Comparisons of forest mapping data sets in Sahel and their implications for the assessment of the pan-African Great Green Wall Initiative

Abstract

Forests in drylands are crucial for ecosystems and restoration projects such as the pan-African Great Green Wall Initiative (GGW). Remote sensing-based forest mapping is essential for assessing GGW progress, but substantial uncertainties exist across these data sets, especially in drylands. This study analyzed the spatiotemporal patterns of seven satellite forest data sets in Sahel, including four percent tree cover data sets and three discrete land cover data sets. Consistency among these data sets was evaluated in terms of forest extent, area, and temporal dynamics. The influencing factors contributing to uncertainties were also discussed. Results revealed notable discrepancies in forest distribution and area. Spatial uncertainties were mainly found in transitional zones such as highlands and agroforestry ecosystems. Forest area estimates for 2020 ranged from 6.65 × 10⁴ km² to 6.90 × 10⁵ km², with discrete data sets generally overestimating compared to percent tree cover data sets. Temporally, these data sets showed varying trends in both area and spatial distribution. Discrete data sets, compared to percent tree cover data sets, exhibited a more stable state and forest increase. Most data sets detected increases in the Guinea Highlands and southern Chad, but discrepancies were evident in areas such as the Ethiopian Highlands. This discrepancy can primarily be attributed to the inherent challenges of forest mapping in the Sahelian dryland ecosystem, characterized by highly sparse tree distribution. Percent tree cover data sets, by quantifying the areal proportion covered by trees within remote sensing pixels, could improve the ability to characterize sparse tree cover in Sahelian dryland ecosystems, facilitating the assessment of ecological initiatives such as the GGW.

Graphical abstract

Keywords

forest mapping datasets / discrete land cover datasets / percent tree cover datasets / spatial consistency / temporal dynamics / Sahel

1 Introduction

Forests in drylands are crucial to the global environment, by contributing to the removal of CO₂ emissions (Grassi et al., 2017), biodiversity conservation (IPBES, 2019), and providing natural resources for humankind (Maestre et al., 2012). However, these dryland ecosystems, particularly in Africa, are increasingly threatened by desertification and degradation, affecting 45%−55% of the continent (Curtis et al., 2018; Baldrian et al., 2023). Sahel, located south of the Sahara Desert with sparse vegetation, is particularly vulnerable, experiencing frequent floods and droughts that have led to widespread deforestation and degradation (Nicholson et al., 2018), threatening the livelihoods of 135 million people dependent on degraded land (Angelsen et al., 2014; Oldekop et al., 2020; UNCCD, 2020). Notable droughts in the 1970s, 2010, and 2012 resulted in famine, significant biodiversity loss, and increased fire frequency (Walther, 2016). The region’s climate-vegetation dynamics (Nicholson et al., 1998) and anthropogenic pressures — such as rapid population growth, economic development, logging, and agricultural expansion — have exacerbated these challenges (Brandt et al., 2017; Brandt et al., 2018).

To address these challenges, the pan-African Great Green Wall Initiative (GGW) was launched in 2007, focused on large-scale tree planting to restore 100 million hectares of degraded lands in Sahel and surrounding areas by 2030 (Mbow, 2017; Goffner et al., 2019; Turner et al., 2021). The initiative involves 11 founding states, including Senegal, Mauritania, Mali, Burkina Faso, Niger, Nigeria, Chad, Sudan, Eritrea, Ethiopia and Djibouti, and it has achieved 17.8 million hectares restored by early 2020 (UNCCD, 2020). The GGW is a pioneering large-scale restoration effort recognized by the United Nations (UN), and emphasizes the need for accurate and reliable forest mapping to monitor progress and assess effectiveness (UNCCD, 2020).

Remote sensing provides large-scale observations with frequent revisits, making it a vital tool for forest monitoring. Satellite-based forest mapping data sets can be broadly categorized into two types: the discrete land cover data sets of forests and the percent tree cover data sets. Discrete data sets focus on forest extent, such as the 30-m resolution global forest canopy height data (Potapov-TH) (Potapov et al., 2021), the 25-m resolution PALSAR/PALSAR-2 Forest/Non-Forest Map (JAXA-FNF) (Shimada et al., 2014), and the 10-m resolution Global Forest Cover map representing forest extent (JRC-FE) released by the European Commission (Bourgoin et al., 2024). Percent tree cover data sets, on the other hand, quantify the proportion of ground covered by trees (Jennings et al., 1999; Majasalmi and Rautiainen, 2021), providing a more detailed representation of forest density heterogeneity (Defries et al., 2000). Global data sets in this category include the 250-m resolution MODIS vegetation continuous field product (MODIS-VCF, MOD44B) (Hansen et al., 2003), the 250-m resolution GLOBMAP fractional tree cover data set (GLOBMAP-FTC) (Liu and Liu, 2020; Liu et al., 2024) and the 30-m resolution Global Forest Change data set (Hansen-GFC) (Hansen et al., 2013). Recently, researchers have developed a 100-m resolution percent tree cover map of the African continent using PlanetScope and U-Net (Brandt et al., 2020; Reiner et al., 2023). These two categories of data sets have been widely used for forest monitoring in Sahel (Souverijns et al., 2020; Li et al., 2021).

Despite their potential, mapping forests in Sahel remains challenging due to the sparse vegetation and high temporal dynamics. Comparing different data sets is essential for evaluating their applicability in drylands, which is crucial for monitoring and assessing future restoration efforts (Bai et al., 2014; Grassi et al., 2017). Significant discrepancies have been reported among forest mapping data sets in drylands (Sano et al., 2021; Adzhar et al., 2022). For instance, MOD44B and Hansen-GFC underestimated dryland forests by 40%−47% (Bastin et al., 2017) and showed marked underestimation in both magnitude and spatial distribution in Sahel (Bastin et al., 2017; Zhang et al., 2019; Adzhar et al., 2022). While recently published data sets like GLOBMAP-FTC, Reiner2019, Potapov-TH, and JRC-FE provide additional perspectives, their applicability in Sahel is still understudied. Incorporating forest dynamics as a key variable is essential for effective monitoring and restoration assessment (Nesha et al., 2021). Temporal comparisons across data sets can help us better understand restoration progress (Bos et al., 2019; Wang et al., 2022), such as the GGW. For example, over the past 15−20 years in Sahel, MOD44B captured much larger variations in tree cover but exhibited poorer temporal stability, while GLOBMAP-FTC showed more stability (Gross et al., 2018; Wei et al., 2023). These discrepancies have implications for restoration assessments (Fagan, 2020; Mirzabaev et al., 2022), such as aligning project goals with on-the-ground realities. Therefore, comparing forest mapping data sets at both spatial and temporal scales is crucial for improving forest monitoring in the Sahel and supporting ecological initiatives such as the GGW.

In this study, we compare seven widely used forest mapping data sets in Sahel, including both discrete data sets and percent tree cover data sets. We evaluate their spatial consistency in forest distribution, temporal dynamics, and forest area estimates, and further analyze the influencing factors contributing to uncertainties within these data sets. Through this comparison, we aim to analyze the ability of satellite forest mapping data sets to accurately delineate forests in tropical drylands, and their applicability in supporting monitoring and assessment of ecological initiatives such as the GGW.

2 Materials and methods

2.1 Study area

Sahel (Fig. 1), bordering the Sahara Desert to the north, predominantly experiences an arid, steppe, hot climate with annual precipitation ranging from 200 to 1000 mm, showing a clear north–south gradient (Beck et al., 2018). The area is characterized by distinct wet and dry seasons, with rainfall primarily occurring between June and October, followed by a long dry season (Nicholson et al., 1998). In this climatic context, Sahel features annual herbaceous vegetation and scattered open woodlands (Le Houerou, 1980; Brandt et al., 2016). Tree cover increases progressively from north to south along the precipitation gradient, transitioning from open woodlands to humid and dense rainforests in the south (Karlson and Ostwald, 2016). Figure 1(b) shows the land cover map of the study area derived from MODIS land cover product (MCD12Q1). The underrepresentation of forests results from the sparse tree cover in the region, with many areas not meeting the IGBP definition of forest (tree cover > 60%) and being classified instead as savannas, grasslands, or shrublands.

