Project Topics
Get Complete Project Material File(s) Now! »
Dynamics of forest-savanna ecotone in Central Africa
Several authors have described a widespread woody encroachment into savannas in Central Africa (Fig. 7 and Fig. 8; Boulvert, 1990; Youta and Bonvallot, 1996; Youta, 1998; Youta et al., 2003; Favier et al., 2004b; Mitchard et al., 2009; Mitchard and Flintrop, 2013; Cuni-Sanchez et al., 2016; Aleman et al., 2017; Devine et al., 2017; Axelsson and Hanan, 2018; Deklerck et al., 2019) with significant impacts on the global carbon budget of the continent (Poulter et al., 2014). Boulvert (1990) described a woody expansion occurring in the forest-savanna ecotones of Central Africa Republic and attributed it to be a consequence of the urbanization of the population. The example of the Bambari region shows that with the increase of the Fulani shepherd community and their herds, the fire regime has been totally modified. In the absence of high intensity bushfires trees took advantage over herbaceous plants, rendering pastures unsuitable and causing the migration of many herders to other areas(Gautier et al., 2005) especially towards humid savannas (Boutrais and Jean, 1990). While studying the dynamics of about 3000 years old isolated savannas enclosed by forest in the eastern part of the Congolese They showed that forest was spreading over savanna at a rate between 14 and 75 m per century which suggested that enclosed savannas could completely disappear in the Mayombe (Democratic Republic of Congo) in about 1000-2000 years. In the same logic Favier et al. (2004b) described two main modes of forest progression into savanna: a linear forest edge progression which is slow (c. 1 m year-1) and a forest clump coalescence with an increase in the man-made fire frequency that can lead to a shift from the coalescence regime to the edge progression one. Axelsson et al. (2018) estimated the mean annual change of woody cover across Sub-Saharan African savannas to be 0.25% per year which was negatively correlated with fire frequency.
When focusing on structural changes that occur in the vegetation structure during forest expansion after fire exclusion, Deklerck et al. (2019) estimated a forest species’ specialist encroachment rate of 9 stems.ha-1.y-1 and a savanna specialist disappearance rate of 16 stems stems.ha-1.y-1 in savanna patches of the Mayombe hills. Average diameter of forest specialists did not change due to an increasing influx of recruits, while average diameter of savanna trees increased due to decreasing recruitment. Carbon stored by forest specialists increased from 3.12 to 5.60 MgC.ha-1, suggesting a forest carbon recovery rate of 0.62 MgC.ha-1.yr-1. They estimated at least 150 years as the time required for a total forest recovery over savanna after fire exclusion. Jeffery et al. (2014) reported in the Lopé National Park in Gabon that savannas can sufficiently thicken up over a 15 year period to reach a structure comparable to a colonising forest when protected from fires. Later Cardoso et al., (2020) evidenced the presence of an ecotonal community in the Loppé National Park that occupies a narrow belt between savanna and forest and stabilises the forest-savanna mosaic even when the savanna is burned regularly.
In Cameroon, Youta (1998) observed the encroachment of gallery forest into the surrounding savannas at a rate of 0.6-2 m.yr-1 between 1950 and 1990 in the Central Region. However the encroachment rate was slower in savannas neighbouring forest established on waterlogged soils compared to savannas that were close to forest found on well-drained ferrallitic soils dominated by Malvaceae and Ulmaceae. Youta (1998) also hypothesized that human presence in Guinean savannahs reduced fire occurrence which favoured the establishment of forest species and later contributed into the formation of forest patches within savannas. Mitchard et al. (2011, 2009, 2013) described a rapid woody encroachment of savannas in central Cameroon (Mbam et Djerem National Park) where forest edges were dominated by young pioneer trees, with dead and dying savanna trees prevalent, which is strong evidence that this constituted young encroaching forests. They hypothesize one of the causes to be either recent reduction in fire frequency due to a reduction in human pressure caused by urbanization, as rainfall did not alter significantly over the study period. Their alternative hypothesis was that increasing atmospheric CO2 concentrations were altering the competitive balance between grasses and trees.
Remote sensing-based modelling of vegetation structure and dynamics in forest-savanna transitional areas
Conventionally, vegetation structure have been assessed using field-based inventory plots (Jayakumar et al., 2011; Higgins and Scheiter, 2012; Arellano et al., 2016). Field inventories are expensive, time consuming, labour intensive and they do not integrate the spatial heterogeneity of the vegetation structure within the landscape especially for remote areas with limited access (Lu, 2006; Maniatis et al., 2011; Clark and Louis, 2012; Réjou-méchain et al., 2019). In Central Africa historical field inventory data of forest-savanna ecotone over several decades are especially rare (Mitchard et al., 2013). Therefore it is difficult to use field plots alone in assessing natural and anthropogenic induced variation in vegetation structure and biomass over large areas. Given the extend of tropical ecosystems, access limitations and structural complexity, Remote Sensing (RS) methods have since long been used in assessing and characterize tropical ecosystems (St-Onge and Cavayas, 1997; Youta et al., 2003). Moreover as new technologies emerged, RS provides a systematic and synoptic view of earth cover for changes in land cover and to reveal aspect of biological diversity directly. Satellite data are important tools in the interdisciplinary study of tropical forests that are increasingly integrated into studies that monitor changes in vegetation cover within tropical forests-savanna ecotones and also applied with other types of data (i.e. geographical, topography, hydrology) to investigate the drivers of land cover changes.
