Lessons from a regional analysis of forest recovery trajectories in West Africa

The sampling covers three phytogeographic areas of Côte d'Ivoire, West Africa (figure 1): (a) evergreen forests are found in the ombrophilous zone; (b) semi-deciduous forests in the mesophilous zone and (c) open dry forests in the Sudanese zone. This gradient corresponds to an annual precipitation gradient that varies between 2500 mm in the south and 1100 mm in the north. The average annual temperature is around 26.5 ^∘C but with much higher daily and seasonal temperature variations in the north than in the south.

We set up eight chronosequences and, in each chronosequence, we established 20 plots of 0.2 ha corresponding to different ages of the plots since their abandonment by the local farmers. Care was taken to ensure that all ages were represented in each chronosequence, with a total of 32 plots per age class (${\lt}11$ years, 11–20 years, 21–30 years, ${\gt}30$ years) and 32 old-growth forests (never logged or burnt in human memory and with no visible signs of disturbance). Botanical names, diameter at breast height (DBH) and height of all stems at least 2.5 cm DBH were measured from October 2016 to October 2020. We chose the threshold of 2.5 cm DBH because young trees are an important component of early successional stages (Swaine and Hall 1983).

We collected a set of environmental variables (table 1) to characterise (a) local variability related to the plot itself (soil hydromorphy, presence of remnant trees), its history (type and duration of previous crop) and its landscape context (forest proximity, connectivity) and (b) regional variability of climate (temperature, precipitation) and soils (density, texture).

• The local variables were collected in situ, i.e. in the field. A survey, prior to the choice of the plot, with the owners (current and former) of the abandoned fields and with Sodefor, the forest service in charge of forest management and protection in Côte d'Ivoire, provided information on the management history of the plot (type and duration of past cultivation) and the status of each tree (remnant or not), as well as the exact age of the secondary forest plots. The landscape context (density of old growth forest, distance to old growth forest) was recorded in situ, details in table 1. Local soil hydromorphy was estimated in the field, i.e. it was generally low-lying areas, plots located near river banks or plots located in areas where the micro topography makes them temporarily retain water.
• The regional variables were acquired in silico, i.e. through the climatic and soil databases available online. The climate variables were extracted from the Worldclim 2 database with a resolution of 10 min (Fick and Hijmans 2017). The intermediate resolution of 10 min was chosen in order to capture the broad regional climatic gradients and not capture particular microclimates. Soil variables were extracted from the SoilGrids version 2.0 databases at a depth of 0–5 cm (Poggio et al 2021), a depth that is crucial for understanding the influence of the topsoil on the early stages of successional dynamics.

Table 1. table of regional and local environmental variables used to predict recovery trajectories in West African Secondary Forests.

Variable	Description	Units	Source
Regional covariates
Annual mean temperature	Average temperature for each month	0.1 ∘C	http:worldclim.org
Temperature annual range	Difference between the Max of the warmest and the Min of the coldest month	0.1 ∘C	http:worldclim.org
Temperature seasonality	Ratio of the standard deviation to the mean monthly temperature	$\%$	http:worldclim.org
Annual precipitation	Total precipitation for each month	mm	http:worldclim.org
Precipitation seasonality	Ratio of the standard deviation to the mean monthly total precipitation	$\%$	http:worldclim.org
Bulk density	Bulk density of the fine earth fraction	kg dm−3	http:soilgrids.org
Coarse content	Volumetric fraction of coarse fragments	$\%$	http:soilgrids.org
Clay content	Proportion of clay particles (< 0.002 mm) in the fine earth fraction	$\%$	http:soilgrids.org
Sand	Proportion of sand particles (> 0.05 mm) in the fine earth fraction	$\%$	http:soilgrids.org
Local covariates
Hydromorphy	Ferralithic (0) or hydromorph (1) soils	/	Field-measured
Remnants	Number of left-alive isolated trees from the former old-growth forest	trees ha−1	Field-measured
Density	Presence of old-growth forest within a 500 m distance on the eight cardinal points	frequency	Field-measured
Proximity	1/Distance to the nearest old-growth forest fragment	1 m−1	Field-measured
Crop	Annual (0) or perennial (1) crops	/	Field-measured
Duration	Duration of the previous crop grown	Year	Field-measured

All calculations have been made without remnant trees, which are the legacy of old-growth forests, i.e. trees that were present before cultivation was established.

