labs) PlanetScope 3-meter-resolution satellite imagery in (a)–(d) are in the same order as in Figure 3. Each map panel covers identical dimensions as the corresponding PlanetScope satellite image.

1.3. Large Area Monitoring of Savannas Using Satellite Remote Sensing

To a certain degree, the theoretical concepts that underpin savanna ecology across large scales remain unchanged as much of our current understanding of savannas comes from field observations [58]. The distinct advantage of satellite technology is that it can extend field observations at regular temporal intervals across large areas that are deficient in, or devoid of, field data. For example, AGB measurements from field inventories have a limited spatial and temporal footprint and can be upscaled using the mathematical relationship between field-collected variables and satellite observations [59] to estimate AGB at different scales [60]. Savannas have played a critical role in the 50-year track record of satellite remote sensing [61] by, for example, serving a test locations for satellite missions such as the Moderate Resolution Imaging Spectroradiometer (MODIS). Furthermore, satellite remote sensing has been widely used in studies of savanna ecology such as drivers of woody plant encroachment [62] and mapping long-term changes in vegetation composition [63]. This has led to satellite remote sensing becoming an indispensable tool for monitoring the composition, structure, and functioning of global savannas.

2. Current Developments in Satellite Remote Sensing of Savannas

A rapid advance in satellite sensor technology has occurred over the past decade, and several satellites have been launched with the capability of observing the Earth at unprecedented levels of detail. This section provides a synthesis of the current (2010–present) state of the art in satellite remote sensing of savannas. The following subsections discuss developments in the mapping and separation of woody and herbaceous components, estimating aboveground carbon stocks, detecting fires and quantifying fuel loads, and estimating land surface phenology and ecosystem-scale plant hydraulics. Some of the global-scale studies in this review use data and methods that are not necessarily specific to savannas, and their inclusion hinges on whether their results include information about savannas. Thus, the studies presented here help, in one way or another, improve our understanding of savannas and will contribute to a better characterization of the biome going forward.

2.1. Separation of Woody and Herbaceous Components

A longstanding technical challenge in the remote sensing of savannas has been the separation of woody and herbaceous components. Insufficient representation of the spatial structure of the woody component (i.e., trees and shrubs) [64, 65] indicates that the heterogeneity and variability of trees in savannas is not fully understood [66]. A likely reason is due to the fact that nonforest trees have previously been ignored as the pixels of commonly used satellite sensors such as MODIS and Landsat are coarser than the spatial resolution of the woody canopy [67]. It is conceivable that this is what led to the woody component receiving less attention relative to the more easily observed herbaceous component. However, the gap in knowledge about the true distribution of woody and herbaceous vegetation led to several large-scale Earth observation studies being conducted with primary focus on woody cover [23, 62, 68–70], and our current understanding is that trees outside forests are more ubiquitous than previously thought [71].

2.1.1. Spatial and Temporal Partitioning of Coarse Resolutions

As savannas are highly heterogeneous, the spatial resolutions of most optical sensors are too coarse to capture the diversity of the landscape, and pixels become a mixture of different land-cover types. The spatial partitioning of different signals within coarse-resolution pixels—termed spectral unmixing—requires information about the spectral properties of the contributing materials called endmembers, which in turn requires the collection of pure spectral signatures within a study area as reference samples [72]. Although there are diverse methods for collecting endmembers from coarse-scale satellite imagery, the availability of time series data helps distinguish between the temporal signals of vegetation embodied within each pixel [73]. However, moisture-driven variability in phenology of both the grass and wood layers can pose a challenge to this approach. Moisture conditions in savannas vary considerably within and between years, which alters the response of the woody leaves and the grass layer [74]. This ultimately influences the signal received by optical satellite sensors.

The dependence of vegetation on moisture availability is a characteristic of savannas, and the strong empirical relationship between them is well-known [75]. Soil moisture controls the phenology of the herbaceous component [76] whereas trees obtain moisture from a range of depths that are inaccessible to grasses [77]. The separation of woody and herbaceous components of savannas based on their sensitivity to moisture conditions is an active area of research. Early work was based on the relationship between vegetation greenness, as indicated by the normalized difference vegetation index (NDVI), and rainfall [78], and it was assumed that trees have a high long-term greenness and low sensitivity to rainfall, whereas grasses have a high sensitivity to rainfall and moderate greenness. This approach was later expanded by Anchang et al. [79] using the rain-use efficiency concept to infer the long-term rainfall sensitivity of woody and herbaceous vegetation. In another study, Kahiu and Hanan [80] leveraged rainfall sensitivity of savanna vegetation using an allometric relationship that included mean annual rainfall and maximum canopy leaf area index (LAI) of dominant tree species to partition MODIS LAI 250 m pixels over Africa into woody and herbaceous components.

Methods to disentangle woody and herbaceous signals, based on temporal trajectories, include time-series decomposition of NDVI data using the slow varying trend component as a baseline for the tree layer and the rapidly varying seasonal component as indicative of the grass layer [81, 82]. These approaches required an evergreen tree canopy as a baseline as well as noise-free data, conditions that are not applicable in many savannas. Furthermore, the signal of tree foliage in an area with a low canopy cover is relatively small and often overshadowed by other factors such as variations in background reflectance from bare ground or the grass layer [83]. Other studies relied on the phenological separability of woody and herbaceous components based on their phenological traits such as the senescence of the grass layer during the dry season, allowing for better detection of the signal from trees [84–86]. However, the uncertainty of these methods can be considerable due to, for example, the presence of deciduous trees that completely shed their canopy during the dry season.

2.1.2. Mapping at an Ecologically Relevant Scale

Supervised approaches to endmember extraction have relied on field collection [73] or very-high-resolution (VHR) imagery. Over the past decade, an increasing number of VHR imagery in Google Earth and cloud computing services such as Earth Engine have streamlined spatial endmember collection. The customization afforded by these services provides a range of possibilities—from manual selection of reference samples through visual interpretation [87] to near-automatic approaches that utilize VHR imagery to select endmembers in medium-resolution imagery [88] and further to a fully operational software suite such as the Food and Agriculture Organization’s Open Foris Collect Earth [89]. A potential caveat to these approaches is that the heterogeneity in leaf density and greenness caused by variability in rainfall and soil moisture can introduce uncertainties into the mapping process. VHR imagery provides more detailed opportunities for mapping savanna vegetation components. For example, data from the WorldView satellite constellation, which provides spatial resolutions of less than 0.5 m, have been successfully used to identify individual tree species [90] and distinguish trees, shrubs, and grasses [91]. However, these data are commercial and proprietary, and the associated costs and large data volumes have largely limited their use to local scales. An exception is a recent work by Brandt et al. [71], who created a wall-to-wall map of individual tree crowns over a sizable region from multitemporal mosaics of VHR satellite data provided under the NextView licensing agreement [92].

Eddy covariance flux towers collect data on micrometeorology, greenhouse gas, and energy exchange between the atmosphere and the terrestrial biosphere from a footprint of around 1 km² around each tower [93]. Although the distribution of these towers in savanna-dominated areas is low relative to other biomes [94], existing sites provide opportunities for coupling satellite and flux tower data [93] to further enable the partitioning and mapping of plant functional types into their constituent trees and grasses [74]. Some flux towers are equipped with phenological cameras (PhenoCams) that provide site-level repeat photography [95, 96]. The so-called “regions of interest” delineated within the field of view of PhenoCam images comprise areas of relatively pure grass and tree canopies that enable the separation of their respective phenology [97] and gross primary production [98]. These offer yet another opportunity for upscaling field measurements with satellite observations [99].

2.1.3. Overcoming Limitations Inherent to Optical Remote Sensing

Multispectral indices used to delineate vegetation greenness saturate in areas with relatively high green biomass [100] such as the savanna of the wet Miombo. This also applies to indices that were designed to be more sensitive to high green biomass such as the enhanced vegetation index [101]. Areas with high biomass often receive large amounts of rainfall and can thus have persistent cloud cover during the wet season, which obscures the optical signal. In contrast, synthetic aperture radar (SAR) microwaves can penetrate the woody canopy regardless of meteorological (cloud) conditions and interact with its structural components such as leaves, branches, and trunks. Direct comparison has shown that L-band ( m) SAR outperforms optical data in mapping woody cover [102]. Despite its complimentary advantages, the use of SAR trails that of optical data. The reasons for this are both historical and technical. Optical data for land applications developed a wider user base early on with the launch of Landsat-1 in 1972, and it has comparatively fewer processing steps before the data are ready for analysis. In contrast, SAR data are difficult to interpret [103] and require a relatively good understanding of several parameters, such as horizontal and vertical polarization, in order to be effectively exploited [104]. However, the present-day availability of analysis-ready SAR data [105] may help narrow this gap in usage. SAR remains a viable complement to multispectral optical data, particularly with the free availability of Sentinel-1 and second-generation Phased Array type L-band Synthetic Aperture Radar (PALSAR-2) data, as well as recent efforts to popularize the technology such as the online SAR Handbook [106].

2.2. Estimation of Aboveground Biomass

It has long been known that the radar backscatter signal at long wavelengths ( m) is sensitive to biomass and interacts with the woody components [107] of the vegetation canopy. Longwave active and passive microwave sensors provide adequate sensitivity to woody structure in savannas and do not saturate in these ecosystems where average AGB densities are less than 100 Mg ha^-1 [69, 108, 109] (Figure 5). Furthermore, microwave observations at longer wavelengths represent the entire vegetation column and include both woody and herbaceous components of savannas [110–112]. In passive microwave systems, the surface energy reaching the sensor is small compared to optical sensors, causing the former to have large spatial footprints (i.e., low spatial resolution). For example, Brandt et al. [66] used 25 km spatial resolution passive microwave L-band data from the Soil Moisture and Ocean Salinity mission to estimate short-term trends in AGB over African ecosystems. They found considerable a loss of AGB (−0.05 petagrams of carbon per year) in savannas that was associated with low rainfall. Comparatively, active microwave systems emit their own signal and thus can provide L-band data at a considerably higher spatial resolution. For example, data from the PALSAR family of sensors are provided as 25 m dual polarization (HH/HV) global mosaics. These have recently been combined with field observations to produce continental-scale baseline maps of aboveground savanna biomass [113].

Figure 5 

Variability in aboveground biomass (AGB) estimates from eight tropical savanna ecosystems. The savanna delineation in the central panel (pink color) depicts subtropical and tropical grasslands, savannas, and shrublands according to the Ecoregions 2017 dataset [33]. (a–h) Show examples of different ecosystems within this area and their corresponding ecosystem-wide AGB in megagrams per hectare (Mg ha^-1): (a) Yucatan Dry Forests (), (b) Sahelian Acacia Savanna (), (c) Central Deccan Plateau Dry Deciduous Forests (), (d) Central Indochina Dry Forests (), (e) Guianan Savanna (), (f) Cerrado (), (g) Eastern Miombo Woodlands (), and (h) Kimberly Tropical Savanna (). The AGB data in this figure is based on Avitabile et al. [118], who harmonized two earlier AGB maps by Baccini et al. [119] and Saatchi et al. [120] with field data at a 1 km spatial resolution.

Spaceborne light detection and ranging (lidar) is a relatively new technology (2003–present), and its applicability in savannas has not been thoroughly investigated. Most studies using lidar data in savannas employ airborne laser scanning as it provides high accuracy and detail [114], but at a considerably higher overall cost and smaller spatial coverage than spaceborne lidar [115]. In dense savannas, lidar has the potential to map horizontal and vertical attributes of woody vegetation such as such as canopy height [116], which can be included in models estimating savanna AGB [117]. The usability of lidar technology in savannas is intricately tied to the presence of woody cover as there is a tradeoff between pulse density and accuracy of the estimated attributes. This means that the herbaceous understory in savannas is not well represented in spaceborne lidar observations, which leads to that component being overlooked in AGB maps based on this technology.

One of the recently launched spaceborne lidars, the Advanced Topographic Laser Altimeter System (ATLAS) on board the Ice, Cloud, and Land Elevation Satellite-2, has three pairs of laser beams with a 90 m beam separation and a 3.3 km distance between pairs. This is an insufficient sampling density for mapping vegetation structure in savannas [121] and other open ecosystems [122] because the pulses are likely to miss trees scattered across the landscape. The other operational lidar, the Global Ecosystem Dynamics Investigation (GEDI) on board the International Space Station (ISS), has three lasers that altogether emit eight beams and measure vegetation structure within a 25 m footprint. Footprints are separated by 60 m along the ISS path, and beams are separated by 600 m [123]. Even though GEDI has a denser sampling rate and unlike ATLAS was specifically designed to measure the structure of temperate and tropical forests, it will directly measure only about 4% of Earth’s land surface [123] during the expected operation lifetime of the mission. That said, preliminary results in South Africa [124] seem promising but further research is required to fully assess its applicability in the diverse vegetation structure of savannas.