The Sahel region is experiencing severe risks of land degradation. To mitigate local ecological and environmental challenges, African nations and international organizations have implemented multiple ecological restoration initiatives. The GGW, launched in 2007, is one of the most prominent among these efforts. Initially, the GGW focused on large-scale tree planting to act as a barrier against the southward expansion of the Sahara Desert (UNCCD, 2020). Over time, its focus has expanded beyond tree plantation to include improving livelihoods and promoting sustainable development, with more than 20 countries participating by 2020 (Turner et al., 2021). Given the challenges of implementing a physical forest wall, specific intervention areas have been continuously adjusted. Researchers have simulated the updated path in 2019 compared to that of 2007 (Fig. 1(a)) by tracking national plans available at grandemurailleverte.org website (Goffner et al., 2019).

Since Sahel does not have clear geographical boundaries, we defined 10°N−20°N and 20°W−45°E regions of Africa as the study area (Mbow, 2017; Zhang et al., 2019), which covers the vegetation zone and the core area of the GGW.

2.2 Data

2.2.1 Data and preprocessing

This study involves 7 global and continental forest mapping data sets, including 4 percent tree cover data sets (MOD44B, GLOBMAP-FTC, Reiner2019, and Hansen-GFC) with spatial resolutions ranging from 30 to 250 m, and 3 discrete land cover data sets of forests (Potapov-TH, JAXA-FNF, and JRC-FE), with spatial resolutions of 10−30 m (Table 1). These data sets offer complete coverage of the study area. Six data sets cover around 2020 and five offer multi-decadal temporal spans since around 2000 (Table 1). All data sets were reprojected to sinusoidal projection.

1) MODIS vegetation continuous field product (MODIS-VCF, MOD44B)

MOD44B (v61), at a resolution of 250 m, provides annual tree cover layers starting from 2000. The data set contains 3 continuous fractional cover layers: tree cover, non-tree vegetation, and non-vegetated surface plus surface water. These layers are derived using measures of land surface reflectance and brightness temperatures as metrics for a supervised regression tree algorithm (Hansen et al., 2003; DiMiceli et al., 2021).

It is worth noting that fractional tree cover in MOD44B is defined as “the ratio of the area of the vertical projection of green vegetation onto the ground to the total area” (DiMiceli et al., 2021), which excludes the gaps between canopies, different from other data sets (Gross et al., 2018; Li et al., 2023). To ensure comparability, MOD44B was normalized by an empirical correction coefficient for subsequent analysis (Adzhar et al., 2022).

2) GLOBMAP fractional tree cover data set (GLOBMAP-FTC)

GLOBMAP-FTC, at a resolution of 250 m, provides annual tree cover layers from 2000 to 2021. Derived from MODIS imagery, the percent tree cover was generated using a feedforward neural network that accounted for differences in spectral seasonal curves between trees and other land cover components (Liu et al., 2024).

3) Tree cover map of Africa (Reiner2019)

Reiner2019, the only continental-scale data set in this study, provides percent tree cover mapping across Africa for the year 2019, derived from PlanetScope nanosatellite constellation imagery. The original imagery, with a resolution of 3−4 m, was resampled to 1 m during modeling, and the predictions aggregated to a 100 m resolution are publicly available. The data set was developed using a U-Net model trained on over 130000 samples (Reiner et al., 2023).

4) Global Forest Change data set (Hansen-GFC)

The 30-m resolution Hansen-GFC data set was derived from Landsat imagery obtained during the growing season and used decision tree methodology to generate tree cover for 2000 and 2010. The data set also includes forest gain layer from 2000 to 2012 and annually updated forest lossyear layer since 2001 (Hansen et al., 2013).

5) Global Forest Canopy Height data set (Potapov-TH)

Potapov-TH, with a 30-m resolution, maps global forest canopy height greater than 3 m, based on a Landsat-based model calibrated by the Global Ecosystem Dynamics Investigation (GEDI) LiDAR (Potapov et al., 2021). Available in 2000, 2019, and 2020, Potapov-TH is advantageous for estimations in forest biomass and carbon storage. In this study, the forest cover maps were extracted using the tree height exceeded the thresholds of 3 m and 5 m, respectively, to analyze forest extent (Patrick, 2001; Brandt et al., 2023). We selected data from 2000 and 2020 for analysis.

6) PALSAR/PALSAR-2 Forest/Non-Forest Map (JAXA-FNF)

JAXA-FNF, derived from JAXA’s L-band Synthetic Aperture Radar (PALSAR/PALSAR-2), provides global Forest/Non-Forest maps from 2007 to 2020 at a 25-m spatial resolution. JAXA-FNF (v200a) updated the classification into 4 classes, including “Dense Forest” (tree cover > 90%) and “Non-dense Forest” (10%−90% tree cover). The data set was conducted based on the differences in backscatter coefficients across forests, and non-forest through the Random Forest (RF) method (Shimada et al., 2014). We selected data from 2007 and 2020 for subsequent analysis.

7) Global Forest Cover map (JRC-FE)

JRC-FE, provided by the Joint Research Centre (JRC) from the European Commission, offers a 10-m resolution map of forest presence and absence for 2020. It is a harmonized, multi-source fusion data set that integrated land cover data sets and tree cover data sets such as JRC Tropical Moist Forest, JRC Global Surface Water, and JRC Global Human Settlement Layer (Bourgoin et al., 2024).

2.2.2 Auxiliary data

This study also used MODIS land cover type products (MCD12Q1), and CHIRPS rainfall data set for supplementary analysis. The land cover type and rainfall data sets were used to analyze forest distribution characteristics.

1) MCD12Q1 land cover type product

MCD12Q1 (v61) is an annual land cover type product based on MODIS imagery since 2001, featuring a 500-m spatial resolution. Land cover types are derived from five classification schemes including the International Geosphere Biosphere Programme (IGBP), University of Maryland (UMD), Leaf Area Index (LAI), BIOME-Biogeochemical Cycles (BGC), and Plant Functional Types (PGT) (Friedl and Sulla-Menashe, 2022). We selected the IGBP (17) classification standard layer of MCD12Q1, in which croplands include croplands (12), cropland/natural vegetation mosaics (14); forests include evergreen needleleaf forests (1), evergreen broadleaf forests (2), deciduous needleleaf forests (3), deciduous broadleaf forests (4), mixed forests (5); grasslands (10); woody savannas (8); savannas (9); shrublands include closed shrublands (6) and open shrublands (7); water bodies (11, 15, 17); built-up lands (13); barren (16). In detail, grasslands, savannas, woody savannas, and forests are distinguished by tree cover of 10%, 30%, and 60% respectively. We would conduct zonal statistics based on the land cover types and compare the spatiotemporal distribution patterns.

According to MCD12Q1 in 2020, the proportions of croplands, forests, grassland, woody savannas, savannas, shrublands, and barren in the study area were 9.91%, 0.09%, 41.98%, 0.02%, 3.17%, 0.70%, and 43.70%, respectively.

2) CHIRPS Rainfall data set

The Climate Hazards Center InfraRed Precipitation with Station (CHIRPS) is a global rainfall data set that combines 0.05° resolution satellite imagery with in situ station data. It is widely used for monitoring and modeling climate dynamics (Funk et al., 2015). We calculated the 2000−2020 average annual rainfall from CHIRPS to qualitatively compare the performance of different data sets across varying rainfall intervals.

MCD12Q1 and CHIRPS data were reprojected to the sinusoidal projection.

2.3 Methods

This study primarily compared spatial consistency in forest distribution around 2020, temporal dynamics from around 2000 to 2020, and forest area estimates in founding states of the GGW.

2.3.1 Spatial consistency

According to the Food and Agriculture Organization (FAO), a “forest” is defined as land with a tree canopy cover of more than 10%, a minimum tree height of 5 m, and an area larger than 0.5 ha (FAO, 2000; Grainger, 2008). To ensure consistency between discrete data sets and percent tree cover data sets, we adopted the FAO definition of forests—areas with more than 10% tree cover—to compare the spatial consistency of forest distribution at a 0.05° scale in 2020. For data sets such as Reiner2019 and Hansen-GFC, where 2020 data were not available, the most recent maps were used to minimize the effect of temporal variations.