2.2.1. Aboveground biomass

2.2.2. Diversity

2.2.3. Composition

2.3.1. Modelling recovery trajectories

We parameterised an ecosystem attribute recovery model that has already been successfully tested to predict the recovery trajectories of biomass (N'Guessan et al 2019), biodiversity (Amani et al 2021) and timber volumes (Doua-Bi et al 2021):

$$\begin{align} X{s,t} &\sim \mathrm{log}{\,}\mathcal{N} \left(\theta_{0,s}^{X}+\left(\theta_{+\infty,s}^{X}-\theta_{0,s}^{X}\right)\right.\nonumber\\&\left.\quad\qquad\quad \times \left(1-\mathrm{e}^{-\kappa_{p}^{X} (t/\alpha^{X})^2}\right), \sigma_{X}^2\right) \end{align} \tag{ 1 }$$

$$\begin{align} \theta_{0,s} \sim \mathrm{log}\,\mathcal{N} \left(\mathrm{log}\theta_{0,z}^{X}, \sigma_{\theta_{0,X}}^2\right) \end{align} \tag{ 2 }$$

$$\begin{align} \theta_{+\infty,s} \sim \mathrm{log}\,\mathcal{N} \left(\mathrm{log}\theta_{+\infty,z}^{X}, \sigma_{\theta_{+\infty,X}}^2\right) \end{align} \tag{ 3 }$$

with

$X_{s,t}$: the value of the attribute X (biomass, diversity or composition) of the plot from chronosequence s of age t

$\theta_{0,s}^X$: the starting point of the attribute X when the recovery starts, in chronosequence s

$\theta_{+\infty,s}^X$: the ending point of the fully recovered attribute X in chronosequence s, namely the asymptotic value of the attribute X on time

α^X : a parameter link to the inflexion point (before, the recovery rate increases with age and after, it decreases) of attribute X

$\sigma_{X}^2$: the variance of the lognormal law used to infer attribute X.

Starting points and ending points were both encapsulated into two lognormal hyperlaws centred around different values for each bioclimatic area, namely:

$\theta_{0,z}^X$: the starting point of the attribute X when the recovery starts, in bioclimatic area z

$\theta_{+\infty,z}^X$: the ending point of the fully recovered attribute X in bioclimatic area z

$\sigma_{\theta_{0,X}}^2$: the variance of the hyperlaw over starting points for attribute X

$\sigma_{\theta_{+\infty,X}}^2$: the variance of the hyperlaw over ending points for attribute X

$\kappa^{X} = e^{(\lambda^X)}$, the intrinsic recovery rate of attribute X, this rate is exponentially-transformed to always be defined on $\mathbb{R}^+$.

In order to infer the model and determine the a posteriori distribution of all above-mentioned model parameters, we chose to use a lognormal likelihood law because $X_{s,t}$ is positive and defined on $\mathbb{R}^+$.

2.3.2. Effects of the local and regional variables

To test the effect of each variable on the rate of recovery κ, we introduce them into the model at the plot level p through this equation:

$$\begin{align} \kappa_{p}^{X} = \mathrm{e}^{\left(\lambda^X+\sum\limits _{i = 1}^{i = n}\beta _{\mathrm{Cov}_{i}}^{X}\times \mathrm{Cov}_{i,p}\right)} \end{align} \tag{ 4 }$$

where $\beta_{\mathrm{Cov}_{i}}^X$, is the effect of variable i on the intrinsic recovery rates of attribute X and $\mathrm{Cov}_{i,p}$ is the standardized value of variable i in plot p.