2.3. Detection and Mapping of Fire and Burned Area

Mapping of fire and burned area from space commenced soon after the launch of the first Landsat satellite in 1972 [125]. Thermal infrared imagery represents outgoing longwave radiation and reflects variability in the land surface energy balance due to loss of vegetation cover caused by fire [126, 127]. Satellite observations in the middle and shortwave infrared region are generally used to detect active fires and subsequent estimation of fire radiative power [128]. The estimation of fire radiative power is based on the notion that heat produced by a fixed quantity of biomass is invariant to the type of vegetation (e.g., woody or herbaceous) consumed by the fire [129], which causes savanna fires to have relatively high fire radiative power [130] due to their spatial and structural heterogeneity [131]. Fires in savannas are generally small [132], and evidence shows that they go largely undetected by MODIS [133] fire products, which have spatial thresholds of 10 ha for active fire detection and 169 ha for burned area mapping [134] (Figure 6). Since savanna fires typically occur during the dry season under fairly clear skies, the Sentinel-2 Multispectral Instrument (MSI) can conceivably detect active flaming fires as small as 1 m² [135] through a combination of the red band (centered at 665 nm) and the shortwave infrared band (centered at 2190 nm) due to the contrasting spectral response of these wavelengths to fire [136]. However, on days with sporadic cloud cover and during intense fires that produce thick smoke, the optical signal can be obscured. In such cases, data from Sentinel-1 SAR can be a complement for mapping active fires by penetrating through clouds and smoke [137] as wells providing observations at night.

Figure 6 

Number of fire ignitions relative to fire size in savannas. This figure was created by dividing mean monthly fire sizes and ignitions for 2016 from the MODIS-based World Fire Atlas [149]. The black outline indicates the boundary of temperate, tropical, and subtropical grasslands, savannas, and dry forests according to the Ecoregions 2017 dataset [33].

Burned area mapping relies on a high temporal resolution in order to capture recently burned areas before vegetation regeneration. Thus, the period between the end of a fire and image acquisition is critical to sense newly burned areas [138]. Progress in mapping burned area has opened opportunities for locally adaptive burned area algorithms using complementary datasets at spatial, temporal, and radiometric resolutions appropriate for small burned areas. The combination of Sentinel-2 MSI 10 m spatial resolution and thermal bands of the Landsat-8 Operational Land Imager (OLI) provides both a high temporal revisit frequency and higher spatial resolutions and radiometric sensitivity to capture smaller burned areas [139, 140]. There has also been progress in cross-sensor fusion, for example, between Sentinel-1 SAR and Sentinel-2 MSI [141], that overcomes limitations such as meteorological conditions and poor sensor performance. The remote sensing of fire activity under tree canopies is challenging using optical satellite sensors, since woody canopies can obscure the view of burned areas [142]. Here, lidar can prove useful in providing three-dimensional information on vegetation structure to differentiate between different height thresholds and the ground surface [143]. A caveat for using spaceborne lidar to map burned area is that measurements need to be made before and after the fire in order to infer its impact [138]. This may not be possible as postfire smoke haze or cloud cover might deflect or block the lidar signal; therefore, assessments of fire impact are done by comparing burned and unburned areas that are close to one another [143].

For more detailed monitoring of the strong diurnal variability of savanna fires [144], geostationary satellites provide near-real-time (e.g., 5–15 minutes) observations at spectral resolutions that are comparable to polar-orbiting satellites. Although the spatial resolution of geostationary satellites is coarser than most polar orbiting satellites, they can still detect fires down to 1 km² [145]. The use of geostationary satellites for fire detection is an active area of research, and currently, operational sensors provide global coverage—the Spinning Enhanced Visible and InfraRed Imager on board the Meteosat Second Generation satellites covers Africa [146]; the Advanced Himawari Imager on board Himawari-8 and Himawari-9 covers Australia and Asia [147]; and the Advanced Baseline Imager on board the Geostationary Operational Environmental Satellites covers the Americas [148].

The herbaceous component of xeric savannas is dominated by C₄ grasses, and the spatial coverage of these grasses increases with fire frequency at a continent scale [150]. The herbaceous component comprises the main fuel source for fires in savannas [11] as standing biomass (e.g., thick-barked trees) is not necessarily affected by fires. Therefore, the fuel load is the biomass amount that is available for burning [151]. Knowledge of the amount and extent of the fuel load is important from a land management perspective, but studies that map savanna fuel load at large scales have been relatively few compared to those that detect fires or map burned area [152]. Recently launched satellite sensors provide new opportunities for mapping fuel loads. For example, in a Brazilian study, spectrally unmixed Landsat-8 OLI and Sentinel-2 MSI data [153] explained 86% of the variation in fuel loads at a subpixel level using a Mixture Tuned Matched Filtering method. Spaceborne lidar is particularly promising for large area mapping fuel load because of its ability to estimate canopy height, which can be used to identify different fuel types. This potential of lidar is exemplified in another Brazilian study [154] where the authors used a GEDI-based modeling framework to predict fuel loads of multiple vegetation layers in a savanna at accuracies of 88% for woody fuels and 71% for the total fuel load.

2.4. Land Surface Phenology and Plant Hydraulics

The estimation of vegetation phenology using satellite data, termed land surface phenology [155], is widely used to study spatial and interannual patterns in growth, maturity, and senescence of vegetation. It is a useful indicator of climatic and edaphic conditions that influence vegetation growth and the overall response of ecosystems to global change [156]. Woody and herbaceous components of savannas have distinct phenologies whereby grasses are mostly annual, and trees can be either deciduous, evergreen, semievergreen, or rain-stimulated [157]. Leaf deployment strategies also differ among the deciduous trees and grasses with the former capable of deploying leaves considerably earlier than the latter [158]. Difference in leaf deployment can be used to estimate the timing of green-up and brown-down between trees and grasses that enables their separation using satellite time series data. For example, seasonal metrics extracted from MODIS and SPOT (Satellite pour l’Observation de la Terre) observations were used to map woody cover in the Sahel based on the fact that signal of evergreen and certain deciduous trees is more readily detectable in the early dry season with the absence of the underlying grass layer [85].

Photosynthetic activity in savannas during the dry season is primarily due to the woody component, because the herbaceous component generally undergoes full senescence. The woody component employs a continuum of strategies to deal with seasonal drought ranging from drought avoidance (isohydric) to drought tolerance (anisohydric) [20]. Isohydric species exhibit deciduous behavior by closing their stomata and shedding their leaves thereby shutting down photosynthesis and transpiration to avoid losing water [159]. Anisohydric species keep their stomata open and continue photosynthesis and transpiration during periods of drought. SAR and passive microwave observations can estimate water content within the leaves, branches, and trunks through an index known as vegetation optical depth (VOD) [160, 161]. The slope of regression between predawn and midday VOD has been shown to approximate an/isohydric conditions across large scales [162], and VOD values for savannas indicate moderate anisohydricity.

The prevalence of deciduous trees in savannas [163] complicates the idea that woody vegetation in savannas is anisohydric. Deciduousness in response to the dry season suggests isohydricity, and MODIS-based satellite observations recently revealed large-scale green-up in African savannas well before the onset of the rainy season, a phenomenon termed as “prerain green-up” [164, 165]. The advantage of an early leaf flush is the availability of a longer period in which trees can assimilate CO₂ and grow with little competition for resources from grasses and late-greening trees [166, 167]. Site-level observation and plant growth models show that trees must consume stored carbon and nitrogen for maintenance and growth respiration to fuel leaf flush and new shoots as well as have sufficient reserves of water through either stem and root storage or deep root systems [159, 168, 169].

The exchange of water between the land and atmosphere is a crucial factor controlling the overall storage of water in the surface, subsurface soil, and aquifers and within the savanna vegetation itself. This moisture component can be approximated using observations from the Gravity Recovery and Climate Experiment (GRACE) satellites through terrestrial water storage (TWS) [170]. TWS is an estimate of groundwater, soil moisture, and wet biomass inferred from small changes in Earth’s gravitational field and is tightly linked to the interannual variability of atmospheric CO₂ growth rate [171]. Using TWS, Madani et al. [172] found that the interannual variability of CO₂ uptake in the moist savannas of Africa is controlled by subsurface water. In another study, GRACE TWS and L-band VOD observations suggest that water stored within the stem and trunk plays a crucial role in sustaining early leaf flush in tropical savannas [173].

Geostationary satellites offer unique opportunities for studying plant hydraulics due to the high temporal resolution of their thermal and optical bands that can capture diurnal processes and be used for assessing evapotranspiration and soil moisture status [174]. Initial work on the use of geostationary data in savannas explored the diurnal canopy water status, daily evapotranspiration, surface soil moisture, and phenology [175–179]. Despite this progress, however, there have been surprisingly few recent studies on the use of geostationary data to study the diurnal progression of ecosystem water use and moisture stress in savannas.

3. Recommendations and Emerging Opportunities

3.1. A Consistent Definition of Savannas in Remote Sensing Research

There is need for a conceptual framework to aid the process of mapping savannas from space. Perhaps, the elephant in the room is the issue of how best to define savannas for application in remote sensing and at what scale that definition should hold [180]. A potential mitigating strategy to the myriad of definitions for savannas is to dissect the biome along key characteristics that are observable from space, e.g., rainfall seasonality, distinct wet and dry seasons, tree-dominated overstory with a grass-dominated understory, and prevalence of fire as a key disturbance mechanism. Using this as a baseline, savannas can conceivably be subdivided along a moisture (e.g., mesic, xeric) and canopy cover (e.g., dense, sparse) gradient along with any region-specific characteristics (e.g., Mopane, Eucalypt). Ideally, the term “forest” should be reserved to ecosystems with a closed or near-closed tree canopy and an understory devoid of grasses, and the term “woodland” for ecosystems with a more open tree-dominated overstory that similarly do not have a grass-dominated understory. But ecosystems are far from being representations of the abovementioned exemplary cases, and there will invariably be cases where these definitions are inapplicable. There needs to be a conceptual space [181] where the disparate characteristics that make up savannas and their relative advantages and weaknesses at different scales are discussed. In light of this, the delineation of savannas as distinct biomes that are not degraded forests [19, 182] but an alternative stable state to forests [183] is crucial to keep them ecologically intact.

3.2. Improving Land-Cover Classification Workflows

Uncertainties in tree [184] and land-cover maps [185] are caused by misclassification, inadequate spatial resolution and minimum mapping unit, thematic inconsistencies, or a combination of all. Incorporating geospatial datasets on key savanna characteristics in land-cover classification workflows can help produce more accurate maps of savanna distribution. Recent evidence points to the fact that ecoregion-based mapping of the terrestrial biosphere is an effective way to delineate biodiversity patterns [186]. We also know that the inclusion of environmental variables increases the accuracy of land-cover maps in complex savanna ecosystems [187, 188]. As is evident from Section 2 of this review, there has been substantial remote sensing research on the structure, composition, and ecophysiology of savannas. Traditional land-cover classification of savannas should be enhanced to include ecologically relevant information such as land surface phenology [189], soil properties [190], rainfall seasonality [191], fire activity [192], and even plant canopy traits and functional diversity [193, 194].

Another approach towards consolidating land-cover classification with ecosystem functioning is the establishment of long-term plots within diverse savannas globally [195] such as the ones that have been instituted in forest science (e.g., http://forestplots.net [196]). A widely distributed global network will be able to capture climatic and edaphic gradients as well as community biogeography and vegetation structure. The integration of such information with satellite observations via machine learning or mechanistic models will invariably lead to better models that are capable to classify differences between savannas and distinguish them from other similar ecosystems at large scales.