Likelihood Assessment Method is widely used to evaluate the spatial consistency of land cover products from multiple data, which can be understood as a pixel-by-pixel voting mechanism for multi-source data (Qin et al., 2017; Wang et al., 2022; Liu et al., 2023). For percent tree cover data sets, forest/non-forest maps were first generated based on the 10% threshold. All data sets were then resampled to 0.05° by the “majority” algorithm to forest/non-forest maps. This algorithm assigns the value that occurs most often of all cells in the raw value raster that belong to the same zone as the output 0.05° cell. Pixels classified as “forests” by nearly all data sets received high frequency values (e.g., value 6 or 7 in this study), indicating high spatial consistency and agreement on the presence of forest (Wang et al., 2022). Conversely, a frequency value of 1 represented the agreement on forest absence, while medium frequency values (2 to 5) reflected disagreement in forest distribution across data sets (Sexton et al., 2016).

2.3.2 Temporal dynamics

To analyze dynamic changes, this study applied change detection to discrete data sets and used the Theil–Sen Median trend analysis along with the Mann–Kendall test for percent tree cover data sets. Specifically, change detection analysis was conducted on JAXA-FNF (2007−2020) and Potapov-TH (2000−2020) to assess forest extent dynamics over the past two decades.

Theil-Sen Median trend analysis is a robust trend statistical method that can reduce the impact of outliers and it calculates the median slopes of time series data (Sen, 1968). The calculation method is as follows:

(1)

where TC_i and TC_j refer to the percent tree cover value of year i and j, respectively, and N refers to the most recent year used in the data calculation, which is 2021 in this study. slope > 0 indicates that the tree cover is increasing while slope < 0 indicates a decreasing trend. Considering data fluctuations, we believe the slope between −0.0005 and 0.0005 to be a stable state.

The Mann-Kendall (MK) trend test method is a non-parametric test method that does not require data to obey certain distribution and is not affected by outliers (Mann, 1945; Kendall, 1948). It is suitable for analyzing dynamics in ecology and climate such as hydrological and meteorological time series. The definition of the statistic Z_MK definition and its calculation formula are as follows:

(2)

where

(3)

(4)\mathrm{V}\mathrm{A}\mathrm{R}(S)=\frac{n(n-1)(2n+5)}{18},

(5)

where n represents the number of time series observations and sgn is a sign function. If | Z_MK | is greater than Z_α/2, where α represents the chosen significance level, it implies that the trend is significant. In statistical analysis, the corresponding critical values are Z_0.05 = 1.65 for α = 0.1, Z_0.025 = 1.96 for α = 0.05, and Z_0.005 = 2.58 for α = 0.001. In this study, we adopted a significance level of α = 0.05 according to previous studies (Jiang et al., 2015; Nguyen et al., 2022); therefore, a trend is considered significant if | Z_MK | exceeds 1.96.

The combination of Theil–Sen Median trend analysis and the Mann–Kendall test is widely used in pixel-level trend analysis in long-term series vegetation studies (Fensholt et al., 2012; Jiang et al., 2015). We classified the results into four categories of tree cover changes (Table 2).

Among all the data sets, we conducted the Theil-Sen Slope analysis and MK trend test on MOD44B (2000−2021) and GLOBMAP-FTC (2000−2021) at a 250 m resolution to analyze temporal dynamic analysis. Calculations were performed using the Python pymannkendall package.

2.3.3 Forest area estimation

Referring to the FAO definition, we evaluated the ability of each data set to depict forest area. For percent tree cover data sets, the forest area was calculated as the product of a pixel’s area and its percent tree cover if the pixel value exceeded 10% (Gross et al., 2018; Wei et al., 2023). For discrete land cover data sets, forest area was determined based on the area of the pixel that classified as forest.

3 Results

3.1 Spatial consistency of forests

We applied a 10% tree cover threshold to define forests and applied 3 m and 5 m thresholds to Potapov-TH to present forest cover distribution at a 0.05° scale. Forest mapping showed a distinct north–south gradient, with forests primarily distributed south of 15°N and tree cover decreased with increasing latitude. Tree cover in western Africa and the Ethiopian Highlands was slightly higher than elsewhere at the same latitude, while most northern areas either had no data available or exhibited extremely sparse tree cover. All data sets captured a peak within the 0−20% forest cover range. The histograms of forest cover show that the percent tree cover data sets displayed a slight increase in the 80%−100% range, while results from discrete land cover data sets were more evenly distributed across the 20%−100% range (Fig. 2). However, in Potapov-TH, the result with the 3 m threshold showed a slight increase near 100%, which was not observed with the 5 m threshold. This suggested that dense forests are predominantly characterized by forest canopy heights between 3 m and 5 m according to this data set.

Figure 3 illustrates the spatial consistency of forest distribution at 0.05° and the maximum forest extent is defined as the frequency value ≥ 1. The maximum extent covered 12.5% of the study area. Areas with high spatial consistency, characterized by values 6 and 7, accounted for 1.2% and 0.9%, respectively, and were mainly distributed in the southern areas. Southern Senegal, Guinea-Bissau, Guinea, and the Ethiopian Highlands have high spatial consistency, aligning with areas of dense tree cover and forest presence. In southern Senegal, Guinea-Bissau, and Guinea, the primary ecoregion is known as the Guinean forest-savanna located in West Africa and characterized by vegetation composed of tropical rainforests and grasslands with scattered trees (Fig. 3(A2)) (Fairhead and Leach, 1996). The distinctiveness of rainforests from herbaceous vegetation facilitates remote sensed forest mapping. In the Ethiopian Highlands, spatial consistency was stronger in the western part of the highlands than in the east. This could be due to the predominance of croplands and a mixed distribution of dry evergreen Afromontane forests and grasslands in the eastern region (White, 1978; Friis et al., 2010; Friedl and Sulla-Menashe, 2022). Areas with median frequencies were typically located in agricultural and pastoral land. These areas are located in the transition zone between humid and arid regions or different ecosystems, where distinguishing herbaceous vegetation from trees is a significant challenge (Fig. 3(A4)).

3.2 Temporal dynamics over the past two decades

The temporal dynamics analysis of discrete data sets revealed contrasting trends. JAXA-FNF showed little forest loss between 2007 and 2020, with a net forest area gain of 6.28 × 10⁵ km², primarily in the Guinea Highlands, southern Chad, and western Ethiopia. In contrast, Potapov-TH showed stability in most areas and exhibited limited forest gain compared to forest loss, with gains primarily concentrated in northern Nigeria. When applying the 3 m tree height threshold, Potapov-TH estimated a forest loss of 2.40 × 10⁴ km² and a gain of 1.24 × 10⁴ km² between 2000 and 2020. When applying the 5 m threshold, the estimated loss increased to 3.33 × 10⁴ km², and the gain slightly rose to 1.43 × 10⁴ km².

The dynamics of percent tree cover data sets were more pronounced. Through Theil–Sen Slope and the Mann–Kendall trend test, MOD44B identified a changing area of 2.23 × 10⁶ km², representing 32.90% of the study area, of which 77.8% of this change was attributed to a decrease in tree cover. In contrast, GLOBMAP-FTC captured only 5.03% change, with 3.41 × 10⁵ km², of which 76.7% was an increase (Fig. 4). Both data sets consistently identified areas of increase in Guinea Highlands and southern Chad but showed discrepancies in the Ethiopian Highlands, where MOD44B showed an increase in the eastern regions, while GLOBMAP-FTC captured changes in the south-west. Differences also emerged in agro-pastoral regions like northern Senegal and central Nigeria; MOD44B detected widespread, significant decreases in these areas, whereas GLOBMAP-FTC failed to represent the northern region due to missing data, instead of meaning a stable state. In addition, GLOBMAP-FTC highlighted a greater proportion of areas with increased tree cover, aligning with the greening trends reported in previous studies.