In order to obtain a model that is as parsimonious as possible and not over-parameterised, we proceeded to a step-by-step forward variable selection stage using the Akaike information criterion (AIC) in a Bayesian framework. All variables were standardised beforehand in order to be able to directly compare their respective effects on the intrinsic recovery rates.

The model was written in Stan (Carpenter et al 2017) and then inferred in a Bayesian framework using the Hamiltonian Monte Carlo algorithm. The priors chosen for the parameters are uninformative, as follows

$$\begin{align} \theta_{+\infty}^{X}\sim \Gamma \left( 0.001 \mathrm{,}{\,}0.001 \right) \end{align} \tag{ 5 }$$

$$\begin{align} \theta_{0}^{X} \sim \Gamma \left( 0.001 \mathrm{,\,} 0.001 \right) \end{align} \tag{ 6 }$$

$$\begin{align} \alpha^{X} \sim \Gamma \left( 0.001 \mathrm{,\,} 0.001 \right) \end{align} \tag{ 7 }$$

$$\begin{align} \sigma_{X}\sim \mathcal{N} \left( 0 \mathrm{,\,} 1 \right) \end{align} \tag{ 8 }$$

$$\begin{align} \beta _{\mathrm{cov}_i}^{X} \sim \Gamma \left( 0.001 \mathrm{,\,} 0.001 \right).\end{align} \tag{ 9 }$$

2.3.3. Relative importance of local vs regional variables

On average, diversity has the highest rate of recovery, followed by composition and then biomass (figure 3). In terms of recovery time, the restoration of 50% of the value of the three attributes studied is reached in less than 35 years at the regional scale (table 2). In contrast, it will take more than 60 years to recover more than 90% of the initial biomass and composition in secondary forests (table 2 and figure 3).

Zoom In Zoom Out Reset image size

Table 2. Values (medians and quantiles) of the parameters of the recovery trajectories' models for the three explored forest attributes (biomass, composition, and diversity). θ₀ is the starting point of the trajectories of recovery and $\theta_{+\infty}$ the final point of the attribute of each site, κ indicates the instantaneous rate of recovery, σ the model error, time (50%) is the recovery time of 50% of the attribute and time (90%) is the recovery time of 90% of the attribute.

Parameters	Biomass	Composition	Diversity
θ0	0 [0 ; 0]	0.16 [0.08 ; 0.31]	3.34 [2.31 ; 4.32]
$\theta_{+\infty}$	163.37 [110.08 ; 240.18]	0.63 [0.57 ; 0.70]	21.79 [17.46 ; 28.27]
κ	0.24 [0.02 ; 0.47]	0.26 [0.03 ; 0.47]	0.36 [0.11 ; 0.48]
α	23.29 [11.27 ; 30.24]	22.97 [10.45 ; 32.01]	6.04 [2.76 ; 8.90]
Time $(50\%)$	34 [31 ; 39]	34 [28 ; 42]	9 [7 ; 12]
Time $(90\%)$	63 [56 ; 71]	62 [52 ; 77]	16 [12 ; 21]

Of all the covariates (15 in total, see table 1) that were used in the selection procedure, 10 were actually retained in at least one of the three models (diversity, composition, biomass) and thus had effects on the corresponding recovery rates. Biomass and diversity were influenced by a large number of covariates (7 each, see figure 4) compared to composition (only 5).