3.3. Incorporating Higher Resolutions and Multisource Observations

Nanosatellites (satellites weighing between 1 and 10 kg) provide opportunities to overcome the traditional tradeoff between high temporal and high spatial resolutions. Operational Earth observation nanosatellites have spatial resolutions ranging from 0.5 m to 5 m and daily revisit times across the globe [197]. According to the Nanosats Database (nanosats.eu), out of the 1674 nanosatellites in orbit (as of June 2021), 620 are dedicated to Earth observation. The company Planet Labs (San Francisco, CA, USA, hereafter Planet) operates 436 of the Earth observation nanosatellites. Savanna ecosystem studies using commercial high-resolution satellite data have typically covered areas less than 250 km² using one or two scenes [90, 198, 199] due to the prohibitive cost of procuring the satellite data. In 2020, Norway’s International Climate and Forest Initiative recently partnered with Planet [200] (http://www.planet.com/nicfi/) to provide free access to analysis-ready monthly mosaics of multispectral satellite imagery covering savannas across most of the tropics. This dataset will invariably provide insights into the role of landscape structure in the ecosystem functioning of savannas as well as unique opportunities for consistent monitoring of a substantial portion of the global savanna biome.

Data from geostationary satellites have been generally underused in savanna vegetation studies compared to polar-orbiting satellites due to their relatively lower spatial resolution. The large number of spectral bands provided by these satellites are comparable to those of MODIS [201] (which will soon be retired) and thus offer unique opportunities for the retrieval of ecosystem-scale metrics. Furthermore, the very high temporal resolution of geostationary satellites can capture diurnal changes in photosynthetic capacity in response to short-term climatic events, which are likely unobserved by polar-orbiting satellites. There is, however, indication of a revived interest in geostationary satellites, for example, for studying vegetation seasonality [202] and vegetation-water interactions at the continental scale [203].

Hyperspectral satellite sensors have the potential to provide synoptic and holistic insight into plant functional processes due to their very narrow spectral bands [204, 205] but have been relatively underused due to limited availability compared to multispectral sensors. Hyperspectral sensors provide a wide range of spectral bands providing a full reflectance spectrum between 400 and 2500 nm spaced at intervals ranging from 1 to 10 nm. There are a handful of such sensors currently in orbit (e.g., Environmental Mapping and Analysis Program (EnMAP), Project for On-Board Autonomy-1 (PROBA-1), Hyperspectral Imager Suite (HISUI), and Precursore IperSpettrale della Missione Applicativa (PRISMA)), and more are planned in the near future (e.g., Spaceborne Hyperspectral Applicative Land and Ocean Mission (SHALOM)). The fine-scale sampling of spectral bands and medium spatial resolution makes hyperspectral data useful for providing diverse information about the vegetation such as canopy nitrogen and phosphorous concentrations, foliar protein, phenology, flowering groups, biomass content, and species composition that can be used to map functional traits at large scales [204–209]. Importantly, these derived traits can be coupled with global ecosystem models or operational satellite-based algorithms to better represent savanna ecosystem functioning.

Two SAR missions with complementary radiometric resolutions relevant for savannas are scheduled for launch in 2023. The European Space Agency’s BIOMASS mission [210] will operate a P-band ( MHz,  m) SAR capable of penetrating the canopy of denser savannas, such as parts of the Miombo, and provide estimates of biomass at a spatial resolution of 200 m. The NISAR mission is a collaboration between NASA and the Indian Space Research Organization and will operate an L-band ( GHz,  m) and an S-band ( GHz,  m) SAR focused on areas with AGB of <100 Mg ha^-1, which encompasses most global savannas. These missions will provide more enhanced opportunities to synergize data not only across SAR platforms but also including lidar and optical sensors, for improved mapping of savanna vegetation structure [211]. For example, because longwave SAR can penetrate through to the understory, it can be fused with observations that capture woody biomass such as GEDI (or Sentinel-1 SAR), which will open up an opportunity to spatially separate herbaceous biomass from woody biomass.

3.4. Exploiting Machine Learning and Big Data Analytics

The efficiency of machine learning (ML) to extract patterns from satellite data made it an attractive tool for the Earth observation community. Breakthroughs in mapping woody vegetation in savannas using ML include detecting individual trees [71], identifying them down to the species level [91], and assessing reforestation effects in savannas [212]. The stage is currently set for the study of ecosystem-scale interactions using ML, which has become increasingly popular in ecology [213], in part due to its relatively few a priori assumptions. Interpreting ecological ML models to understand underlying relationships is a challenge [214] as there is generally no direct assessment of how the predictions are generated. ML models often have higher predictive accuracies than traditional statistical learning methods, such as (non-)linear regression. The implication here is that ML models embody functional relationships between the parameters, provided they are not spurious [215]. This is exemplified in Schwieder et al. [216], where the authors used a random forest regression algorithm to link savanna land surface phenology to aboveground carbon and identify the key phenological metrics that define this relationship at each of their study sites.

A noteworthy advantage of ML is scalability and repeatability, which could help break the entry barrier for developing countries to use satellite remote sensing for monitoring savanna resources. Such a framework was proposed by Anchang et al. [49], who combined Sentinel-1 and Sentinel-2 observations using a random forest regression to map percent woody canopy cover in savannas. More recent studies have exploited capabilities of convolutional neural networks to detect individual savanna trees [71] and elephants [217] from VHR satellite observations. These developments represent a research frontier when it comes to satellite remote sensing of savannas and open the door for hyperlocal analysis of the role of large herbivory in these ecosystems. These models, however, are rather opaque, but there are ongoing efforts to make their predictions more transparent, including the development of interpretable machine learning frameworks [218] as an alternative to black-box models.

A challenge for the use of ML with satellite observations to infer ecological relationships is model transferability [219]. Ecological models should ideally be generalizable in order to be applied outside the spatiotemporal space of the data used to train them. In savannas, the large spatial and temporal heterogeneity within ecosystems can limit model generalizability [220]. A potential solution here is model-data assimilation whereby satellite observations and process-based models, such as dynamic global vegetation models, are integrated to improve parametrization and lead to better representation of ecosystem functioning [221]. This is particularly pertinent for savannas where process-based models have generally underperformed [65]. Similarly, outputs from process-based models can be used to constrain ML predictions, creating a bidirectional benefit that could help reduce uncertainties in both ML and process-based model outputs.

The increasing availability of large volumes of data (termed “big data”), many of which are ecologically relevant for savannas, has opened up new avenues for research [222]. For example, plant trait data provide a strong response to ecosystem functioning and various plant traits show stable relationships with both optical reflectance [223] and SAR backscatter [224]. These could be parameterized using ML to develop trait-backscatter-spectral models capable of differentiating plant functional types and allowing for their global mapping [225]. The global plant trait database (ver. 5) [226] has nearly 12 million trait records covering 280,000 plant taxa, which can be coupled with both raw satellite observations and derived products (vegetation properties, climate, soil, elevation, disturbance, etc.) within an ML framework to infer undiscovered relationships between biotic and abiotic components of savannas, in essence, allowing for the discovery of unknown unknowns that can then be either proven or disproven with targeted field experiments.

In 2020, the Copernicus Open Access Hub (http://scihub.copernicus.eu) published 6.3 petabytes of Sentinel-1 SAR and Sentinel-2 MSI data. These large volumes of satellite data pose new challenges to data curation, storage, and computing capacity. Hosting of data and analytical tools on public cloud infrastructures such as the Copernicus Data and Information Access Services (copernicus.eu/en/access-data/dias) as well as private cloud computing services such as Amazon Web Services, Microsoft Azure, and Google Earth Engine provides timely opportunities for continental-scale analysis [227] of savannas. For example, recent work has shown the potential of cloud-based analysis of Sentinel-1 SAR and Sentinel-2 MSI time series to map woody canopies in savannas [68]. A key advantage of cloud-based services is the potential for efficient operational workflows since both the data and algorithms are stored in the same infrastructure. This not only allows for near-real time monitoring of savannas, but it is also scalable and can be replicated as per user needs [49]. Cloud-based services such as Google’s Earth Engine (http://earthengine.google.com) platform are also equitable because they level the playing field for remote sensing scientists in developing countries by eliminating the need for costly computational infrastructure to run large-scale analyses.

4. Conclusions

Our understanding of large-scale savanna ecology and its response to global change has dramatically improved in recent decades, and satellite remote sensing has played a key role in this process. The marked advantage of satellite technology is that it provides continuous observations across large areas. However, it is clear that traditional methods of land-cover mapping, whereby training samples for specific classes are collected from field surveys or photointerpretation, are insufficient for delineating savannas. More holistic methods that include the unique climatic, edaphic, and disturbance characteristics of savannas are necessary for a more complete description of the role of savannas in the terrestrial biosphere. These include, for example, incorporating big geospatial data on vegetation, rainfall, fire, soil, and topography as covariates in classification models along with satellite data.

The past twelve years (2010–2021) have witnessed remarkable progress in new satellites and analytical methods that have helped improve our understanding of savannas. Better representation of woody and herbaceous components remains a crucial aspect that underlines many of the structural and functional attributes of savannas. Earth observation studies of savannas have generally used coarse- to medium-resolution imagery with more attention paid to one or the other component. This will likely change going forward as higher resolution imagery becomes more readily available enabling distinct spatial separation between trees and grasses.

Although new satellite sensors will invariably add to the ecological synthesis of savannas, currently availably geostationary, hyperspectral, synthetic aperture radar (SAR), and light detection and ranging (lidar) technologies remain relatively underused. This potentially hampers synergies of new and old data, particularly where one technology can complement or extend the other in one or more dimensions. To better understand savanna ecosystem responses to climate change and disturbances, both historical and current data need to be utilized. Despite their limitations, we are bound to using historical multisensor satellite observations for insight into change dynamics, whereas current and new sensor systems provide high spatial, temporal, and radiometric resolutions for more detailed analysis of ecosystem response. Indeed, the complexity of the savanna biome necessitates spatial (VHR), spectral (hyperspectral), temporal (geostationary), and structural (lidar, SAR) synergies.

Advancements in the observation, mapping, and ecological synthesis of savannas from space will gain momentum in the future, which is a promising outlook for the improved understanding of the large-scale processes that occur in the biome. As remote sensing scientists and ecologists continue to develop ecological frameworks to understand savanna vegetation functioning at large scales, improved communication between disciplines and flexible ontologies are needed to link satellite observations and ecological data. Such cross-disciplinary communication will result in enhanced maps of savanna distribution as well as better predictions of savanna response to climate change and anthropogenic disturbance. This will ultimately lead to more informed conservation efforts of these special ecosystems.

Data Availability

Data on satellite observations are as follows: Sentinel-1 and Sentinel-2 (https://scihub.copernicus.eu/dhus/); Landsat-8 (https://earthexplorer.usgs.gov/); MODIS (https://lpdaac.usgs.gov/product_search/); PALSAR-2 (https://www.eorc.jaxa.jp/ALOS/en/index_e.htm); GEDI (https://gedi.umd.edu/data/download/ and https://lpdaac.usgs.gov/product_search/); ATLAS/ICESat-2 (https://icesat-2.gsfc.nasa.gov/science/data-products); Meteosat Second Generation (https://www.eumetsat.int/0-degree-service); Himawari-8 (https://registry.opendata.aws/noaa-himawari/); Geostationary Operational Environmental Satellites (https://registry.opendata.aws/noaa-goes/); NICFI Analysis Ready Mosaics (https://www.planet.com/nicfi/#sign-up). Satellite data products are as follows: ESA CCI Land Cover (https://www.esa-landcover-cci.org/); CGLS Land Cover (https://land.copernicus.eu/global/products/lc); GlobeLand30 (http://www.globallandcover.com/defaults_en.html); Global Land Cover SHARE (https://data.apps.fao.org/map/catalog/srv/eng/catalog.search#/metadata/ba4526fd-cdbf-4028-a1bd-5a559c4bff38); World Fire Atlas (http://www.globalfiredata.org/).

Conflicts of Interest

The authors declare no conflict of interest.

Authors’ Contributions

A.M.A. conceived the study, led the writing, and designed the graphics (except Figure 1, which was commissioned). All authors (A.M.A., M.B., C.A., and R.F.) contributed to substantial discussion of content and extensive edits to the manuscript, its revisions, and the final production.

Acknowledgments

We thank Dr. Nicola Stevens, Dr. Nathalie Pettorelli, and Dr. Torbern Tagesson for their helpful comments during the drafting of the manuscript. This work was supported by the Swedish Research Council–Vetenskapsrådet (grant 2018-00430 to A.M.A); Independent Research Fund Denmark–DFF Sapere Aude (grant 9064-00049B to M.B.); Villum Foundation through the project “Deep Learning and Remote Sensing for Unlocking Global Ecosystem Resource Dynamics” (DeReEco to R.F.).