Hansen-GFC provides a forest lossyear layer from 2001 and a forest gain layer from 2000 to 2012 instead of annual tree cover layers. Forest loss here is defined as a replacement, defined as a pixel where crown cover decreases by more than 50% to 0% within a year, while forest gain represents the inverse of forest loss or an increase from non-forest to forest over the 12 years. In our study area, Hansen-GFC captured 3.83 × 10⁴ km² of forest loss from 2000 to 2021, primarily occurring between 2013 and 2021 (Fig. 5(a)) in the southern areas with denser tree cover, and almost no forest gain was observed.

3.3 Forest area estimation

There are clear differences in forest area estimates in 2020 across the data sets (Fig. 5(b)). Estimates from discrete data sets were 3 to 10 times higher than those from percent tree cover data sets, especially in areas with lower rainfall. JAXA-FNF captured the largest forest area in the study area (6.90 × 10⁵ km²), while MOD44B-2020 identified the smallest forest area, at only 6.65 × 10⁴ km². Temporally, MOD44B exhibited poor temporal stability with no discernible trend, whereas GLOBMAP-FTC showed a significant increase in forest area over time.

Figure 6 compares forest area estimates around 2020 from various remote sensing data sets across Senegal, Mauritania, Mali, Burkina Faso, Niger, Nigeria, Chad, Eritrea, Ethiopia, and Djibouti. Sudan was not considered due to territorial conflicts. The results revealed substantial discrepancies among the data sets. In most countries, discrete data sets produced similar estimates, while the percent tree cover data sets also showed internal consistency. In general, the forest area estimates from discrete data sets were higher than those from percent tree cover data sets. In the case of JAXA-FNF, forest area was significantly overestimated in most countries, with the most pronounced overestimation in Djibouti and Eritrea. Among these data sets, Reiner2019, derived from 3−4 m resolution PlanetScope imagery, was expected to be capable of capturing the distribution of sparse trees in the study area, making it a useful reference for data set comparisons. Compared to Reiner2019, discrete data sets significantly overestimated forest area, while MOD44B and GLOBMAP-FTC tended to underestimate forest area. Notably, the three percent tree cover data sets showed good consistency in Ethiopia, Eritrea, and Chad. These discrepancies may be attributable to the coarse spatial resolution of MOD44B and GLOBMAP-FTC, which may lead to the omission of low-value data.

4 Discussion

4.1 Differences and uncertainties of spatial consistency

By comparing four percent tree cover data sets and three discrete forest mapping data sets, this study found that the distribution of forests in Sahel varies across different data sets, although they generally exhibited a pattern of lower forest cover in the north and higher in the south. The sparse distribution of forests in Sahel is a primary factor contributing to these discrepancies. Additional influencing factors of uncertainties include differences in forest definitions, algorithmic approaches, mixed-pixel effects, and the remote sensing data sources used.

The definition of “forest” directly influences these differences (Sexton et al., 2016; Ferrer Velasco et al., 2022). Discrete data sets classify forests according to various criteria, such as tree height or tree canopy cover, with pixels classified as “forests” if the properties exceed certain thresholds (Wei et al., 2023; Yang et al., 2023; Table 1). This has a direct impact on the results, as shown by the differences with thresholds of 3 m or 5 m in the Potapov-TH data set (Fig. 2).

Algorithms and processing methods can also influence mapping results. For example, forests can refer to one type of land cover or a type of land use in different algorithm processing (Hansen et al., 2013; Bastin et al., 2017). Potapov-TH, JAXA-FNF, and JRC-FE all pre-masked agricultural or urban lands, whereas MOD44B, GLOBMAP-FTC, Reiner2019, and Hansen-GFC only masked water bodies (Table 1), leading to data gaps in certain areas for the former (Fig. 3(A1)). Agroforestry is widespread and crucial in Sahel through the balance between woody vegetation and soil cover (Bayala et al., 2014). In a regional study in Africa (Miller et al., 2017), researchers found that more than a quarter of smallholder farmers cultivated trees and trees/forests in farming systems, which can improve both production and farmers’ incomes (Karlson and Ostwald, 2016). Moreover, Sahal has experienced rapid cropland expansion in recent decades (Curtis et al., 2018). Consequently, masking can lead to significant uncertainties in forest mapping.

The mixed-pixel problem, that multiple land cover types influence the spectral response of a single pixel, is particularly common in Sahel where forests and herbaceous vegetation are mixed and sparsely distributed. This problem poses a significant challenge to forest mapping in the region, resulting in substantial uncertainties in discrete data sets (Yang et al., 2017; Estoque et al., 2018). For example, pixels with an actual tree cover of 20% and 90% might both be classified as the forest category (Figs. 3(A1 and A2) and 6), leading to overestimation (Wei et al., 2023). In contrast, percent tree cover data sets provide continuous and quantitative measurements, capturing the transition features and internal structure of forests at a sub-pixel scale. These data sets are particularly effective in sparsely vegetated areas and also provide flexibility in applying various thresholds for defining forests, demonstrating clear advantages for forest mapping and monitoring (Estoque et al., 2022; Melo et al., 2023; Reiner et al., 2023).

Spatial resolution can also affect the characterization capacity of forests and the results of analysis (Higginbottom et al., 2018; Chen et al., 2020). The seven forest data sets compared in this study feature spatial resolutions ranging from 10 to 250 m. Data with resolutions between 100 and 250 m provide improved spatial details compared to earlier kilometer-scale data sets while also offering the advantage of time-series analysis (Qin et al., 2017; Abdi et al., 2022). However, these resolutions are insufficient to capture certain details of sparse forests in drylands (Jung et al., 2006; Zhang et al., 2019; Yang et al., 2023), such as isolated trees in croplands and forest edges (Fig. 3(A1)). Data sets at 10 to 30 m resolution further enhance spatial mapping by reducing the phenomena of mixed pixels (Estoque et al., 2018; Liu et al., 2023). However, satellite with tens-meter resolution often present limitations in acquiring dense temporal observations, thereby introducing challenges for accurate tree extraction and the detection of interannual dynamics. Additionally, to mitigate the impact of scaling effects in our comparative analysis, all data sets with varying spatial resolutions were resampled to 0.05° grid. This process may compromise the data sets’ capacity to represent spatial patterns of trees, potentially resulting in information loss and introducing uncertainties into the analysis.

Optical imagery and radar data are the two main data sources for forest mapping. Spectral properties and phenological characteristics from time-series data are essential and widely utilized, while radar data are becoming increasingly popular due to their insensitivity to clouds, which poses significant challenges to optical remote sensing in tropical regions (Shimada et al., 2014; Estoque et al., 2018; Zhang et al., 2019). In this study, Potapov-TH used GEDI canopy height metrics to calibrate Landsat data to estimate forest canopy height, and JAXA-FNF used the threshold classification of HH and HV polarizations of ALOS-PALSAR to map forests. However, radar data tend to easily confuse forests with other elevated features such as settlements (Thapa et al., 2014) and rocks in highlands (Van Zyl, 1993). For example, JAXA-FNF incorrectly classified some rocks as forest cover in Niger Aïr Mountains (Fig. 3(A3)).

4.2 Cross-comparison of percent tree cover data sets

For further analysis, we averaged three percent tree cover data set, MOD44B-2020, GLOBMAP-FTC-2020, and Reiner2019, to a 0.05° scale for comparison based on MCD12Q1-2020 land cover data (Fig. 7). The results showed a significant linear correlation across all categories, except for “Forests” and “Woody Savannas”, which covered 0.09% and 0.02% of the study area, respectively. MOD44B was found to underestimate tree cover, consistent with previous studies indicating that MOD44B is not suitable for analysis in areas with less than 30% tree cover, with greater uncertainty in areas below 10% (Hansen et al., 2005; Staver and Hansen, 2015), despite preprocessing. The differences are exacerbated by the training data. MOD44B used discrete classification of Landsat data for 4 tree cover classes (0, 25%, 50%, 80%+), averaged to 250-m resolution fractional values as the training data, and restricted trees to those taller than 5 m (Yang and Crews, 2019). However, trees in the study area are predominantly between 3 m and 5 m (Fig. 2).