Zoom In Zoom Out Reset image size

3.2.1. Regional covariates

3.2.2. Local covariates

Overall, the general signal is that variability in local conditions (figure 5(A)) as well as variability in regional conditions (figure 5(B)) lead to significant changes in recovery trajectories. Beyond this general signal, we note that composition is slightly more influenced by regional covariates and diversity is slightly more influenced by local covariates. When local conditions are favourable (figure 5(A), green curves), diversity is the attribute that recovers the fastest, followed by composition and biomass. When regional conditions are favourable, composition recovers first (figure 5(B), green curves), followed by diversity and biomass. Nearly 100% of diversity and composition are recovered in less than 25 years when both local and regional conditions are favourable. In contrast, under unfavourable regional conditions, composition hardly recovers at all over the time scale studied.

Zoom In Zoom Out Reset image size

Estimated recovery times vary considerably between attributes (figure 3(A)), a result observed in most secondary forests worldwide (Poorter et al 2021a). The recovery of 90% of diversity in less than 20 years (figure 3(B)) shows the very high resilience of West African forest biodiversity. Similar results have already been observed in neotropical forests where only 20 years were sufficient for secondary forests to recover 90% of their species richness and diversity (Lennox et al 2018, Barros et al 2020). The fast recovering of diversity may be linked to the intermediate disturbance hypothesis where biodiversity quickly peaks in recovering forests because of the co-occurrence of persisting pioneer species and shade-tolerant species that established in the shade of these pioneers (Rozendaal et al 2019).

In contrast, secondary forests take, on average, more than six decades to recover 90% of the composition of old-growth forests in our study. This may be explained by the high similarity index between old-growth forests and can be explained by the late arrival of understory forest species in secondary forests that remain thus highly dynamic in composition for a long time. These results corroborate observations made in the central Congo basin where the convergence of the floristic composition of young forests towards the community composition of old-growth forests was slower as compared to diversity (Makelele et al 2021) and echo work done in the neotropics (Rozendaal et al 2019) and in Southeast Asia (Deng et al 2018) where the return of old-growth forest species can take hundreds of years, mainly because of their poor dispersal abilities, especially since a number of these species are dispersed by large animals that also suffered during the time the area was deforested for agriculture. For biomass, it also takes several decades for secondary forests to recover 90% of the biomass of old-growth forests. This long time lag could be explained by the fact that young secondary forests in West Africa have low biomass productivity (N'Guessan et al 2019), and because the regeneration of very large emergent trees, typical of West African landscapes, highly depends on the presence, in the plots, of remnant trees that provide seeds typical late-successional species (Doua-Bi et al 2021). However, these results contrast somewhat with those of the Congo Basin where biomass recovery is much slower, with only 32% recovered by secondary forests in 60 years (Makelele et al 2021), a discrepancy that can be linked to the greater biomass in the old-growth forests of Central Africa (Avitabile et al 2016).

By far the most important variable at the regional scale is the precipitation seasonality, which has a very marked positive effect on the recovery rates of the three attributes studied (figure 4). This precipitation seasonality gradient corresponds, in West Africa, to a South-North gradient with, in the South, evergreen forests where precipitation is regularly distributed over the year and, in the North, open and dry forests where precipitation is concentrated over a few months (Feng et al 2013). It has already been noted several times that recovery trajectories are totally different in tropical rainforests and dry forests, linked to a particular flora and more drastic biophysical constraints in dry forest (Poorter et al 2019). However, it has often been argued that total precipitation and thus water availability positively influence recovery rates (Becknell et al 2012, Rozendaal and Chazdon 2015, Poorter et al 2016). This is not the case in our study where biomass, diversity and composition recover faster in areas with high precipitation seasonality (figure 4) and different, non-exclusive explanations can be mobilised. (a) In areas with high seasonality, herbaceous species are mainly annual and only live for a few months so that the soil is completely bare at the end of the long dry season (Hérault and Hiernaux 2004), which offers the opportunity for woody species to germinate more easily every year when the rains return. In contrast, in low-seasonality areas, perennial herbaceous plants quickly colonise the abandoned field and make it more difficult for tree seeds to germinate. (b) In low-seasonality evergreen forests, the early part of the succession process is dominated by heliophilous woody species (Herault et al 2010, Hérault et al 2011, Mirabel et al 2019) or pioneer lianas that transiently prevent the germination of late-successional forest species (Finegan 1996, Hérault and Piponiot 2018). These cohorts of pioneer species are absent from dry forests and succession begins at abandonment with slow-growing, hardwood species (Poorter et al 2019), most of which are found at the end of succession (figure 2). (c) In terms of structure and biodiversity, West African dry forests are simpler than their evergreen or semi-deciduous counterparts, so their reconstitution may be less complex.