References

1. S. Archibald, W. J. Bond, W. Hoffmann, C. Lehmann, C. Staver, and N. Stevens, “Distribution and determinants of savannas,” in Savanna Woody Plants and Large Herbivores, P. F. Scogings and M. Sankaran, Eds., pp. 1–24, John Wiley & Sons Ltd., Hoboken, NJ, USA, 2020. View at: Google Scholar
2. I. Campbell, K. Behrens, C. Hesse, and P. Chaon, Habitats of the World: A Field Guide for Birders, Naturalists, and Ecologists, Princeton University Press, Princeton, NJ, USA, 2021.
3. J. Mistry, World Savannas: Ecology and Human Use, Taylor & Francis, New York, NY, USA, 2000.
4. R. T. Pennington, C. E. R. Lehmann, and L. M. Rowland, “Tropical savannas and dry forests,” Current Biology, vol. 28, no. 9, pp. R541–R545, 2018. View at: Publisher Site | Google Scholar
5. I. Harris, T. J. Osborn, P. Jones, and D. Lister, “Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset,” Scientific Data, vol. 7, no. 1, p. 109, 2020. View at: Publisher Site | Google Scholar
6. M. A. Huston and S. Wolverton, “The global distribution of net primary production: resolving the paradox,” Ecological Monographs, vol. 79, no. 3, pp. 343–377, 2009. View at: Publisher Site | Google Scholar
7. A. Ahlström, M. R. Raupach, G. Schurgers et al., “The dominant role of semi-arid ecosystems in the trend and variability of the land CO2sink,” Science, vol. 348, no. 6237, pp. 895–899, 2015. View at: Publisher Site | Google Scholar
8. B. Poulter, D. Frank, P. Ciais et al., “Contribution of semi-arid ecosystems to interannual variability of the global carbon cycle,” Nature, vol. 509, no. 7502, pp. 600–603, 2014. View at: Publisher Site | Google Scholar
9. S. Bojinski, M. Verstraete, T. C. Peterson, C. Richter, A. Simmons, and M. Zemp, “The concept of essential climate variables in support of climate research, applications, and policy,” Bulletin of the American Meteorological Society, vol. 95, no. 9, pp. 1431–1443, 2014. View at: Publisher Site | Google Scholar
10. D. Eamus, A. Huete, and Q. Yu, Vegetation Dynamics: A Synthesis of Plant Ecophysiology, Remote Sensing and Modelling, Cambridge University Press, 2016.
11. M. N. Kahiu and N. P. Hanan, “Fire in sub-Saharan Africa: the fuel, cure and connectivity hypothesis,” Global Ecology and Biogeography, vol. 27, no. 8, pp. 946–957, 2018. View at: Publisher Site | Google Scholar
12. G. R. McPherson, Ecology and Management of North American Savannas, University of Arizona Press, Tuscon, AZ, USA, 1997.
13. J. Ratnam, C. Sheth, and M. Sankaran, “African and Asian savannas,” in Savanna Woody Plants and Large Herbivores, P. F. Scogings and M. Sankaran, Eds., pp. 25–49, John Wiley & Sons Ltd., Hoboken, NJ, USA, 2020. View at: Google Scholar
14. J. S. Landry and H. D. Matthews, “Non-deforestation fire vs. fossil fuel combustion: the source of CO₂ emissions affects the global carbon cycle and climate responses,” Biogeosciences, vol. 13, no. 7, pp. 2137–2149, 2016. View at: Publisher Site | Google Scholar
15. G. J. Lipsett-Moore, N. H. Wolff, and E. T. Game, “Emissions mitigation opportunities for savanna countries from early dry season fire management,” Nature Communications, vol. 9, no. 1, p. 2247, 2018. View at: Publisher Site | Google Scholar
16. Y. Zhou, J. Singh, J. R. Butnor et al., “Limited increases in savanna carbon stocks over decades of fire suppression,” Nature, vol. 603, no. 7901, pp. 445–449, 2022. View at: Publisher Site | Google Scholar
17. M. F. Simon and T. Pennington, “Evidence for adaptation to fire regimes in the tropical savannas of the Brazilian Cerrado,” International Journal of Plant Sciences, vol. 173, no. 6, pp. 711–723, 2012. View at: Publisher Site | Google Scholar
18. A. C. Staver, S. Archibald, and S. A. Levin, “The global extent and determinants of savanna and forest as alternative biome states,” Science, vol. 334, no. 6053, pp. 230–232, 2011. View at: Publisher Site | Google Scholar
19. J. Ratnam, W. J. Bond, R. J. Fensham et al., “When is a ‘forest’ a savanna, and why does it matter?” Global Ecology and Biogeography, vol. 20, no. 5, pp. 653–660, 2011. View at: Publisher Site | Google Scholar
20. M. Sankaran, “Droughts and the ecological future of tropical savanna vegetation,” Journal of Ecology, vol. 107, no. 4, pp. 1531–1549, 2019. View at: Publisher Site | Google Scholar
21. N. Stevens, C. E. R. Lehmann, B. P. Murphy, and G. Durigan, “Savanna woody encroachment is widespread across three continents,” Global Change Biology, vol. 23, no. 1, pp. 235–244, 2017. View at: Publisher Site | Google Scholar
22. A. P. Devine, R. A. McDonald, T. Quaife, and I. M. D. Maclean, “Determinants of woody encroachment and cover in African savannas,” Oecologia, vol. 183, no. 4, pp. 939–951, 2017. View at: Publisher Site | Google Scholar
23. T. M. Rosan, L. E. O. C. Aragão, I. Oliveras et al., “Extensive 21st-century woody encroachment in South America's savanna,” Geophysical Research Letters, vol. 46, no. 12, pp. 6594–6603, 2019. View at: Publisher Site | Google Scholar
24. M. Berdugo, M. Delgado-Baquerizo, S. Soliveres et al., “Global ecosystem thresholds driven by aridity,” Science, vol. 367, no. 6479, pp. 787–790, 2020. View at: Publisher Site | Google Scholar
25. A. Kulmatiski and K. H. Beard, “Woody plant encroachment facilitated by increased precipitation intensity,” Nature Climate Change, vol. 3, no. 9, pp. 833–837, 2013. View at: Publisher Site | Google Scholar
26. M. J. Hill and N. P. Hanan, Ecosystem Function in Savannas: Measurement and Modeling at Landscape to Global Scales, CRC Press, 2010. View at: Publisher Site
27. C. L. Parr, C. E. R. Lehmann, W. J. Bond, W. A. Hoffmann, and A. N. Andersen, “Tropical grassy biomes: misunderstood, neglected, and under threat,” Trends in Ecology & Evolution, vol. 29, no. 4, pp. 205–213, 2014. View at: Publisher Site | Google Scholar
28. A. Dobson, G. Hopcraft, S. Mduma et al., “Savannas are vital but overlooked carbon sinks,” Science, vol. 375, no. 6579, pp. 392–392, 2022. View at: Publisher Site | Google Scholar
29. J. G. Pausas and W. J. Bond, “Alternative biome states in terrestrial ecosystems,” Trends in Plant Science, vol. 25, no. 3, pp. 250–263, 2020. View at: Publisher Site | Google Scholar
30. C. E. R. Lehmann, S. A. Archibald, W. A. Hoffmann, and W. J. Bond, “Deciphering the distribution of the savanna biome,” New Phytologist, vol. 191, no. 1, pp. 197–209, 2011. View at: Publisher Site | Google Scholar
31. B. P. Murphy and D. M. J. S. Bowman, “What controls the distribution of tropical forest and savanna?” Ecology Letters, vol. 15, no. 7, pp. 748–758, 2012. View at: Publisher Site | Google Scholar
32. J. W. Veldman, “Clarifying the confusion: old-growth savannahs and tropical ecosystem degradation,” Philosophical Transactions of the Royal Society B: Biological Sciences, vol. 371, no. 1703, article 20150306, 2016. View at: Publisher Site | Google Scholar
33. E. Dinerstein, D. Olson, A. Joshi et al., “An ecoregion-based approach to protecting half the terrestrial realm,” Bioscience, vol. 67, no. 6, pp. 534–545, 2017. View at: Publisher Site | Google Scholar
34. FAO, Global Ecological Zones for FAO Forest Reporting: 2010 Update, FAO, Rome, Italy, 2012.
35. UNEP, World Atlas of Desertification, United Nations Environmental Programme, New York, NY, USA, 1992.
36. W. J. Bond, N. Stevens, G. F. Midgley, and C. E. R. Lehmann, “The trouble with trees: afforestation plans for Africa,” Trends in Ecology & Evolution, vol. 34, no. 11, pp. 963–965, 2019. View at: Publisher Site | Google Scholar
37. World Resources Institute, “African Forest Landscape Restoration Initiative (AFR100),” 2019; https://www.wri.org/our-work/project/AFR100/about-afr100. View at: Google Scholar
38. G. W. Fernandes, M. S. Coelho, R. B. Machado et al., “Afforestation of savannas: an impending ecological disaster,” Natureza & Conservação, vol. 14, no. 2, pp. 146–151, 2016. View at: Publisher Site | Google Scholar
39. J.-F. Bastin, N. Berrahmouni, A. Grainger et al., “The extent of forest in dryland biomes,” Science, vol. 356, no. 6338, pp. 635–638, 2017. View at: Publisher Site | Google Scholar
40. J.-F. Bastin, Y. Finegold, C. Garcia et al., “The global tree restoration potential,” Science, vol. 365, no. 6448, pp. 76–79, 2019. View at: Publisher Site | Google Scholar
41. P. Friedlingstein, M. Allen, J. G. Canadell, G. P. Peters, and S. I. Seneviratne, “Comment on “The global tree restoration potential”,” Science, vol. 366, no. 6463, article eaay8060, 2019. View at: Publisher Site | Google Scholar
42. J. W. Veldman, J. C. Aleman, S. T. Alvarado et al., “Comment on “The global tree restoration potential”,” Science, vol. 366, no. 6463, article eaay7976, 2019. View at: Publisher Site | Google Scholar
43. S. L. Lewis, E. T. A. Mitchard, C. Prentice, M. Maslin, and B. Poulter, “Comment on “The global tree restoration potential”,” Science, vol. 366, no. 6463, article eaaz0388, 2019. View at: Publisher Site | Google Scholar
44. A. Grainger, L. R. Iverson, G. H. Marland, and A. Prasad, “Comment on “The global tree restoration potential”,” Science, vol. 366, no. 6463, article eaay8334, 2019. View at: Publisher Site | Google Scholar
45. E. Luedeling, J. Börner, W. Amelung, K. Schiffers, K. Shepherd, and T. Rosenstock, “Forest restoration: overlooked constraints,” Science, vol. 366, no. 6463, pp. 315–315, 2019. View at: Publisher Site | Google Scholar
46. D. M. Griffith, C. E. R. Lehmann, C. A. E. Strömberg et al., “Comment on “The extent of forest in dryland biomes”,” Science, vol. 358, no. 6365, article eaao1309, 2017. View at: Publisher Site | Google Scholar
47. M. de la Cruz, P. F. Quintana-Ascencio, L. Cayuela, C. I. Espinosa, and A. Escudero, “Comment on “The extent of forest in dryland biomes”,” Science, vol. 358, no. 6364, article eaao0369, 2017. View at: Publisher Site | Google Scholar
48. L. Eigentler and J. A. Sherratt, “Spatial self-organisation enables species coexistence in a model for savanna ecosystems,” Journal of Theoretical Biology, vol. 487, article 110122, 2020. View at: Google Scholar
49. J. Y. Anchang, L. Prihodko, W. Ji et al., “Toward operational mapping of woody canopy cover in tropical savannas using Google Earth Engine,” Frontiers in Environmental Science, vol. 8, no. 4, 2020. View at: Publisher Site | Google Scholar
50. M. E. Fagan, “A lesson unlearned? Underestimating tree cover in drylands biases global restoration maps,” Global Change Biology, vol. 26, no. 9, pp. 4679–4690, 2020. View at: Publisher Site | Google Scholar
51. J. Latham, R. Cumani, I. Rosati, and M. Bloise, Global Land Cover SHARE (GLC-SHARE) database beta-release version 1.0-2014, FAO, 2014.
52. D. Sulla-Menashe, J. M. Gray, S. P. Abercrombie, and M. A. Friedl, “Hierarchical mapping of annual global land cover 2001 to present: the MODIS Collection 6 Land Cover product,” Remote Sensing of Environment, vol. 222, pp. 183–194, 2019. View at: Publisher Site | Google Scholar
53. ESA, Land Cover Climate Change Initiative Product Version 2, European Space Agency, 2017.
54. M. Buchhorn, M. Lesiv, N.-E. Tsendbazar, M. Herold, L. Bertels, and B. Smets, “Copernicus global land cover layers—collection 2,” Remote Sensing, vol. 12, no. 6, p. 1044, 2020. View at: Publisher Site | Google Scholar