GLOBMAP-FTC-2020 behaved similarly overall to Reiner2019, although it was less effective in the northern Sahel. GLOBMAP-FTC’s training data were primarily derived from different global forest/tree cover data sets such as JAXA-FNF (25 m) and the ESA WorldCover land cover data set (10 m) (Liu et al., 2024). The researchers aggregated high-resolution discrete classification forest data sets into fractional tree cover at 250 m, excluding areas with high uncertainty. This method may lead to data gaps in complex landscapes and regions with sparse vegetation. Additionally, GLOBMAP-FTC has stronger detection capabilities in areas with over 50% tree cover, due to the clear distinction between dense forest and herbaceous vegetation in seasonal NDVI curves (Liu and Liu, 2020).

Reiner2019 captured more valid values in regions with sparse vegetation. Global data sets usually behave poorly in specific regions, due to the extensive distribution of training data which can introduce regional biases (Tropek et al., 2014; Tucker et al., 2023). In Reiner2019, the researchers used imagery at the beginning of the dry season in 2019 as the data source and training data from the manual interpretation of Google Maps and Bing Maps in the African continent (Reiner et al., 2023). High-resolution imagery enabled Reiner2019 to perform better at mapping trees outside forests in Africa. However, PlanetScope imagery is limited by 4 spectral bands and excludes time-series information. Reiner2019 therefore tends to misclassify shrubs or herbaceous vegetation as trees and presents challenges for historical mapping due to the data source used.

For further comparison, we selected three sample areas around the study area. According to the results of temporal dynamics, sample areas showed a decrease, increase, or stable state relatively and were displayed differently in Google Earth historical images (Fig. 8). The sample area of southern Senegal showed a “Significant Decrease” from MOD44B and GLOBMAP-FTC temporal dynamics results from 2000 to 2021 with tree cover decrease from Hansen-GFC between 2000 and 2010. This trend was consistent across three data sets, and it could be inferred from the historical images that the trend was driven by the expansion of cropland and built-up areas (Brandt et al., 2018). The sample area in Nigeria covers part of the national forest reserve, and it showed the same trend of tree cover increase. However, results from the sample area in southern Sudan displayed the differences, with no trend information from GLOBMAP-FTC, a decrease in MOD44B, and an increase in Hansen-GFC. High-resolution images showed that this area has been continuously covered croplands and built-up lands over the past two decades, interfering with tree cover mapping.

Apart from the fact that Hansen-GFC only recorded pixels with over 50% tree cover change, differences in temporal dynamics are also influenced by the time-series algorithms used. We found that MOD44B tended to overestimate change areas and exhibited poor inter-annual stability (Gross et al., 2018; Yang and Crews, 2019). GLOBMAP-FTC showed a significant increase in tree cover over the past 20 years, consistent with the observed “greening” trend. Although both data sets use spectral metrics including time series properties as inputs, MOD44B relied on 8 composites per year to eliminate clouds, while GLOBMAP-FTC used every 8-day clear-sky reflectance from MODIS data to generate 4 key phenology time phases each year, with gaps filled and smoothed (Liu et al., 2024). Given the low likelihood of significant and repeated changes in forest cover over consecutive years, effective monitoring requires the ability to detect both disturbances and recovery while maintaining temporal stability.

4.3 Impacts on the assessment of the GGW

The GGW initiative, initially intended to create a green barrier to halt the encroachment of the Sahara Desert, has evolved into a comprehensive regional strategy for Sustainable Land Management (SLM). SLM encompasses forestry and agriculture, water resources, and soil management, with forests playing a pivotal role (UNCCD, 2020). In drylands, forest management and restoration are crucial for maintaining the balance between precipitation and runoff, mitigating the impacts of climate change, and combating desertification (Mbow, 2017; Jones et al., 2022). Additionally, forests and trees in agriculture can enhance productivity through processes like nitrogen fixation (Machado-Silva et al., 2022) while providing resources that support economic development (Bayala et al., 2014).

The assessment of the GGW initiative requires information on forest extent, coverage, and spatiotemporal dynamics. Country reports from national authorities usually serve as the primary reference, such as the Global Forest Resources Assessment (FAO-FRA), a periodic assessment of forest resources led by FAO. However, systematic differences may arise due to varying definitions, standards, and methodologies among states (Ochieng et al., 2016; Sexton et al., 2016). For instance, FAO-FRA data on forest area excludes orchards, oil palm plantations, agroforestry, and urban trees, which are instead classified as “Other Woodlands” (FAO, 2020).

Satellites provide comparable dynamic data for forest monitoring for the GGW. Previous studies have mainly utilized a few forest mapping products to investigate forests in the Sahel region. For example, JAXA-FNF has been used to define forest extent in Sahel, revealing tree expansion in certain regions (Li et al., 2021). JRC-FE has been employed to support the monitoring of deforestation and degradation (Mansuy et al., 2024), while Potapov-TH has been utilized to estimate carbon sequestration and forest regeneration (Fagan et al., 2022). This study conducted a comparative analysis of currently available and widely used forest mapping data sets, enhancing our understanding of the applicability and limitations of different data sets in this region. We found that discrete data sets performed more poorly in monitoring and applicability in Sahel, consistent with prior findings (Verhelst et al., 2021; Wei et al., 2023). Percent tree cover data sets offer better applicability in sparse forests by reflecting the mixed pixel phenomenon and their potential to overcome the ambiguity in forest definitions (Estoque et al., 2022; Melo et al., 2023; Reiner et al., 2023).

The importance of time-series monitoring for the GGW initiative cannot be overstated. FAO-FRA, which is updated every 5−10 years, generally reflects a downward trend in forest areas across most states. However, some data are outdated in Africa (Grainger, 2008; Gross et al., 2018). Given the importance of monitoring survival rates 5 to 10 years after plantation and that intervention areas for restoration have undergone spatial adjustments over the past 20 years (Goffner et al., 2019), the limitations of FAO-FRA data are becoming evident. For example, though the sample area in Nigeria is located outside the GGW path and it was observed an obvious increase in the tree cover trend (Fig. 8). Considering the distance between the sample area and the updated GGW path, it remains unclear whether the increase was influenced by the indirect effects of the spatial adjustments. As a result, remote sensing enhances the ability to track real-time changes and identify potential restoration areas, and assess degradation risks.

Different data sets, however, vary in their definitions, spatial resolutions, and temporal coverage, which can influence their ability to monitor forest distribution and changes in this dryland region. These differences are particularly significant for the assessments of ecological restoration projects such as the GGW, where accurate mapping and monitoring of forest dynamics are essential to evaluate progress and identify priority areas for intervention. When selecting data sets, it is critical to align the type, spatial resolution and temporal coverage of the data set with the specific objectives of restoration monitoring to ensure robust and meaningful assessments.

5 Conclusions

This study comprehensively compared the spatiotemporal patterns of seven satellite forest mapping data sets in Sahel, including MOD44B (2000−2021), GLOBMAP-FTC (2000−2021), Reiner2019 (2019), Hansen-GFC (2000 and 2010), Potapov-TH (2000 and 2020), JAXA-FNF (2007 and 2020) and JRC-FE (2020). Results showed notable spatial and temporal discrepancies among forest mapping data sets in Sahel, with sparse vegetation and transitional ecosystems. Spatially, all data sets exhibited a consistent north–south gradient. Using a 10% tree cover threshold to define “forest”, areas of low consistency in forest distribution were primarily found in transitional zones, such as highlands and agroforestry ecosystems, and dense forests were often dominated by trees of 3−5 m in height. Temporally, while data sets indicated different trends in forests/tree cover, GLOBMAP-FTC demonstrated superior temporal stability and an overall increase in tree cover. The forest area in 2020 ranges from 6.65 × 10⁴ km² to 6.90 × 10⁵ km², with discrete data sets generally overestimating compared to percent tree cover data sets. These discrepancies are primarily attributed to the sparse vegetation in Sahel, as well as differences in forest definitions, algorithmic approaches, mixed-pixel effects, and the remote sensing data sources. These findings underscore the potential of remote sensing, offering critical insights for assessing ecological restoration initiatives, such as the GGW.