Two additional variables appear to positively influence recovery rates: temperature annual range and soil bulk density. Concerning temperature, as with our results for precipitation, the mean annual temperature had no effect on the recovery rates of the three attributes evaluated (figure 4), whereas its variability through the temperature annual range positively influences the rate of composition recovery. Regarding bulk density, our results may seem contrary to those suggesting that soils that are too compact prevent good vegetation development (Gu et al 2018). It is remarkable that the most drastic soil and climatic conditions (compact soils and high temperature annual range) lead, somewhat paradoxically, to a faster recovery due to the impossibility of following a classical succession path with successive cohorts of pioneers, long-lived pioneers and mature forest species (Finegan 1996). It is also interesting to note that, for these two variables, the development of the forest will profoundly modify their values. Indeed, the reconstitution of the forest will (a) generate an increasingly buffered microclimate where daily thermal amplitudes will be less perceptible (Ellison et al 2017) and (b) decompact the soil through the action of the tree roots which will increase the porosity and stability of the aggregates (Gu et al 2018), thus improving the soil's water conduction capacity (Ajayi and Horn 2016).

Regardless of the attributes, the local variables retained by the selection procedure are comparable and show similar effects on recovery rates (figure 4). Two variables appear to be very important because they were selected for all of the attributes, namely remnant trees and local soil hydromorphy. (a) Remnant trees have a very positive influence on forest recovery rates (figure 4). Presence of remnant trees improves environmental conditions (da Duarte et al 2010, Pérez-Cárdenas et al 2021) by attracting seed dispersers such as birds (de la Peña-Domene et al 2014) and bats (Galindo-González and Sosa et al 2003, Derroire et al 2016). Recent work has shown that the abundance of remnant trees on a given site directly impacts seed rainfall, and thus affects the density of the stand being replenished (Wu et al 2022). Indeed, these remnant trees left (in)voluntarily in the field allow for rapid initiation of secondary succession (Yarranton and Morrison 1974) and may also facilitate the regeneration of slow-growing species that are typically late-trajectory species, allowing for faster passage through the early stages of succession (Tucker and Kashian 2018). (b) Local soil hydromorphy has a negative effect on biomass recovery rates, composition, and diversity (figure 4). Hydromorphic soils have a particular flora that is adapted to temporary excesses of water. For instance, the presence on hydromorphic soils of extremely competitive herbaceous species of the Maranthaceae family (Cuni-Sanchez et al 2016), which form monospecific mats, hinders the regeneration of other plants.

Among other local variables, but of lesser importance, the duration of agricultural cultivation has a negative effect on the rates of biomass and diversity recovery (figure 4). This duration of cultivation directly influences the abundance of seed and shoot banks, and the soil global fertility (Kennard et al 2002) which are essential for a rapid return of biomass and biodiversity. Secondary forests that have developed on long cultivated sites are dominated by species that are resistant or adapted to repeated disturbance and low soil fertility, sometimes resulting in irregular successional trajectories (de Cássia Guimarães Mesquita et al 2015, Jakovac et al 2016). Secondly, the density of forests in the landscape and/or the proximity of old-growth forests also positively influences recovery. Old-growth forests around secondary forests constitute a seed bank of various trees and shelters for certain animals (Rozendaal et al 2019), they ensure landscape connectivity and provide corridors for the movement of seed dispersing animals. However, as old-growth forest cover has strongly decreased in West African landscapes, secondary plots move from being surrounded by old-growth forest to being surrounded by less and less old-growth forest and more secondary forests. This leads to limited availability of old-growth species as succession proceeds (Melo et al 2013, Gilroy et al 2014, Mendenhall et al 2014) and, in turn, limits our ability to relate the observed succession trajectory (especially for aged secondary plots) to the current environment, which may be very different from the environment in which succession began.