55. J. Chen, J. Chen, A. Liao et al., “Global land cover mapping at 30 m resolution: A POK-based operational approach,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 103, pp. 7–27, 2015. View at: Publisher Site | Google Scholar
56. F. White, The Vegetation of Africa: A Descriptive Memoir to Accompany the UNESCO/AETFAT/UNSO Vegetation Map of Africa 7 Place de Fontenoy, UNESCO, Paris, France, 1983.
57. H. E. Beck, N. E. Zimmermann, T. R. McVicar, N. Vergopolan, A. Berg, and E. F. Wood, “Present and future Koppen-Geiger climate classification maps at 1-km resolution,” Scientific Data, vol. 5, no. 1, article 180214, 2018. View at: Publisher Site | Google Scholar
58. M. Greve, “Tree-grass ratios in savannas - challenging paradigms,” Frontiers of Biogeography, vol. 5, no. 3, 2013. View at: Publisher Site | Google Scholar
59. M. Réjou-Méchain, N. Barbier, P. Couteron et al., “Upscaling forest biomass from field to satellite measurements: sources of errors and ways to reduce them,” Surveys in Geophysics, vol. 40, no. 4, pp. 881–911, 2019. View at: Publisher Site | Google Scholar
60. P. Mondal, S. S. McDermid, and A. Qadir, “A reporting framework for Sustainable Development Goal 15: multi-scale monitoring of forest degradation using MODIS, Landsat and Sentinel data,” Remote Sensing of Environment, vol. 237, article 111592, 2020. View at: Publisher Site | Google Scholar
61. M. J. Hill, “Remote sensing of savannas and woodlands: editorial,” Remote Sensing, vol. 13, no. 8, p. 1490, 2021. View at: Publisher Site | Google Scholar
62. Z. S. Venter, M. D. Cramer, and H. J. Hawkins, “Drivers of woody plant encroachment over Africa,” Nature Communications, vol. 9, no. 1, p. 2272, 2018. View at: Publisher Site | Google Scholar
63. A. Alencar, J. Z. Shimbo, F. Lenti et al., “Mapping three decades of changes in the Brazilian savanna native vegetation using Landsat data processed in the Google Earth Engine platform,” Remote Sensing, vol. 12, no. 6, p. 924, 2020. View at: Publisher Site | Google Scholar
64. M. Baudena, S. C. Dekker, P. M. van Bodegom et al., “Forests, savannas, and grasslands: bridging the knowledge gap between ecology and dynamic global vegetation models,” Biogeosciences, vol. 12, no. 6, pp. 1833–1848, 2015. View at: Publisher Site | Google Scholar
65. R. Whitley, J. Beringer, L. B. Hutley et al., “Challenges and opportunities in land surface modelling of savanna ecosystems,” Biogeosciences, vol. 14, no. 20, pp. 4711–4732, 2017. View at: Publisher Site | Google Scholar
66. M. Brandt, J.-P. Wigneron, J. Chave et al., “Satellite passive microwaves reveal recent climate-induced carbon losses in African drylands,” Nature Ecology & Evolution, vol. 2, no. 5, pp. 827–835, 2018. View at: Publisher Site | Google Scholar
67. H. S. Nghiyalwa, M. Urban, J. Baade et al., “Spatio-temporal mixed pixel analysis of savanna ecosystems: a review,” Remote Sensing, vol. 13, no. 19, p. 3870, 2021. View at: Publisher Site | Google Scholar
68. W. Zhang, M. Brandt, Q. Wang et al., “From woody cover to woody canopies: how Sentinel-1 and Sentinel-2 data advance the mapping of woody plants in savannas,” Remote Sensing of Environment, vol. 234, article 111465, 2019. View at: Publisher Site | Google Scholar
69. F. Tian, M. Brandt, Y. Y. Liu, K. Rasmussen, and R. Fensholt, “Mapping gains and losses in woody vegetation across global tropical drylands,” Global Change Biology, vol. 23, no. 4, pp. 1748–1760, 2017. View at: Publisher Site | Google Scholar
70. X. Yang, K. A. Crews, and B. Yan, “Analysis of the pattern of potential woody cover in Texas savanna,” International Journal of Applied Earth Observation and Geoinformation, vol. 52, pp. 527–531, 2016. View at: Publisher Site | Google Scholar
71. M. Brandt, C. J. Tucker, A. Kariryaa et al., “An unexpectedly large count of trees in the West African Sahara and Sahel,” Nature, vol. 587, no. 7832, pp. 78–82, 2020. View at: Publisher Site | Google Scholar
72. C. Quintano, A. Fernández-Manso, Y. E. Shimabukuro, and G. Pereira, “Spectral unmixing,” International Journal of Remote Sensing, vol. 33, no. 17, pp. 5307–5340, 2012. View at: Publisher Site | Google Scholar
73. T. Meyer and G. S. Okin, “Evaluation of spectral unmixing techniques using MODIS in a structurally complex savanna environment for retrieval of green vegetation, nonphotosynthetic vegetation, and soil fractional cover,” Remote Sensing of Environment, vol. 161, pp. 122–130, 2015. View at: Publisher Site | Google Scholar
74. X. Ma, A. Huete, C. E. Moore et al., “Spatiotemporal partitioning of savanna plant functional type productivity along NATT,” Remote Sensing of Environment, vol. 246, article 111855, 2020. View at: Publisher Site | Google Scholar
75. M. Sankaran, N. P. Hanan, R. J. Scholes et al., “Determinants of woody cover in African savannas,” Nature, vol. 438, no. 7069, pp. 846–849, 2005. View at: Publisher Site | Google Scholar
76. S. Archibald and R. J. Scholes, “Leaf green-up in a semi-arid African savanna -separating tree and grass responses to environmental cues,” Journal of Vegetation Science, vol. 18, no. 4, pp. 583–594, 2007. View at: Publisher Site | Google Scholar
77. Y. Zhou, B. J. Wigley, M. F. Case, C. Coetsee, and A. C. Staver, “Rooting depth as a key woody functional trait in savannas,” New Phytologist, vol. 227, no. 5, pp. 1350–1361, 2020. View at: Publisher Site | Google Scholar
78. T. M. Scanlon, J. D. Albertson, K. K. Caylor, and C. A. Williams, “Determining land surface fractional cover from NDVI and rainfall time series for a savanna ecosystem,” Remote Sensing of Environment, vol. 82, no. 2-3, pp. 376–388, 2002. View at: Publisher Site | Google Scholar
79. J. Y. Anchang, L. Prihodko, A. T. Kaptué et al., “Trends in woody and herbaceous vegetation in the savannas of West Africa,” Remote Sensing, vol. 11, no. 5, p. 576, 2019. View at: Publisher Site | Google Scholar
80. M. N. Kahiu and N. P. Hanan, “Estimation of woody and herbaceous leaf area index in sub-Saharan Africa using MODIS data,” Journal of Geophysical Research: Biogeosciences, vol. 123, no. 1, pp. 3–17, 2018. View at: Publisher Site | Google Scholar
81. M. L. Roderick, I. R. Noble, and S. W. Cridland, “Estimating woody and herbaceous vegetation cover from time series satellite observations,” Global Ecology and Biogeography, vol. 8, no. 6, pp. 501–508, 1999. View at: Publisher Site | Google Scholar
82. H. Lu, M. R. Raupach, T. R. McVicar, and D. J. Barrett, “Decomposition of vegetation cover into woody and herbaceous components using AVHRR NDVI time series,” Remote Sensing of Environment, vol. 86, no. 1, pp. 1–18, 2003. View at: Publisher Site | Google Scholar
83. U. Gessner, M. Machwitz, C. Conrad, and S. Dech, “Estimating the fractional cover of growth forms and bare surface in savannas. A multi-resolution approach based on regression tree ensembles,” Remote Sensing of Environment, vol. 129, pp. 90–102, 2013. View at: Publisher Site | Google Scholar
84. S. Horion, R. Fensholt, T. Tagesson, and A. Ehammer, “Using earth observation-based dry season NDVI trends for assessment of changes in tree cover in the Sahel,” International Journal of Remote Sensing, vol. 35, no. 7, pp. 2493–2515, 2014. View at: Publisher Site | Google Scholar
85. M. Brandt, P. Hiernaux, T. Tagesson et al., “Woody plant cover estimation in drylands from Earth Observation based seasonal metrics,” Remote Sensing of Environment, vol. 172, pp. 28–38, 2016. View at: Publisher Site | Google Scholar
86. D. Helman, I. M. Lensky, N. Tessler, and Y. Osem, “A phenology-based method for monitoring woody and herbaceous vegetation in Mediterranean forests from NDVI time series,” Remote Sensing, vol. 7, no. 9, pp. 12314–12335, 2015. View at: Publisher Site | Google Scholar
87. W. Wu, E. De Pauw, and U. Helldén, “Assessing woody biomass in African tropical savannahs by multiscale remote sensing,” International Journal of Remote Sensing, vol. 34, no. 13, pp. 4525–4549, 2013. View at: Publisher Site | Google Scholar
88. R. L. Nagelkirk and K. M. Dahlin, “Woody cover fractions in African savannas from Landsat and high-resolution imagery,” Remote Sensing, vol. 12, no. 5, p. 813, 2020. View at: Publisher Site | Google Scholar
89. A. Bey, A. Sánchez-Paus Díaz, D. Maniatis et al., “Collect Earth: land use and land cover assessment through augmented visual interpretation,” Remote Sensing, vol. 8, no. 10, p. 807, 2016. View at: Publisher Site | Google Scholar
90. M. Karlson, M. Ostwald, H. Reese, H. R. Bazié, and B. Tankoano, “Assessing the potential of multi-seasonal WorldView-2 imagery for mapping West African agroforestry tree species,” International Journal of Applied Earth Observation and Geoinformation, vol. 50, pp. 80–88, 2016. View at: Publisher Site | Google Scholar
91. Ż. Kaszta, R. van de Kerchove, A. Ramoelo et al., “Seasonal separation of African savanna components using Worldview-2 imagery: a comparison of pixel- and object-based approaches and selected classification algorithms,” Remote Sensing, vol. 8, no. 9, p. 763, 2016. View at: Publisher Site | Google Scholar
92. C. S. R. Neigh, J. G. Masek, and J. E. Nickeson, “High-resolution satellite data open for government research,” Eos, Transactions American Geophysical Union, vol. 94, no. 13, pp. 121–123, 2013. View at: Publisher Site | Google Scholar
93. D. Schimel, F. D. Schneider, and JPL Carbon and Ecosystem Participants, “Flux towers in the sky: global ecology from space,” New Phytologist, vol. 224, no. 2, pp. 570–584, 2019. View at: Publisher Site | Google Scholar
94. N. Haughton, G. Abramowitz, M. G. de Kauwe, and A. J. Pitman, “Does predictability of fluxes vary between FLUXNET sites?” Biogeosciences, vol. 15, no. 14, pp. 4495–4513, 2018. View at: Publisher Site | Google Scholar
95. R. A. Paloschi, D. M. Ramos, D. J. Ventura et al., “Environmental drivers of water use for Caatinga woody plant species: combining remote sensing phenology and sap flow measurements,” Remote Sensing, vol. 13, no. 1, p. 75, 2021. View at: Publisher Site | Google Scholar
96. B. Alberton, R. da Silva Torres, T. Sanna Freire Silva, H. Rocha, M. S. B. Moura, and L. Morellato, “Leafing patterns and drivers across seasonally dry tropical communities,” Remote Sensing, vol. 11, no. 19, p. 2267, 2019. View at: Publisher Site | Google Scholar
97. Y. Liu, M. J. Hill, X. Zhang et al., “Using data from Landsat, MODIS, VIIRS and PhenoCams to monitor the phenology of California oak/grass savanna and open grassland across spatial scales,” Agricultural and Forest Meteorology, vol. 237-238, pp. 311–325, 2017. View at: Publisher Site | Google Scholar
98. C. E. Moore, J. Beringer, B. Evans, L. B. Hutley, and N. J. Tapper, “Tree–grass phenology information improves light use efficiency modelling of gross primary productivity for an Australian tropical savanna,” Biogeosciences, vol. 14, no. 1, pp. 111–129, 2017. View at: Publisher Site | Google Scholar
99. D. M. Browning, J. W. Karl, D. Morin, A. Richardson, and C. Tweedie, “Phenocams Bridge the Gap between Field and Satellite Observations in an Arid Grassland Ecosystem,” Remote Sensing, vol. 9, no. 10, p. 1071, 2017. View at: Publisher Site | Google Scholar