Laser scanning methods such as LiDAR (Light Detection and Ranging; Fig. 9) have also emerged as a promising technology for estimating forest height, volume and AGB in boreal, temperate and tropical forests (Drake et al., 2002; Clark et al., 2004). Able to provide direct descriptors of forest structure including tree height, crown size, and tree density (Heurich and Thoma, 2008; Bergen et al., 2009), LiDAR sensors are of particular interest for the estimation of forest biomass and carbon stocks (Corona and Fattorini, 2008; Steinmann et al., 2013). Airborne LiDAR are essential assets in the sense that they supply very high spatial and geometrical resolution (centimetre resolution) data on the tri-dimensional organization of the cover of areas that are too large or inaccessible. Recent studies proved the potentiality of very high resolution LiDAR images in accurate description of tropical forest cover, heterogeneity, cover change and biomass assessment (Vincent et al., 2010).
Mapping species diversity and vegetation types
Optical sensors with a series of contiguous bands covering narrow spectral ranges allows to record information related to a range of plant properties such as the age, the water content, the leaf pigment and the chemical composition (Curran, 1989; Martin and Aber, 1997; Townsend et al., 2008). Recent studies in tropical forests showed that tree species often have unique spectral signatures based on their structural and biochemical properties (Colgan et al., 2012b; Rocchini et al., 2016). Therefore the variability in high spatial resolution multispectral information can be used to differentiate species or group of species at a landscape scale, even in complex tropical ecosystems based on the optical traits corresponding to the reflectance of each pixel (Clark et al., 2005; Ustin and Gamon, 2010; Mbobda et al., 2018; Neba et al., 2020). However the spectral species described from multispectral data do not directly refer to field-based species diversity but rather, the spectral species distribution within a specific area can approximate its biological species richness (Féret and Asner, 2014) according to the spectral variation hypothesis (SVH). The SVH postulates that the spectral variation of a site can be related to its ecosystem heterogeneity (Palmer et al., 2002) since greater heterogeneity allows a higher number of species to coexist (Wilson, 2000; Huggett, 2002). Féret and Boissieu (2020) generated α‐ and β‐diversity indicators from Sentinel 2 optical imagery, based on SVH (Fig. 13).
Airborne LiDAR sampling
Light Detection And Ranging (LiDAR) is a remote sensing system that uses lasers to measure distances between a sensor and targets of interest (Fig. 21). LiDAR instrument consists of a high frequency laser that has the ability to emit tens of thousands of laser pulses per second, from which corresponding range measurements can be derived. Rate of laser pulse emission is quantified and referred to as the “pulse repetition frequency”. Pulses are delivered by scanning, so that pulses are emitted within a set number of degrees from nadir; each laser pulse and associated returned energy has an associated “scan angle”, which describes the angle from nadir of the pulse. LiDAR sensors for earth applications are frequently flown aboard fixed-wing aircraft at relatively low elevations (~1000 m above ground level when mounted on aircrafts), and unmanned aerial vehicles (UAVs) are becoming more popular as LiDAR platforms (generally flown at elevations < 150 m). Because LiDAR systems use lasers to interact with targets, LiDAR is considered a form of active remote sensing, as opposed to passive remote sensing, such as sensors aboard satellites that passively measure reflected sunlight (Klauberg, 2018).
Land cover mapping
Land cover mapping remains a difficult task and it is especially challenging in heterogeneous landscapes such as forest-savanna transitions. In addition to its location within a forest-savanna transitional zone, the anthropogenic pressure over the Nachtigal area results into a complex combination of different landforms such as urban areas, agricultural fields, pasture/scrublands, bare areas, agroforest, forest and natural areas, water surfaces (Alexandre, 2013) to name but a few. One of the main issues when generating land cover maps from such complex areas is the confusion of spectral responses from different features as it can be seen in Table III. The classification accuracy from satellite images therefore depends upon both the quality of the sensor and the classification method used (Poursanidis et al., 2015; Ustuner et al., 2015).
In this study Spot 6/7 image was used to map land cover types over the Nachtigal area as it has already proven its potentiality in mapping heterogeneous landscapes with good accuracy (Kuzucu et al., 2017). Properties of the Spot 6/7 image were given in Table II. Additional vegetation indices i.e. the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation index (EVI) were computed and added as additional bands to enhance the spectral information and increase the spectral separability between classes.