Percent tree cover mapping has greater potential in monitoring forests in drylands and is valuable for initiatives such as the GGW, as they effectively depict forest structure and the transitional details of tree cover. In addition, percent data sets can accommodate various definitions across tree cover thresholds for forests. In this study, Reiner2019 excels in identifying tree cover in sparse regions, and GLOBMAP-FTC demonstrates greater stability in time series monitoring. Future efforts in forest mapping should emphasize percent tree cover at medium to high spatial resolution over long time series. These advancements will not only provide robust quantitative data support, but also improve biomass estimation, climate impact assessment, and other critical analysis.

References

[1] Abdi A M, Brandt M, Abel C, Fensholt R (2022). Satellite remote sensing of savannas: current status and emerging opportunities.J Remote Sens, 2022: 9835284

[2] Adzhar R, Kelley D I, Dong N, George C, Torello Raventos M, Veenendaal E, Feldpausch T R, Phillips O L, Lewis S L, Sonké B, Taedoumg H, Schwantes Marimon B, Domingues T, Arroyo L, Djagbletey G, Saiz G, Gerard F (2022). MODIS vegetation continuous fields tree cover needs calibrating in tropical savannas.Biogeosciences, 19(5): 1377–1394

[3] Angelsen A, Jagger P, Babigumira R, Belcher B, Hogarth N J, Bauch S, Börner J, Smith-Hall C, Wunder S (2014). Environmental income and rural livelihoods: a global-comparative analysis.World Dev, 64: S12–S28

[4] Bai Y, Feng M, Jiang H, Wang J, Zhu Y, Liu Y (2014). Assessing consistency of five global land cover data sets in China.Remote Sens, 6(9): 8739–8759

[5] Baldrian P, López-Mondéjar R, Kohout P (2023). Forest microbiome and global change.Nat Rev Microbiol, 21(8): 487–501

[6] Bastin J F, Berrahmouni N, Grainger A, Maniatis D, Mollicone D, Moore R, Patriarca C, Picard N, Sparrow B, Abraham E M, Aloui K, Atesoglu A, Attore F, Bassüllü Ç, Bey A, Garzuglia M, García-Montero L G, Groot N, Guerin G, Laestadius L, Lowe A J, Mamane B, Marchi G, Patterson P, Rezende M, Ricci S, Salcedo I, Diaz A S P, Stolle F, Surappaeva V, Castro R (2017). The extent of forest in dryland biomes.Science, 356(6338): 635–638

[7] Bayala J, Sanou J, Teklehaimanot Z, Kalinganire A, Ouédraogo S J (2014). Parklands for buffering climate risk and sustaining agricultural production in the Sahel of West Africa.Curr Opin Environ Sustain, 6: 28–34

[8] Beck H E, Zimmermann N E, McVicar T R, Vergopolan N, Berg A, Wood E F (2018). Present and future Köppen-Geiger climate classification maps at 1-km resolution.Sci Data, 5(1): 180214

[9] Bos A B, De Sy V, Duchelle A E, Herold M, Martius C, Tsendbazar N E (2019). Global data and tools for local forest cover loss and REDD+ performance assessment: accuracy, uncertainty, complementarity and impact.Int J Appl Earth Obs Geoinf, 80: 295–311

[10] Bourgoin C, Ameztoy I, Verhegghen A, Descl E B, Carboni S, Bastin J-F, Beuchle R, Brink A, Defourny P, Delhez B, Fritz S, Gond V, Herold M, Lamarche C, Mansuy N, Mollicone D, Oom D, Peedell S, San-Miguel J, Colditz R, Achard F (2024). Mapping global forest cover of the year 2020 to support the EU regulation on deforestation-free supply chains. Luxembourg: Publications Office of the European Union

[11] Brandt J, Ertel J, Spore J, Stolle F (2023). Wall-to-wall mapping of tree extent in the tropics with Sentinel-1 and Sentinel-2.Remote Sens Environ, 292: 113574

[12] Brandt M, Hiernaux P, Tagesson T, Verger A, Rasmussen K, Diouf A A, Mbow C, Mougin E, Fensholt R (2016). Woody plant cover estimation in drylands from Earth Observation based seasonal metrics.Remote Sens Environ, 172: 28–38

[13] Brandt M, Rasmussen K, Hiernaux P, Herrmann S, Tucker C J, Tong X, Tian F, Mertz O, Kergoat L, Mbow C, David J L, Melocik K A, Dendoncker M, Vincke C, Fensholt R (2018). Reduction of tree cover in West African woodlands and promotion in semi-arid farmlands.Nat Geosci, 11(5): 328–333

[14] Brandt M, Rasmussen K, Peñuelas J, Tian F, Schurgers G, Verger A, Mertz O, Palmer J R B, Fensholt R (2017). Human population growth offsets climate-driven increase in woody vegetation in sub-Saharan Africa.Nat Eco Evolut, 1(4): 0081

[15] Brandt M, Tucker C J, Kariryaa A, Rasmussen K, Abel C, Small J, Chave J, Rasmussen L V, Hiernaux P, Diouf A A, Kergoat L, Mertz O, Igel C, Gieseke F, Schöning J, Li S, Melocik K, Meyer J, Sinno S, Romero E, Glennie E, Montagu A, Dendoncker M, Fensholt R (2020). An unexpectedly large count of trees in the West African Sahara and Sahel.Nature, 587(7832): 78–82

[16] Chen T, Zhou S, Liang C, Hagan D F T, Zeng N, Wang J, Shi T, Chen X, Dolman A J (2020). The greening and wetting of the Sahel have leveled off since about 1999 in relation to SST.Remote Sens (Basel), 12(17): 2723

[17] Curtis P G, Slay C M, Harris N L, Tyukavina A, Hansen M C (2018). Classifying drivers of global forest loss.Science, 361(6407): 1108–1111

[18] Defries R S, Hansen M C, Townshend J R G, Janetos A C, Loveland T R (2000). A new global 1-km dataset of percentage tree cover derived from remote sensing.Glob Change Biol, 6(2): 247–254

[19] DiMiceli C, Townshend J, Carroll M, Sohlberg R (2021). Evolution of the representation of global vegetation by vegetation continuous fields.Remote Sens Environ, 254: 112271

[20] Estoque R C, Johnson B A, Dasgupta R, Gao Y, Matsuura T, Toma T, Hirata Y, Lasco R D (2022). Rethinking forest monitoring for more meaningful global forest landscape change assessments.J Environ Manage, 317: 115478

[21] Estoque R C, Pontius R G Jr, Murayama Y, Hou H, Thapa R B, Lasco R D, Villar M A (2018). Simultaneous comparison and assessment of eight remotely sensed maps of Philippine forests.Int J Appl Earth Obs Geoinf, 67: 123–134

[22] Fagan M E (2020). A lesson unlearned? Underestimating tree cover in drylands biases global restoration maps.Glob Change Biol, 26(9): 4679–4690

[23] Fagan M E, Kim D H, Settle W, Ferry L, Drew J, Carlson H, Slaughter J, Schaferbien J, Tyukavina A, Harris N L, Goldman E, Ordway E M (2022). The expansion of tree plantations across tropical biomes.Nat Sustain, 5(8): 681–688

[24] Fairhead J, Leach M (1996). Misreading the African Landscape: Society and Ecology in a Forest-Savanna Mosaic. Cambridge: Cambridge University Press

[25] FAO (2000). Comparison of Forest Area and Forest Area Change Estimates Derived from FRA 1990 and FRA 2000

[26] FAO (2020). Global Forest Resources Assessment 2020: Main report

[27] Fensholt R, Langanke T, Rasmussen K, Reenberg A, Prince S D, Tucker C, Scholes R J, Le Q B, Bondeau A, Eastman R, Epstein H, Gaughan A E, Hellden U, Mbow C, Olsson L, Paruelo J, Schweitzer C, Seaquist J, Wessels K (2012). Greenness in semi-arid areas across the globe 1981–2007 — An earth observing satellite based analysis of trends and drivers.Remote Sens Environ, 121: 144–158

[28] Ferrer Velasco R, Lippe M, Tamayo F, Mfuni T, Sales-Come R, Mangabat C, Schneider T, Günter S (2022). Towards accurate mapping of forest in tropical landscapes: A comparison of datasets on how forest transition matters.Remote Sens Environ, 274: 112997