100. C. Song, “Optical remote sensing of forest leaf area index and biomass,” Progress in Physical Geography: Earth and Environment, vol. 37, no. 1, pp. 98–113, 2013. View at: Publisher Site | Google Scholar
101. T. K. Alexandridis, G. Ovakoglou, and J. G. P. W. Clevers, “Relationship between MODIS EVI and LAI across time and space,” Geocarto International, vol. 35, no. 13, pp. 1385–1399, 2020. View at: Publisher Site | Google Scholar
102. L. Naidoo, R. Mathieu, R. Main, K. Wessels, and G. P. Asner, “L-band Synthetic Aperture Radar imagery performs better than optical datasets at retrieving woody fractional cover in deciduous, dry savannahs,” International Journal of Applied Earth Observation and Geoinformation, vol. 52, pp. 54–64, 2016. View at: Publisher Site | Google Scholar
103. I. H. Woodhouse, E. T. A. Mitchard, M. Brolly, D. Maniatis, and C. M. Ryan, “Radar backscatter is not a 'direct measure' of forest biomass,” Nature Climate Change, vol. 2, no. 8, pp. 556-557, 2012. View at: Publisher Site | Google Scholar
104. S. Sinha, C. Jeganathan, L. K. Sharma, and M. S. Nathawat, “A review of radar remote sensing for biomass estimation,” International Journal of Environmental Science and Technology, vol. 12, no. 5, pp. 1779–1792, 2015. View at: Publisher Site | Google Scholar
105. J. Truckenbrodt, T. Freemantle, C. Williams et al., “Towards Sentinel-1 SAR analysis-ready data: a best practices assessment on preparing backscatter data for the cube,” Data, vol. 4, no. 3, p. 93, 2019. View at: Publisher Site | Google Scholar
106. A. I. Flores-Anderson, K. E. Herndon, R. B. Thapa, and E. Cherrington, The SAR Handbook: Comprehensive Methodologies for Forest Monitoring and Biomass Estimation, SERVIR Global Science Coordination Office, National Space Science and Technology Center, Huntsville, AL, USA, 2019.
107. S. A. Sader, “Forest biomass, canopy structure, and species composition relationships with multipolarization L-band synthetic aperture radar data,” Photogrammetric Engineering and Remote Sensing, vol. 53, no. 2, pp. 193–202, 1987. View at: Google Scholar
108. S. Mermoz, T. Le Toan, L. Villard, M. Réjou-Méchain, and J. Seifert-Granzin, “Biomass assessment in the Cameroon savanna using ALOS PALSAR data,” Remote Sensing of Environment, vol. 155, pp. 109–119, 2014. View at: Publisher Site | Google Scholar
109. S. D. de Miranda, M. Bustamante, M. Palace, S. Hagen, M. Keller, and L. G. Ferreira, “Regional variations in biomass distribution in Brazilian savanna woodland,” Biotropica, vol. 46, no. 2, pp. 125–138, 2014. View at: Publisher Site | Google Scholar
110. J.-P. Wigneron, L. Fan, P. Ciais et al., “Tropical forests did not recover from the strong 2015–2016 El Niño event,” Science Advances, vol. 6, no. 6, 2020. View at: Publisher Site | Google Scholar
111. L. Fan, J.-P. Wigneron, P. Ciais et al., “Satellite-observed pantropical carbon dynamics,” Nature Plants, vol. 5, no. 9, pp. 944–951, 2019. View at: Publisher Site | Google Scholar
112. Y. Y. Liu, A. I. J. M. van Dijk, R. A. M. de Jeu et al., “Recent reversal in loss of global terrestrial biomass,” Nature Climate Change, vol. 5, no. 5, pp. 470–474, 2015. View at: Publisher Site | Google Scholar
113. A. Bouvet, S. Mermoz, T. le Toan et al., “An above-ground biomass map of African savannahs and woodlands at 25 m resolution derived from ALOS PALSAR,” Remote Sensing of Environment, vol. 206, pp. 156–173, 2018. View at: Publisher Site | Google Scholar
114. D. Gwenzi, “Lidar remote sensing of savanna biophysical attributes: opportunities, progress, and challenges,” International Journal of Remote Sensing, vol. 38, no. 1, pp. 235–257, 2017. View at: Publisher Site | Google Scholar
115. S. Hancock, C. McGrath, C. Lowe, I. Davenport, and I. Woodhouse, “Requirements for a global lidar system: spaceborne lidar with wall-to-wall coverage,” Royal Society Open Science, vol. 8, no. 12, article 211166, 2021. View at: Publisher Site | Google Scholar
116. D. Gwenzi and M. A. Lefsky, “Modeling canopy height in a savanna ecosystem using spaceborne lidar waveforms,” Remote Sensing of Environment, vol. 154, pp. 338–344, 2014. View at: Publisher Site | Google Scholar
117. D. Gwenzi and M. A. Lefsky, “Spatial modeling of Lidar-derived woody biomass estimates collected along transects in a heterogeneous savanna landscape,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 10, no. 1, pp. 372–384, 2017. View at: Publisher Site | Google Scholar
118. V. Avitabile, M. Herold, G. B. M. Heuvelink et al., “An integrated pan-tropical biomass map using multiple reference datasets,” Global Change Biology, vol. 22, no. 4, pp. 1406–1420, 2016. View at: Publisher Site | Google Scholar
119. A. Baccini, S. J. Goetz, W. S. Walker et al., “Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps,” Nature Climate Change, vol. 2, no. 3, pp. 182–185, 2012. View at: Publisher Site | Google Scholar
120. S. S. Saatchi, N. L. Harris, S. Brown et al., “Benchmark map of forest carbon stocks in tropical regions across three continents,” Proceedings of the National Academy of Sciences of the United States of America, vol. 108, no. 24, pp. 9899–9904, 2011. View at: Publisher Site | Google Scholar
121. D. Gwenzi, M. A. Lefsky, V. P. Suchdeo, and D. J. Harding, “Prospects of the ICESat-2 laser altimetry mission for savanna ecosystem structural studies based on airborne simulation data,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 118, pp. 68–82, 2016. View at: Publisher Site | Google Scholar
122. P. M. Montesano, J. Rosette, G. Sun et al., “The uncertainty of biomass estimates from modeled ICESat-2 returns across a boreal forest gradient,” Remote Sensing of Environment, vol. 158, pp. 95–109, 2015. View at: Publisher Site | Google Scholar
123. R. Dubayah, J. B. Blair, S. Goetz et al., “The Global Ecosystem Dynamics Investigation: High-resolution laser ranging of the Earth's forests and topography,” Science of Remote Sensing, vol. 1, article 100002, 2020. View at: Publisher Site | Google Scholar
124. X. Li, K. J. Wessels, J. D. Armston et al., “First validation of GEDI vegetation structure metrics in South African savannas,” AGU Fall Meeting Abstracts, American Geophysical UnionPage Range, 2020. View at: Google Scholar
125. P. M. Seevers, P. N. Jensen, and J. V. Drew, “Satellite imagery for assessing range fire damage in the Nebraska Sandhills,” Journal of Range Management, vol. 26, no. 6, pp. 462-463, 1973. View at: Publisher Site | Google Scholar
126. C. Quintano, A. Fernández-Manso, L. Calvo, E. Marcos, and L. Valbuena, “Land surface temperature as potential indicator of burn severity in forest Mediterranean ecosystems,” International Journal of Applied Earth Observation and Geoinformation, vol. 36, pp. 1–12, 2015. View at: Publisher Site | Google Scholar
127. L. Vlassova, F. Pérez-Cabello, M. R. Mimbrero, R. Llovería, and A. García-Martín, “Analysis of the relationship between land surface temperature and wildfire severity in a series of Landsat images,” Remote Sensing, vol. 6, no. 7, pp. 6136–6162, 2014. View at: Publisher Site | Google Scholar
128. M. J. Wooster, G. J. Roberts, L. Giglio et al., “Satellite remote sensing of active fires: history and current status, applications and future requirements,” Remote Sensing of Environment, vol. 267, article 112694, 2021. View at: Publisher Site | Google Scholar
129. B. M. Jenkins, L. L. Baxter, T. R. Miles, and T. R. Miles, “Combustion properties of biomass,” Fuel Processing Technology, vol. 54, no. 1-3, pp. 17–46, 1998. View at: Publisher Site | Google Scholar
130. G. Roberts, M. J. Wooster, and E. Lagoudakis, “Annual and diurnal African biomass burning temporal dynamics,” Biogeosciences, vol. 6, no. 5, pp. 849–866, 2009. View at: Publisher Site | Google Scholar
131. P. Laurent, F. Mouillot, M. V. Moreno, C. Yue, and P. Ciais, “Varying relationships between fire radiative power and fire size at a global scale,” Biogeosciences, vol. 16, no. 2, pp. 275–288, 2019. View at: Publisher Site | Google Scholar
132. J. T. Randerson, Y. Chen, G. R. van der Werf, B. M. Rogers, and D. C. Morton, “Global burned area and biomass burning emissions from small fires,” Journal of Geophysical Research: Biogeosciences, vol. 117, no. G4, 2012. View at: Publisher Site | Google Scholar
133. R. Ramo, E. Roteta, I. Bistinas et al., “African burned area and fire carbon emissions are strongly impacted by small fires undetected by coarse resolution satellite data,” Proceedings of the National Academy of Sciences of the United States of America, vol. 118, no. 9, article e2011160118, 2021. View at: Publisher Site | Google Scholar
134. E. J. Fusco, J. T. Finn, J. T. Abatzoglou, J. K. Balch, S. Dadashi, and B. A. Bradley, “Detection rates and biases of fire observations from MODIS and agency reports in the conterminous United States,” Remote Sensing of Environment, vol. 220, pp. 30–40, 2019. View at: Publisher Site | Google Scholar
135. X. Hu, Y. Ban, and A. Nascetti, “Sentinel-2 MSI data for active fire detection in major fire-prone biomes: a multi-criteria approach,” International Journal of Applied Earth Observation and Geoinformation, vol. 101, article 102347, 2021. View at: Publisher Site | Google Scholar
136. S. S. Kumar and D. P. Roy, “Global operational land imager Landsat-8 reflectance-based active fire detection algorithm,” International Journal of Digital Earth, vol. 11, no. 2, pp. 154–178, 2018. View at: Publisher Site | Google Scholar
137. Y. Ban, P. Zhang, A. Nascetti, A. R. Bevington, and M. A. Wulder, “Near real-time wildfire progression monitoring with Sentinel-1 SAR time series and deep learning,” Scientific Reports, vol. 10, no. 1, p. 1322, 2020. View at: Publisher Site | Google Scholar
138. E. Chuvieco, F. Mouillot, G. R. van der Werf et al., “Historical background and current developments for mapping burned area from satellite Earth observation,” Remote Sensing of Environment, vol. 225, pp. 45–64, 2019. View at: Publisher Site | Google Scholar
139. D. P. Roy, H. Huang, L. Boschetti et al., “Landsat-8 and Sentinel-2 burned area mapping - A combined sensor multi- temporal change detection approach,” Remote Sensing of Environment, vol. 231, article 111254, 2019. View at: Publisher Site | Google Scholar
140. F. Ngadze, K. S. Mpakairi, B. Kavhu, H. Ndaimani, and M. S. Maremba, “Exploring the utility of Sentinel-2 MSI and Landsat 8 OLI in burned area mapping for a heterogenous savannah landscape,” PLoS One, vol. 15, no. 5, article e0232962, 2020. View at: Publisher Site | Google Scholar
141. M. A. Belenguer-Plomer, M. A. Tanase, E. Chuvieco, and F. Bovolo, “CNN-based burned area mapping using radar and optical data,” Remote Sensing of Environment, vol. 260, article 112468, 2021. View at: Publisher Site | Google Scholar
142. C. Yin, B. He, M. Yebra et al., “Improving burn severity retrieval by integrating tree canopy cover into radiative transfer model simulation,” Remote Sensing of Environment, vol. 236, article 111454, 2020. View at: Publisher Site | Google Scholar