Image classification
A supervised image classification was used to distinguish between the different land cover types as it usually warrants higher quality of the final mapping product (Khatami, Mountrakis, and Stehman, 2016). Supervised classification requires labelled training data for each land cover type to establish the statistics to identify spectral classes (or clusters) in a multiband image. All classes have to be derived usually through a training stage with the use of training samples. Training samples for each land cover type was therefore manually delineated based on (i) the knowledge of the land cover type from the field; (ii) their spectral responses from Spot 6/7 satellite image and (iii) the height distribution of their canopies as obtained through the CHM (Table III). A total of approximately 2640 pixels for the different land-cover types was generated. The image was classified into 13 land cover types using the maximum likelihood algorithm from the ENVI 5.0 image classification software. Finally, a 3 x 3 majority filter was applied to each classification to recode isolated pixels classified differently than the majority class of the window.
Table of contents :
CHAPTER I. GENERALITIES
I.1. INTRODUCTION
I.1.1 Context and justification
I.1.2 General and specific objectives
I.1.1.1. General objective
I.1.1.2. Specific objectives
I.1.3 Research hypothesis and questions
I.2. LITERATURE REVIEW
I.2.1. Forest-savanna ecotone in tropical areas: definitions and concepts
I.2.2. Major drivers and processes shaping the distribution of forest-savanna vegetation
I.2.2.1. Role of climate
I.2.2.2. Role of fire
I.2.2.3. Role of herbivory
I.2.2.4. Role of soil and topography
I.2.3. Dynamics of forest-savanna ecotone in Central Africa
I.2.4. Remote sensing-based modelling of vegetation structure and dynamics in forest-savanna transitional areas
I.2.4.1. Mapping aboveground biomass
I.2.4.2. Mapping species diversity and vegetation types
I.2.4.3. Monitoring land cover changes and disturbances
CHAPTER II. MATERIAL AND METHODS
II.1. MATERIAL
II.1.1. Study site
II.1.1.1. Location
II.1.1.2. Vegetation
II.2. METHODS
II.2.1. Data collection
II.2.1.1. Field data collection
II.2.2.2.1. Forest sampling
II.2.2.2.2. Savanna sampling
II.2.2.2.3. Estimating AGB from plot data
II.2.1.2. Spatial data collection
II.2.2.2.1. Airborne LiDAR sampling
II.2.2.2.2. Satellite data sampling
II.2.2. Spatial and statistical analysis
II.2.2.1. Land cover mapping
II.2.2.2.1. Image classification
II.2.2.2.2. Accuracy assessment
II.2.2.2. Mapping functional attributes of the vegetation
II.2.2.2.1. Aboveground biomass
II.2.2.2.2. Spectral floristic assemblages
II.2.3. Monitoring vegetation dynamics in Google Earth Engine
II.2.3.1. Aggregating temporal images for landcover change analysis
II.2.3.2. Mapping forest and savanna using supervised methods
II.2.3.3. Mapping forest and savanna using automated-unsupervised method
II.2.3.4. Generating transition maps
II.2.3.5. Mapping fire frequencies
CHAPTER III. RESULTS AND DISCUSSION
III.1. RESULTS
III.1.1. Distribution of the vegetation types and structure within the study area
III.1.1.1. Spatial distribution of the vegetation types from satellite data
III.1.1.2. AGB estimation from ALS data
III.1.1.3. AGB estimation from optical satellite data
III.1.2. Variation in the vegetation structure after four decades of monitoring
III.1.2.1. Vegetation cover change
III.1.2.2. Functional change
III.1.3. Fire influence on land cover dynamics
III.1.3.1. Influence of fire frequency on forest transition
III.1.3.2. Influence of fire frequency on savanna structure
III.1.4. Species succession dynamics
III.1.4.1. Gradients in floristic composition
III.1.4.2. Species succession
III.2. DISCUSSION
III.2.1. Variation in the vegetation types and structure across the study area
III.2.1.1. Spatial distribution of the vegetation types in the Nachtigal area
III.2.1.2. Landscape-scale AGB estimation
III.2.1.3. Differences in the structure and floristic composition between the Nachtigal area and the Mpem et Djim NP
III.2.2. Vegetation change patterns
III.2.2.1. Long-term (1975-2020) forest expansion
III.2.2.2. Performance of automated cloud computing and Landsat image archives in modelling land cover dynamics
III.2.2.3. Spectral composition structuring along a forest succession
III.2.2.4. AGB recovery along a forest succession
III.2.3. Influence of fire on vegetation dynamics
III.2.3.1. Performance of Landsat data in characterising fire frequency
III.2.3.2. Fire influence on savanna structure and dynamics
III.2.4. Species succession dynamics
III.2.5. Implication for conservation and management
CHAPTER IV. CONCLUSION AND PERSPECTIVES
IV.1. CONCLUSION
IV.2. PERSPECTIVES
REFERENCES