[29] Friedl M, Sulla-Menashe D (2022). MODIS/Terra+ Aqua land cover type yearly L3 global 500m SIN grid V061. DAAC N E L P

[30] Friis I, Demissew S, Van Breugel P (2010). Atlas of the Potential Vegetation of Ethiopia. Copenhagen: Det Kongelige Danske Videnskabernes Selskab

[31] Funk C, Peterson P, Landsfeld M, Pedreros D, Verdin J, Shukla S, Husak G, Rowland J, Harrison L, Hoell A, Michaelsen J (2015). The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes.Sci Data, 2(1): 150066

[32] Goffner D, Sinare H, Gordon L J (2019). The Great Green Wall for the Sahara and the Sahel Initiative as an opportunity to enhance resilience in Sahelian landscapes and livelihoods.Reg Environ Change, 19(5): 1417–1428

[33] Grainger A (2008). Difficulties in tracking the long-term global trend in tropical forest area.Proc Natl Acad Sci USA, 105(2): 818–823

[34] Grassi G, House J, Dentener F, Federici S, den Elzen M, Penman J (2017). The key role of forests in meeting climate targets requires science for credible mitigation.Nat Clim Chang, 7(3): 220–226

[35] Gross D, Achard F, Dubois G, Brink A, Prins H H T (2018). Uncertainties in tree cover maps of Sub-Saharan Africa and their implications for measuring progress towards CBD Aichi Targets.Remote Sens Ecol Conserv, 4(2): 94–112

[36] Hansen M C, DeFries R S, Townshend J R G, Carroll M, Dimiceli C, Sohlberg R A (2003). Global percent tree cover at a spatial resolution of 500 meters: first results of the MODIS vegetation continuous fields algorithm.Earth Interact, 7(10): 1–15

[37] Hansen M C, Potapov P V, Moore R, Hancher M, Turubanova S A, Tyukavina A, Thau D, Stehman S V, Goetz S J, Loveland T R, Kommareddy A, Egorov A, Chini L, Justice C O, Townshend J R G (2013). High-resolution global maps of 21st-century forest cover change.Science, 342(6160): 850–853

[38] Hansen M C, Potapov P V, Pickens A H, Tyukavina A, Hernandez-Serna A, Zalles V, Turubanova S, Kommareddy I, Stehman S V, Song X P, Kommareddy A (2022). Global land use extent and dispersion within natural land cover using Landsat data.Environ Res Lett, 17(3): 034050

[39] Hansen M C, Townshend J R G, DeFries R S, Carroll M (2005). Estimation of tree cover using MODIS data at global, continental and regional/local scales.Int J Remote Sens, 26(19): 4359–4380

[40] Higginbottom T P, Symeonakis E, Meyer H, van der Linden S (2018). Mapping fractional woody cover in semi-arid savannahs using multi-seasonal composites from Landsat data.ISPRS J Photogramm Remote Sens, 139: 88–102

[41] IPBES (2019). Summary for policymakers of the global assessment report on biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. In: Díaz S, Settele J, Brondízio E S, Ngo H T, Guèze M, Agard J, Arneth A, Balvanera P, Brauman K A, Butchart S H M, Chan K M A, Garibaldi L A, Ichii K, Liu J, Subramanian S M, Midgley G F, Miloslavich P, Molnár Z, Obura D, Pfaff A, Polasky S, Purvis A, Razzaque J, Reyers B, Roy Chowdhury R, Shin Y J, Visseren-Hamakers I J, Willis K J, Zayas C N, eds. IPBES secretariat, Bonn, Germany,1–56

[42] Jennings S, Brown N, Sheil D (1999). Assessing forest canopies and understorey illumination: canopy closure, canopy cover and other measures.Forestry: An International Journal of Forest Research, 72(1): 59–74

[43] Jiang W, Yuan L, Wang W, Cao R, Zhang Y, Shen W (2015). Spatio-temporal analysis of vegetation variation in the Yellow River Basin.Ecol Indic, 51: 117–126

[44] Jones J, Ellison D, Ferraz S, Lara A, Wei X, Zhang Z (2022). Forest restoration and hydrology.For Ecol Manage, 520: 120342

[45] Jung M, Henkel K, Herold M, Churkina G (2006). Exploiting synergies of global land cover products for carbon cycle modeling.Remote Sens Environ, 101(4): 534–553

[46] Karlson M, Ostwald M (2016). Remote sensing of vegetation in the Sudano-Sahelian zone: a literature review from 1975 to 2014.J Arid Environ, 124: 257–269

[47] Kendall M G (1948). Rank correlation methods. Oxford, England: Griffin

[48] Le Houerou H N (1980). The rangelands of the Sahel.J Range Manage, 33(1): 41–46

[49] Li L, Mu X, Jiang H, Chianucci F, Hu R, Song W, Qi J, Liu S, Zhou J, Chen L, Huang H, Yan G (2023). Review of ground and aerial methods for vegetation cover fraction (fCover) and related quantities estimation: definitions, advances, challenges, and future perspectives.ISPRS J Photogramm Remote Sens, 199: 133–156

[50] Li W, Guo W, Qin Y, Wang L, Niu Z, Svenning J C (2021). Mapping spatio-temporal patterns in global tree cover heterogeneity: links with forest degradation and recovery.Int J Appl Earth Obs Geoinf, 104: 102583

[51] Liu B, Yang X, Wang Z, Ding Y, Zhang J, Meng D (2023). A comparison of six forest mapping products in Southeast Asia, aided by field validation data.Remote Sens (Basel), 15(18): 4584

[52] Liu Y, Liu R (2020). A simple approach for mapping forest cover from time series of satellite data.Remote Sens (Basel), 12(18): 2918

[53] Liu Y, Liu R, Qi L, Chen J, Dong J, Wei X (2024). Global mapping of fractional tree cover for forest cover change analysis.ISPRS J Photogramm Remote Sens, 211: 67–82

[54] Machado-Silva F, Neres-Lima V, Oliveira A F, Moulton T P (2022). Forest cover controls the nitrogen and carbon stable isotopes of rivers.Sci Total Environ, 817: 152784

[55] Maestre F T, Salguero-Gómez R, Quero J L (2012). It is getting hotter in here: determining and projecting the impacts of global environmental change on drylands.Philos Trans R Soc Lond B Biol Sci, 367(1606): 3062–3075

[56] Majasalmi T, Rautiainen M (2021). Representation of tree cover in global land cover products: Finland as a case study area.Environ Monit Assess, 193(3): 121

[57] Mann H B (1945). Nonparametric tests against trend.Econometrica, 13(3): 245–259

[58] Mansuy N, Barredo J I, Migliavacca M, Pilli R, Leverkus A B, Janouskova K, Mubareka S (2024). Reconciling the different uses and values of deadwood in the European Green Deal.One Earth, 7(9): 1542–1558

[59] Mbow C (2017). The Great Green Wall in the Sahel. Oxford: Oxford University Press

[60] Melo J, Baker T, Nemitz D, Quegan S, Ziv G (2023). Satellite-based global maps are rarely used in forest reference levels submitted to the UNFCCC.Environ Res Lett, 18(3): 034021

[61] Miller D C, Muñoz-Mora J C, Christiaensen L (2017). Prevalence, economic contribution, and determinants of trees on farms across Sub-Saharan Africa.For Policy Econ, 84: 47–61

[62] Mirzabaev A, Sacande M, Motlagh F, Shyrokaya A, Martucci A (2022). Economic efficiency and targeting of the African Great Green Wall.Nat Sustain, 5(1): 17–25

[63] Nesha K, Herold M, De Sy V, Duchelle A E, Martius C, Branthomme A, Garzuglia M, Jonsson O, Pekkarinen A (2021). An assessment of data sources, data quality and changes in national forest monitoring capacities in the Global Forest Resources Assessment 2005–2020.Environ Res Lett, 16(5): 054029

[64] Nguyen H M, Ouillon S, Vu V D (2022). Sea level variation and trend analysis by comparing Mann–Kendall test and innovative trend analysis in front of the Red River Delta, Vietnam (1961–2020).Water, 14(11): 1709