143. M. Liu, S. Popescu, and L. Malambo, “Feasibility of burned area mapping based on ICESAT−2 photon counting data,” Remote Sensing, vol. 12, no. 1, p. 24, 2020. View at: Publisher Site | Google Scholar
144. P. H. Freeborn, M. J. Wooster, G. Roberts, B. D. Malamud, and W. Xu, “Development of a virtual active fire product for Africa through a synthesis of geostationary and polar orbiting satellite data,” Remote Sensing of Environment, vol. 113, no. 8, pp. 1700–1711, 2009. View at: Publisher Site | Google Scholar
145. B. Hally, L. Wallace, K. Reinke, S. Jones, and A. Skidmore, “Advances in active fire detection using a multi-temporal method for next-generation geostationary satellite data,” International Journal of Digital Earth, vol. 12, no. 9, pp. 1030–1045, 2019. View at: Publisher Site | Google Scholar
146. M. Amraoui, C. C. DaCamara, and J. M. C. Pereira, “Detection and monitoring of African vegetation fires using MSG-SEVIRI imagery,” Remote Sensing of Environment, vol. 114, no. 5, pp. 1038–1052, 2010. View at: Publisher Site | Google Scholar
147. W. Xu, M. J. Wooster, T. Kaneko, J. He, T. Zhang, and D. Fisher, “Major advances in geostationary fire radiative power (FRP) retrieval over Asia and Australia stemming from use of Himarawi-8 AHI,” Remote Sensing of Environment, vol. 193, pp. 138–149, 2017. View at: Publisher Site | Google Scholar
148. W. Xu, M. J. Wooster, G. Roberts, and P. Freeborn, “New GOES imager algorithms for cloud and active fire detection and fire radiative power assessment across North, South and Central America,” Remote Sensing of Environment, vol. 114, no. 9, pp. 1876–1895, 2010. View at: Publisher Site | Google Scholar
149. N. Andela, D. C. Morton, L. Giglio et al., “The Global Fire Atlas of individual fire size, duration, speed and direction,” Earth System Science Data, vol. 11, no. 2, pp. 529–552, 2019. View at: Publisher Site | Google Scholar
150. D. D'Onofrio, J. von Hardenberg, and M. Baudena, “Not only trees: grasses determine African tropical biome distributions via water limitation and fire,” Global Ecology and Biogeography, vol. 27, no. 6, pp. 714–725, 2018. View at: Publisher Site | Google Scholar
151. T. T. van Leeuwen, G. R. van der Werf, A. A. Hoffmann et al., “Biomass burning fuel consumption rates: a field measurement database,” Biogeosciences, vol. 11, no. 24, pp. 7305–7329, 2014. View at: Publisher Site | Google Scholar
152. D. M. J. S. Bowman, C. A. Kolden, J. T. Abatzoglou, F. H. Johnston, G. R. van der Werf, and M. Flannigan, “Vegetation fires in the Anthropocene,” Nature Reviews Earth & Environment, vol. 1, no. 10, pp. 500–515, 2020. View at: Publisher Site | Google Scholar
153. J. Franke, A. C. S. Barradas, M. A. Borges et al., “Fuel load mapping in the Brazilian Cerrado in support of integrated fire management,” Remote Sensing of Environment, vol. 217, pp. 221–232, 2018. View at: Publisher Site | Google Scholar
154. R. V. Leite, C. A. Silva, E. N. Broadbent et al., “Large scale multi-layer fuel load characterization in tropical savanna using GEDI spaceborne lidar data,” Remote Sensing of Environment, vol. 268, article 112764, 2022. View at: Publisher Site | Google Scholar
155. D. Helman, “Land surface phenology: What do we really 'see' from space?” Science of The Total Environment, vol. 618, pp. 665–673, 2018. View at: Publisher Site | Google Scholar
156. S. Piao, Q. Liu, A. Chen et al., “Plant phenology and global climate change: current progresses and challenges,” Global Change Biology, vol. 25, no. 6, pp. 1922–1940, 2019. View at: Publisher Site | Google Scholar
157. N. D. Masia, N. Stevens, and S. Archibald, “Identifying phenological functional types in savanna trees,” African Journal of Range & Forage Science, vol. 35, no. 2, pp. 81–88, 2018. View at: Publisher Site | Google Scholar
158. S. I. Higgins, M. D. Delgado-Cartay, E. C. February, and H. J. Combrink, “Is there a temporal niche separation in the leaf phenology of savanna trees and grasses?” Journal of Biogeography, vol. 38, no. 11, pp. 2165–2175, 2011. View at: Publisher Site | Google Scholar
159. S. Manzoni, G. Vico, S. Thompson, F. Beyer, and M. Weih, “Contrasting leaf phenological strategies optimize carbon gain under droughts of different duration,” Advances in Water Resources, vol. 84, pp. 37–51, 2015. View at: Publisher Site | Google Scholar
160. M. Owe, R. de Jeu, and J. Walker, “A methodology for surface soil moisture and vegetation optical depth retrieval using the microwave polarization difference index,” IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 8, pp. 1643–1654, 2001. View at: Publisher Site | Google Scholar
161. Y. Y. Liu, R. A. M. de Jeu, M. F. McCabe, J. P. Evans, and A. I. J. M. van Dijk, “Global long-term passive microwave satellite-based retrievals of vegetation optical depth,” Geophysical Research Letters, vol. 38, no. 18, 2011. View at: Publisher Site | Google Scholar
162. Y. Li, K. Guan, P. Gentine et al., “Estimating global ecosystem isohydry/anisohydry using active and passive microwave satellite data,” Journal of Geophysical Research: Biogeosciences, vol. 122, no. 12, pp. 3306–3321, 2017. View at: Publisher Site | Google Scholar
163. K. M. Dahlin, D. D. Ponte, E. Setlock, and R. Nagelkirk, “Global patterns of drought deciduous phenology in semi-arid and savanna-type ecosystems,” Ecography, vol. 40, no. 2, pp. 314–323, 2017. View at: Publisher Site | Google Scholar
164. C. M. Ryan, M. Williams, J. Grace, E. Woollen, and C. E. R. Lehmann, “Pre‐rain green-up is ubiquitous across southern tropical Africa: implications for temporal niche separation and model representation,” New Phytologist, vol. 213, no. 2, pp. 625–633, 2017. View at: Publisher Site | Google Scholar
165. T. Adole, J. Dash, and P. M. Atkinson, “Large-scale prerain vegetation green-up across Africa,” Global Change Biology, vol. 24, no. 9, pp. 4054–4068, 2018. View at: Publisher Site | Google Scholar
166. M. A. Whitecross, E. T. F. Witkowski, and S. Archibald, “No two are the same: assessing variability in broad-leaved savanna tree phenology, with watering, from 2012 to 2014 at Nylsvley, South Africa,” South African Journal of Botany, vol. 105, pp. 123–132, 2016. View at: Publisher Site | Google Scholar
167. M. A. Whitecross, E. T. F. Witkowski, and S. Archibald, “Savanna tree-grass interactions: a phenological investigation of green-up in relation to water availability over three seasons,” South African Journal of Botany, vol. 108, pp. 29–40, 2017. View at: Publisher Site | Google Scholar
168. R. Rani, K. Abramowicz, D. S. Falster, F. Sterck, and Å. Brännström, “Effects of bud-flushing strategies on tree growth,” Tree Physiology, vol. 38, no. 9, pp. 1384–1393, 2018. View at: Publisher Site | Google Scholar
169. E. C. February and S. I. Higgins, “Rapid Leaf Deployment Strategies in a Deciduous Savanna,” PLoS One, vol. 11, no. 6, article e0157833, 2016. View at: Publisher Site | Google Scholar
170. B. D. Tapley, S. Bettadpur, J. C. Ries, P. F. Thompson, and M. M. Watkins, “GRACE measurements of mass variability in the earth system,” Science, vol. 305, no. 5683, pp. 503–505, 2004. View at: Publisher Site | Google Scholar
171. V. Humphrey, J. Zscheischler, P. Ciais, L. Gudmundsson, S. Sitch, and S. I. Seneviratne, “Sensitivity of atmospheric CO₂ growth rate to observed changes in terrestrial water storage,” Nature, vol. 560, no. 7720, pp. 628–631, 2018. View at: Publisher Site | Google Scholar
172. N. Madani, J. S. Kimball, N. C. Parazoo et al., “Below-surface water mediates the response of African forests to reduced rainfall,,” Environmental Research Letters, vol. 15, no. 3, article 034063, 2020. View at: Publisher Site | Google Scholar
173. F. Tian, J.-P. Wigneron, P. Ciais et al., “Coupling of ecosystem-scale plant water storage and leaf phenology observed by satellite,” Nature Ecology & Evolution, vol. 2, no. 9, pp. 1428–1435, 2018. View at: Publisher Site | Google Scholar
174. T. Tagesson, S. Horion, H. Nieto et al., “Disaggregation of SMOS soil moisture over West Africa using the Temperature and Vegetation Dryness Index based on SEVIRI land surface parameters,” Remote Sensing of Environment, vol. 206, pp. 424–441, 2018. View at: Publisher Site | Google Scholar
175. Z. Sun, M. Gebremichael, J. Ardö, and H. A. R. de Bruin, “Mapping daily evapotranspiration and dryness index in the East African highlands using MODIS and SEVIRI data,” Hydrology and Earth System Sciences, vol. 15, no. 1, pp. 163–170, 2011. View at: Publisher Site | Google Scholar
176. R. Fensholt, S. Huber, S. R. Proud, and C. Mbow, “Detecting canopy water status using shortwave infrared reflectance data from polar orbiting and geostationary platforms,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 3, no. 3, pp. 271–285, 2010. View at: Publisher Site | Google Scholar
177. W. Zhao, J. Labed, X. Zhang, and Z. L. Li, “Surface soil moisture estimation from SEVIRI data onboard MSG satellite,” in 2010 IEEE International Geoscience and Remote Sensing Symposium, pp. 3865–3868, Honolulu, HI, USA, 2010. View at: Publisher Site | Google Scholar
178. M. García, I. Sandholt, P. Ceccato et al., “Actual evapotranspiration in drylands derived from in-situ and satellite data: assessing biophysical constraints,” Remote Sensing of Environment, vol. 131, pp. 103–118, 2013. View at: Publisher Site | Google Scholar
179. K. Guan, D. Medvigy, E. F. Wood, K. K. Caylor, S. Li, and S. J. Jeong, “Deriving vegetation phenological time and trajectory information over Africa using SEVIRI daily LAI,” IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 2, pp. 1113–1130, 2014. View at: Publisher Site | Google Scholar
180. A. Cord, C. Conrad, M. Schmidt, and S. Dech, “Standardized FAO-LCCS land cover mapping in heterogeneous tree savannas of West Africa,” Journal of Arid Environments, vol. 74, no. 9, pp. 1083–1091, 2010. View at: Publisher Site | Google Scholar
181. A. Baglatzi and W. Kuhn, “On the formulation of conceptual spaces for land cover classification systems,” in Geographic Information Science at the Heart of Europe, D. Vandenbroucke, B. Bucher, and J. Crompvoets, Eds., pp. 173–188, Springer International Publishing, Cham, Switzerland, 2013. View at: Google Scholar
182. F. E. Putz and K. H. Redford, “The importance of defining ‘forest’: tropical forest degradation, deforestation, long-term phase shifts, and further transitions,” Biotropica, vol. 42, no. 1, pp. 10–20, 2010. View at: Publisher Site | Google Scholar
183. A. C. Staver, S. Archibald, and S. Levin, “Tree cover in sub-Saharan Africa: rainfall and fire constrain forest and savanna as alternative stable states,” Ecology, vol. 92, no. 5, pp. 1063–1072, 2011. View at: Publisher Site | Google Scholar
184. D. Gross, F. Achard, G. Dubois, A. Brink, and H. H. T. Prins, “Uncertainties in tree cover maps of sub-Saharan Africa and their implications for measuring progress towards CBD Aichi Targets,” Remote Sensing in Ecology and Conservation, vol. 4, no. 2, pp. 94–112, 2018. View at: Publisher Site | Google Scholar
185. R. G. Congalton, J. Gu, K. Yadav, P. Thenkabail, and M. Ozdogan, “Global land cover mapping: a review and uncertainty analysis,” Remote Sensing, vol. 6, no. 12, pp. 12070–12093, 2014. View at: Publisher Site | Google Scholar
