Reduction of tree cover in West African woodlands and promotion in semi-arid farmlands

Martin Brandt, Kjeld Rasmussen, Pierre Hiernaux, Stefanie Herrmann, Compton J. Tucker, Xiaoye Tong, Feng Tian, Ole Mertz, Laurent Kergoat, Cheikh Mbow, John David, K. A. Melocik, Morgane Dendoncker, Caroline Vincke, Rasmus FensholtNature Geoscience 2018 · DOI

TL;DR

This study leverages high-resolution satellite imagery to map woody cover across West African drylands, revealing that farmlands, contrary to common assumptions, often support significant woody vegetation. The research highlights that while climate is the primary driver of woody cover, human management plays a crucial role, particularly in semi-arid regions, where it can increase tree density.

What Problem Does This Paper Address?

For decades, the narrative surrounding West African drylands has been dominated by concerns of desertification and a decline in woody vegetation, often attributed to agricultural expansion and population growth. This "Sahel syndrome" has driven policy and conservation efforts. However, localized studies have suggested a more nuanced reality, with some areas showing increased woody cover. The challenge has been to move beyond these case studies to a regional-scale understanding of woody cover dynamics. Previous large-scale assessments using lower-resolution satellite data struggled to differentiate between herbaceous and woody vegetation and often overlooked dryland ecosystems, leading to an incomplete picture of tree cover and the factors influencing it. This paper aims to bridge this gap by providing a detailed, wall-to-wall assessment of woody cover across West African drylands and identifying the key drivers, including both environmental factors and human management.

Methods

The study employs a multi-faceted approach combining very high-resolution (VHR) satellite imagery with broader-scale remote sensing data and environmental datasets.

  • High-Resolution Woody Cover Mapping: The core of the methodology involves using VHR satellite imagery (DigitalGlobe QuickBird-2, GeoEye-1, WorldView-2) with resolutions as fine as 0.3m. A modified soil-adjusted vegetation index (MSAVI) was calculated, and a texture-based feature extraction method was used to identify and quantify woody canopy cover at a 1.7m resolution. This was further refined by a weighted approach to account for sub-pixel cover and prevent overestimation.
  • Regional Woody Cover Prediction: To create a wall-to-wall map at 100m resolution, the VHR-derived woody cover data was used to train a gradient boost decision tree (GBR) regressor. This model was then applied to broader-scale data from ALOS PALSAR and PROBA-V NDVI imagery. Phenological differences between evergreen and deciduous species were accounted for by using a ratio of NDVI values from different seasons.
  • Determinant Analysis: A decision tree ensemble model was used to analyze the relationships between predicted woody cover and various environmental and management factors. This model helped to identify the relative importance of factors like rainfall, terrain, soil, human population density, distance to villages, and fire frequency.

Datasets

  • Very High Resolution (VHR) Satellite Imagery: DigitalGlobe QuickBird-2, GeoEye-1, and WorldView-2 (0.3m resolution) were used for training the high-resolution woody cover mapping. These were acquired under the NextView license program.
  • PROBA-V NDVI Data: Used for large-scale woody vegetation assessment at 100m resolution.
  • ALOS PALSAR Data: Also used for large-scale woody vegetation assessment at 100m resolution.
  • Worldpop Population Data: Human population density data at 1km resolution.
  • CHIRPS Rainfall Data: Mean annual rainfall data used for defining rainfall zones and as an explanatory variable.
  • MODIS MCD45A1 Burned Area Product: Used for fire frequency analysis.
  • African Soil Map: Used for soil texture analysis.
  • SRTM Digital Elevation Model: Used for terrain analysis.
  • Globeland30 and ESA LC CCI Land Cover Maps: Used for masking out wetlands and irrigated areas.
  • Field Data: Collected from various sites in Senegal, Mali, and Niger (2000-2015) for validation of the remote sensing-derived woody cover maps.

Key Insights and Ideas

  • Farmlands as Carbon Sinks: Contrary to the assumption that agricultural expansion leads to a net loss of trees, this study reveals that West African farmlands, particularly in semi-arid zones, often maintain or even increase woody cover compared to adjacent savannas.
  • Human Management's Nuanced Role: While climate (rainfall, terrain, soil) is the dominant driver of woody cover (explaining ~80%), human management is still significant (~20%). Importantly, in semi-arid regions, management can actively promote woody cover, a finding that challenges the prevailing "Sahel syndrome" narrative.
  • Village Proximity Matters: Woody cover is densest within and around villages, suggesting that farmers actively protect or plant trees for various benefits, rather than solely exploiting them. This effect is more pronounced in semi-arid areas.
  • Dryland Woody Cover is Underestimated: The study highlights that previous global tree cover assessments often focus on forests in humid areas and underestimate woody cover in drylands, leading to a skewed understanding of global tree cover dynamics.

Key Results

  • Regional Woody Cover Patterns: Woody cover averages 3% in arid zones, 9% in semi-arid zones, and over 20% in sub-humid zones.
  • Farmland vs. Savanna Woody Cover: In semi-arid Sahel, farmland management increases woody cover to 12%, compared to 6% in savannas. In sub-humid zones, farmlands have reduced woody cover (20%) compared to savannas (30%).
  • Determinants of Woody Cover: The decision tree model explained 67% of the predicted woody cover. The most important factors were mean annual rainfall (32%), terrain (23%), and human population density (13%). Climate and edaphic factors together accounted for 79% of the variation, while management factors accounted for 21%.
  • Population Density and Woody Cover: In arid and semi-arid regions, woody cover generally increases with population density up to a certain threshold, contradicting the idea that high population pressure always leads to tree loss. However, in sub-humid zones, higher population densities are associated with lower woody cover.
  • Conservation Areas: Woody cover is generally higher in conservation areas (29%) compared to surrounding areas (21%).

Limitations and Future Work

  • Short Time Series for Trend Analysis: The availability of VHR satellite data is relatively recent, limiting the ability to estimate long-term trends in woody cover. The study focuses on spatial correlations.
  • Definition of Woody Cover: The study defines woody cover as projected canopy coverage. For some applications, additional variables like species composition, carbon stocks, woody volume, or fodder production would be beneficial.
  • Future Work: The authors suggest that future research could exploit VHR imagery to map individual trees, providing even finer detail. Combining wall-to-wall analysis with local process studies is crucial for developing more robust environmental policies. The study's methodology and findings can be applied to other dryland regions globally.

Summary updated

BibTeX
@inproceedings{brandt2018reduction,
  author = {Martin Brandt and Kjeld Rasmussen and Pierre Hiernaux and Stefanie Herrmann and Compton J. Tucker and Xiaoye Tong and Feng Tian and Ole Mertz and Laurent Kergoat and Cheikh Mbow and John David and K. A. Melocik and Morgane Dendoncker and Caroline Vincke and Rasmus Fensholt},
  title = {{Reduction of tree cover in West African woodlands and promotion in semi-arid farmlands}},
  booktitle = {Nature Geoscience},
  year = {2018},
  doi = {10.1038/s41561-018-0092-x},
}
Figure