[65] Nicholson S E, Fink A H, Funk C (2018). Assessing recovery and change in West Africa’s rainfall regime from a 161-year record.Int J Climatol, 38(10): 3770–3786

[66] Nicholson S E, Tucker C J, Ba M B (1998). Desertification, drought, and surface vegetation: an example from the West African Sahel.Bull Am Meteorol Soc, 79(5): 815–830

[67] Ochieng R M, Visseren-Hamakers I J, Arts B, Brockhaus M, Herold M (2016). Institutional effectiveness of REDD+ MRV: countries progress in implementing technical guidelines and good governance requirements.Environ Sci Policy, 61: 42–52

[68] Oldekop J A, Rasmussen L V, Agrawal A, Bebbington A J, Meyfroidt P, Bengston D N, Blackman A, Brooks S, Davidson-Hunt I, Davies P, Dinsi S C, Fontana L B, Gumucio T, Kumar C, Kumar K, Moran D, Mwampamba T H, Nasi R, Nilsson M, Pinedo-Vasquez M A, Rhemtulla J M, Sutherland W J, Watkins C, Wilson S J (2020). Forest-linked livelihoods in a globalized world.Nat Plants, 6(12): 1400–1407

[69] Patrick G (2001). Desertification and a shift of forest species in the West African Sahel.Clim Res, 17(2): 217–228

[70] Potapov P, Li X, Hernandez-Serna A, Tyukavina A, Hansen M C, Kommareddy A, Pickens A, Turubanova S, Tang H, Silva C E, Armston J, Dubayah R, Blair J B, Hofton M (2021). Mapping global forest canopy height through integration of GEDI and Landsat data.Remote Sens Environ, 253: 112165

[71] Qin Y, Xiao X, Dong J, Zhou Y, Wang J, Doughty R B, Chen Y, Zou Z, Moore B III (2017). Annual dynamics of forest areas in South America during 2007–2010 at 50-m spatial resolution.Remote Sens Environ, 201: 73–87

[72] Reiner F, Brandt M, Tong X, Skole D, Kariryaa A, Ciais P, Davies A, Hiernaux P, Chave J, Mugabowindekwe M, Igel C, Oehmcke S, Gieseke F, Li S, Liu S, Saatchi S, Boucher P, Singh J, Taugourdeau S, Dendoncker M, Song X P, Mertz O, Tucker C J, Fensholt R (2023). More than one quarter of Africa’s tree cover is found outside areas previously classified as forest.Nat Commun, 14(1): 2258

[73] Sano E E, Rizzoli P, Koyama C N, Watanabe M, Adami M, Shimabukuro Y E, Bayma G, Freitas D M (2021). Comparative analysis of the global forest/non-forest maps derived from SAR and optical sensors. Case studies from Brazilian Amazon and Cerrado biomes.Remote Sens (Basel), 13(3): 367

[74] Sen P K (1968). Estimates of the regression coefficient based on Kendall's tau.J Am Stat Assoc, 63(324): 1379–1389

[75] Sexton J O, Noojipady P, Song X P, Feng M, Song D X, Kim D H, Anand A, Huang C, Channan S, Pimm S L, Townshend J R (2016). Conservation policy and the measurement of forests.Nat Clim Chang, 6(2): 192–196

[76] Shimada M, Itoh T, Motooka T, Watanabe M, Shiraishi T, Thapa R, Lucas R (2014). New global forest/non-forest maps from ALOS PALSAR data (2007–2010).Remote Sens Environ, 155: 13–31

[77] Souverijns N, Buchhorn M, Horion S, Fensholt R, Verbeeck H, Verbesselt J, Herold M, Tsendbazar N-E, Bernardino P N, Somers B, Van De Kerchove R (2020). Thirty years of land cover and fraction cover changes over the Sudano-Sahel using Landsat time series.Remote Sens (Basel), 12(22): 3817

[78] Staver A C, Hansen M C (2015). Analysis of stable states in global savannas: is the CART pulling the horse? - A comment.Glob Ecol Biogeogr, 24(8): 985–987

[79] Thapa R B, Itoh T, Shimada M, Watanabe M, Takeshi M, Shiraishi T (2014). Evaluation of ALOS PALSAR sensitivity for characterizing natural forest cover in wider tropical areas.Remote Sens Environ, 155: 32–41

[80] Tropek R, Sedláček O, Beck J, Keil P, Musilová Z, Šímová I, Storch D (2014). Comment on “High-resolution global maps of 21st-century forest cover change”.Science, 344(6187): 981

[81] Tucker C, Brandt M, Hiernaux P, Kariryaa A, Rasmussen K, Small J, Igel C, Reiner F, Melocik K, Meyer J, Sinno S, Romero E, Glennie E, Fitts Y, Morin A, Pinzon J, McClain D, Morin P, Porter C, Loeffler S, Kergoat L, Issoufou B A, Savadogo P, Wigneron J P, Poulter B, Ciais P, Kaufmann R, Myneni R, Saatchi S, Fensholt R (2023). Sub-continental-scale carbon stocks of individual trees in African drylands.Nature, 615(7950): 80–86

[82] Turner M D, Carney T, Lawler L, Reynolds J, Kelly L, Teague M S, Brottem L (2021). Environmental rehabilitation and the vulnerability of the poor: the case of the Great Green Wall.Land Use Policy, 111: 105750

[83] UNCCD (2020). The Great Green Wall implementation status and way ahead to 2030 advanced version. Bonn, Germany

[84] Van Zyl J J (1993). The effect of topography on radar scattering from vegetated areas.IEEE Trans Geosci Remote Sens, 31(1): 153–160

[85] Verhelst K, Gou Y, Herold M, Reiche J (2021). Improving forest baseline maps in tropical wetlands using GEDI-based forest height information and Sentinel-1.Forests, 12(10): 1374

[86] Walther B A (2016). A review of recent ecological changes in the Sahel, with particular reference to land-use change, plants, birds and mammals.Afr J Ecol, 54(3): 268–280

[87] Wang H, Cai L, Wen X, Fan D, Wang Y (2022). Land cover change and multiple remotely sensed datasets consistency in China.Ecosyst Health Sustain, 8(1): 2040385

[88] Wei X, Liu Y, Qi L, Chen J, Wang G, Zhang L, Liu R (2023). Monitoring forest dynamics in Africa during 2000–2020 using a remotely sensed fractional tree cover dataset.Int J Digit Earth, 16(1): 2212–2232

[89] White F (1978). The Afromontane Region. In: Werger M J A, ed. Biogeography and Ecology of Southern Africa. Dordrecht: Springer Netherlands, 463–513

[90] Yang F, Jiang X, Ziegler A D, Estes L D, Wu J, Chen A, Ciais P, Wu J, Zeng Z (2023). Improved fine-scale tropical forest cover mapping for Southeast Asia using Planet-NICFI and Sentinel-1 imagery.Journal of Remote Sensing, 3: 0064

[91] Yang X, Crews K (2019). Applicability analysis of MODIS tree cover product in Texas savanna.Int J Appl Earth Obs Geoinf, 81: 186–194

[92] Yang Y, Xiao P, Feng X, Li H (2017). Accuracy assessment of seven global land cover datasets over China.ISPRS J Photogramm Remote Sens, 125: 156–173

[93] Zanaga D, Van De Kerchove R, De Keersmaecker W, Souverijns N, Brockmann C, Quast R, Wevers J, Grosu A, Paccini A, Vergnaud S, Cartus O, Santoro M, Fritz S, Georgieva I, Lesiv M, Carter S, Herold M, Li L, Tsendbazar N.E, Ramoino F, Arino O (2021). ESA WorldCover 10 m 2020 v100

[94] Zhang W, Brandt M, Wang Q, Prishchepov A V, Tucker C J, Li Y, Lyu H, Fensholt R (2019). From woody cover to woody canopies: How Sentinel-1 and Sentinel-2 data advance the mapping of woody plants in savannas.Remote Sens Environ, 234: 111465

RIGHTS & PERMISSIONS

Higher Education Press