186. J. R. Smith, A. D. Letten, P.-J. Ke et al., “A global test of ecoregions,” Nature Ecology & Evolution, vol. 2, no. 12, pp. 1889–1896, 2018. View at: Publisher Site | Google Scholar
187. L. B. Zeferino, L. F. T. . Souza, C. H. . Amaral, E. I. Fernandes Filho, and T. S. . Oliveira, “Does environmental data increase the accuracy of land use and land cover classification?” International Journal of Applied Earth Observation and Geoinformation, vol. 91, article 102128, 2020. View at: Publisher Site | Google Scholar
188. P. Hurskainen, H. Adhikari, M. Siljander, P. K. E. Pellikka, and A. Hemp, “Auxiliary datasets improve accuracy of object-based land use/land cover classification in heterogeneous savanna landscapes,” Remote Sensing of Environment, vol. 233, article 111354, 2019. View at: Publisher Site | Google Scholar
189. W. Wu, Y. Sun, K. Xiao, and Q. Xin, “Development of a global annual land surface phenology dataset for 1982-2018 from the AVHRR data by implementing multiple phenology retrieving methods,” International Journal of Applied Earth Observation and Geoinformation, vol. 103, article 102487, 2021. View at: Publisher Site | Google Scholar
190. L. Poggio, L. M. de Sousa, N. H. Batjes et al., “SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty,” The Soil, vol. 7, no. 1, pp. 217–240, 2021. View at: Publisher Site | Google Scholar
191. C. Funk, P. Peterson, M. Landsfeld et al., “The climate hazards infrared precipitation with stations--a new environmental record for monitoring extremes,” Scientific Data, vol. 2, no. 1, article 150066, 2015. View at: Publisher Site | Google Scholar
192. G. Otón, J. Lizundia-Loiola, M. L. Pettinari, and E. Chuvieco, “Development of a consistent global long-term burned area product (1982-2018) based on AVHRR-LTDR data,” International Journal of Applied Earth Observation and Geoinformation, vol. 103, article 102473, 2021. View at: Publisher Site | Google Scholar
193. X. Ma, M. D. Mahecha, M. Migliavacca et al., “Inferring plant functional diversity from space: the potential of Sentinel-2,” Remote Sensing of Environment, vol. 233, article 111368, 2019. View at: Publisher Site | Google Scholar
194. J. Aguirre-Gutiérrez, S. Rifai, A. Shenkin et al., “Pantropical modelling of canopy functional traits using Sentinel-2 remote sensing data,” Remote Sensing of Environment, vol. 252, article 112122, 2021. View at: Publisher Site | Google Scholar
195. A. C. Staver, “Prediction and scale in savanna ecosystems,” New Phytologist, vol. 219, no. 1, pp. 52–57, 2018. View at: Publisher Site | Google Scholar
196. C. Blundo, J. Carilla, R. Grau et al., “Taking the pulse of Earth's tropical forests using networks of highly distributed plots,” Biological Conservation, vol. 260, article 108849, 2021. View at: Google Scholar
197. D. P. Roy, H. Huang, R. Houborg, and V. S. Martins, “A global analysis of the temporal availability of PlanetScope high spatial resolution multi-spectral imagery,” Remote Sensing of Environment, vol. 264, article 112586, 2021. View at: Publisher Site | Google Scholar
198. S. Madonsela, M. A. Cho, R. Mathieu et al., “Multi-phenology WorldView-2 imagery improves remote sensing of savannah tree species,” International Journal of Applied Earth Observation and Geoinformation, vol. 58, pp. 65–73, 2017. View at: Publisher Site | Google Scholar
199. T. Dube, T. W. Gara, O. Mutanga et al., “Estimating forest standing biomass in savanna woodlands as an indicator of forest productivity using the new generation WorldView-2 sensor,” Geocarto International, vol. 33, no. 2, pp. 178–188, 2018. View at: Publisher Site | Google Scholar
200. NICFI, “New satellite images to allow anyone, anywhere, to monitor tropical deforestation,” 2020, https://www.nicfi.no/current/new-satellite-images-to-allow-anyone-anywhere-to-monitor-tropical-deforestation/. View at: Google Scholar
201. T. Miura and S. Nagai, “Monitoring terrestrial vegetation and the environment with new-generation geostationary satellites,” Journal of the Remote Sensing Society of Japan, vol. 39, no. 5, pp. 377–383, 2019. View at: Google Scholar
202. H. Hashimoto, W. Wang, J. L. Dungan et al., “New generation geostationary satellite observations support seasonality in greenness of the Amazon evergreen forests,” Nature Communications, vol. 12, no. 1, p. 684, 2021. View at: Publisher Site | Google Scholar
203. Ç. Küçük, S. Koirala, N. Carvalhais, D. G. Miralles, M. Reichstein, and M. Jung, “Characterising the response of vegetation cover to water limitation in Africa using geostationary satellites,” Earth and Space Science Open Archive, vol. 14, 2022. View at: Google Scholar
204. A. Ramoelo, A. K. Skidmore, M. Schlerf, I. M. A. Heitkönig, R. Mathieu, and M. A. Cho, “Savanna grass nitrogen to phosphorous ratio estimation using field spectroscopy and the potential for estimation with imaging spectroscopy,” International Journal of Applied Earth Observation and Geoinformation, vol. 23, pp. 334–343, 2013. View at: Publisher Site | Google Scholar
205. T. Landmann, R. Piiroinen, D. M. Makori et al., “Application of hyperspectral remote sensing for flower mapping in African savannas,” Remote Sensing of Environment, vol. 166, pp. 50–60, 2015. View at: Publisher Site | Google Scholar
206. A. D. Jacon, L. S. Galvão, R. Dalagnol, and J. R. dos Santos, “Aboveground biomass estimates over Brazilian savannas using hyperspectral metrics and machine learning models: experiences with Hyperion/EO-1,” GIScience & Remote Sensing, vol. 58, no. 7, pp. 1112–1129, 2021. View at: Publisher Site | Google Scholar
207. L. Naidoo, M. A. Cho, R. Mathieu, and G. Asner, “Classification of savanna tree species, in the Greater Kruger National Park region, by integrating hyperspectral and LiDAR data in a Random Forest data mining environment,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 69, pp. 167–179, 2012. View at: Publisher Site | Google Scholar
208. A. K. Skidmore, J. G. Ferwerda, O. Mutanga et al., “Forage quality of savannas -- Simultaneously mapping foliar protein and polyphenols for trees and grass using hyperspectral imagery,” Remote Sensing of Environment, vol. 114, no. 1, pp. 64–72, 2010. View at: Publisher Site | Google Scholar
209. M. J. Hill, A. Millington, R. Lemons, and C. New, “Functional Phenology of a Texas Post Oak Savanna from a CHRIS PROBA Time Series,” Remote Sensing, vol. 11, no. 20, p. 2388, 2019. View at: Publisher Site | Google Scholar
210. S. Quegan, T. Le Toan, J. Chave et al., “The European Space Agency BIOMASS mission: measuring forest above-ground biomass from space,” Remote Sensing of Environment, vol. 227, pp. 44–60, 2019. View at: Publisher Site | Google Scholar
211. T. Fatoyinbo, J. Armston, M. Simard et al., “The NASA AfriSAR campaign: airborne SAR and lidar measurements of tropical forest structure and biomass in support of current and future space missions,” Remote Sensing of Environment, vol. 264, article 112533, 2021. View at: Publisher Site | Google Scholar
212. R. Doi, “Assessing the reforestation effects of plantation plots in the Thai savanna based on 45 cm resolution true-color images and machine learning,” Environmental Research Letters, vol. 16, no. 1, article 014030, 2021. View at: Google Scholar
213. S. Christin, É. Hervet, and N. Lecomte, “Applications for deep learning in ecology,” Methods in Ecology and Evolution, vol. 10, no. 10, pp. 1632–1644, 2019. View at: Publisher Site | Google Scholar
214. T. C. D. Lucas, “A translucent box: interpretable machine learning in ecology,” Ecological Monographs, vol. 90, no. 4, article e01422, 2020. View at: Publisher Site | Google Scholar
215. Q. Yu, W. Ji, L. Prihodko, C. W. Ross, J. Y. Anchang, and N. P. Hanan, “Study becomes insight: ecological learning from machine learning,” Methods in Ecology and Evolution, vol. 12, no. 11, pp. 2117–2128, 2021. View at: Publisher Site | Google Scholar
216. M. Schwieder, P. J. Leitão, J. R. R. Pinto et al., “Landsat phenological metrics and their relation to aboveground carbon in the Brazilian Savanna,” Carbon Balance and Management, vol. 13, no. 1, p. 7, 2018. View at: Publisher Site | Google Scholar
217. I. Duporge, O. Isupova, S. Reece, D. W. Macdonald, and T. Wang, “Using very-high-resolution satellite imagery and deep learning to detect and count African elephants in heterogeneous landscapes,” Remote Sensing in Ecology and Conservation, vol. 7, no. 3, pp. 369–381, 2021. View at: Publisher Site | Google Scholar
218. W. J. Murdoch, C. Singh, K. Kumbier, R. Abbasi-Asl, and B. Yu, “Definitions, methods, and applications in interpretable machine learning,” Proceedings of the National Academy of Sciences of the United States of America, vol. 116, no. 44, pp. 22071–22080, 2019. View at: Publisher Site | Google Scholar
219. K. L. Yates, P. J. Bouchet, M. J. Caley et al., “Outstanding challenges in the transferability of ecological models,” Trends in Ecology & Evolution, vol. 33, no. 10, pp. 790–802, 2018. View at: Publisher Site | Google Scholar
220. S. J. Wenger and J. D. Olden, “Assessing transferability of ecological models: an underappreciated aspect of statistical validation,” Methods in Ecology and Evolution, vol. 3, no. 2, pp. 260–267, 2012. View at: Publisher Site | Google Scholar
221. M. Reichstein, G. Camps-Valls, B. Stevens et al., “Deep learning and process understanding for data-driven Earth system science,” Nature, vol. 566, no. 7743, pp. 195–204, 2019. View at: Publisher Site | Google Scholar
222. S. S. Farley, A. Dawson, S. J. Goring, and J. W. Williams, “Situating ecology as a big-data science: current advances, challenges, and solutions,” Bioscience, vol. 68, no. 8, pp. 563–576, 2018. View at: Publisher Site | Google Scholar
223. T. Kattenborn, F. E. Fassnacht, and S. Schmidtlein, “Differentiating plant functional types using reflectance: which traits make the difference?” Remote Sensing in Ecology and Conservation, vol. 5, no. 1, pp. 5–19, 2019. View at: Publisher Site | Google Scholar
224. Q. Chang, S. Zwieback, B. DeVries, and A. Berg, “Application of L-band SAR for mapping tundra shrub biomass, leaf area index, and rainfall interception,” Remote Sensing of Environment, vol. 268, article 112747, 2022. View at: Publisher Site | Google Scholar
225. Á. Moreno-Martínez, G. Camps-Valls, J. Kattge et al., “A methodology to derive global maps of leaf traits using remote sensing and climate data,” Remote Sensing of Environment, vol. 218, pp. 69–88, 2018. View at: Publisher Site | Google Scholar
226. J. Kattge, G. Bönisch, S. Díaz et al., “TRY plant trait database – enhanced coverage and open access,” Global Change Biology, vol. 26, no. 1, pp. 119–188, 2020. View at: Publisher Site | Google Scholar
227. M. Sudmanns, D. Tiede, S. Lang et al., “Big Earth data: disruptive changes in Earth observation data management and analysis?” International Journal of Digital Earth, vol. 13, no. 7, pp. 832–850, 2020. View at: Publisher Site | Google Scholar

Copyright

Copyright © 2022 Abdulhakim M. Abdi et al. Exclusive Licensee Aerospace Information Research Institute, Chinese Academy of Sciences. Distributed under a Creative Commons Attribution License (CC BY 4.0).