West Africa is home to a large rural population for whom subsistence agriculture is at the expense of forests (Somorin 2010). In other words, deforestation and forest degradation are largely attributable to small-scale, non-mechanised clearing for agriculture (Appiah et al 2007, Curtis et al 2018, Duguma et al 2019). Secondary forests resulting from this deforestation represent a large part of the West African landscape and are valuable to local people. These secondary forests provide valuable ecosystem services, including timber and fuelwood, fodder for livestock, non-timber forest products, and also contribute to water purification (Amani et al 2021, Doua-Bi et al 2021, Sanial et al 2022). In order to ensure that these secondary forests can continue to provide these services and cleared land is reduced to a few suitable areas, it would be necessary to

• Promote sustainable, i.e. non-shifting, agricultural practices such as agroforestry in an agro-ecological transition to ensure the maintenance of household income
• Develop community-based management practices for secondary forests, which do not currently exist, so as not to tie the future of a given forest to an individual decision
• Support commercial chains of non-timber forest products to increase economic benefits from secondary forests (which cannot yet produce timber)
• Assess soil fertility to decide first the suitability of an area for agriculture, and if the site is not suitable compared to other nearby areas, conserve the area to recover forest.

For rapid restoration, local conditions are very important as they condition the initiation of secondary succession after land abandonment. Hence, we have demonstrated the great importance of remnant trees in agricultural fields and fragments of old-growth forests in the nearby environment of abandoned fields to sustain the seed rain and to create landscape connectivity that ensures the movement of disseminating animals. In order to keep the landscape suitable for rapid secondary succession, one should therefore

• Avoid large-scale deforestation, partly caused by concessions granted to large private companies, which totally destructure the landscape, eliminate all remnant trees and thus handicap natural recovery once industrial cultivation is abandoned.
• Protect remnant trees in small farmers' fields by giving the latter true ownership of these trees, so as to consider them as standing capital to be preserved, so that these trees can no longer be sold as concessions granted to timber companies by the governments.
• Systematically protect the few remaining old-growth forests as the main seed sources for natural reforestation.

Regional conditions determine recovery trajectories that are specific to each studied attribute and phytogeographic zone (figure 2). Overall, climatic conditions in the northern part of the study area (dry forests with high variability of precipitation and temperature) are much more suitable for rapid forest restoration (especially of composition, figure 4) than conditions in the south (evergreen forests with low climatic variability). These results suggest that there is a need for politicians, decision-makers and other actors involved in forest management to take into account both the variability of regional climatic conditions and local conditions related to plot history and landscape context in a decision-making framework for the choice of more or less active restoration techniques. The main lines that can be drawn from our study are as follows:

• In the North, when the local context is favourable (good connectivity, presence of numerous remnant trees), we propose to allow secondary succession to take place and to better value these forests in recovery thanks to a community management of the natural resources which are therein.
• In the North, when the local context is unfavourable (poor connectivity, no or few remnant trees), we propose to (a) leave the possibility of converting a part of these surfaces for subsistence agriculture and (b) for the remainder, use assisted natural restoration techniques (Chomba et al 2020) to accelerate the installation/growth of trees. Assisted regeneration have already proved their worth in Sahelian and Sudanian zones and could be implemented at little cost in dry forests where the flora is relatively similar.
• In the South, when the local context is favourable, we propose cleaning, depressing and pruning young trees. Indeed, the relative slowness of reconstitution is not linked here to a lack of regeneration but to competition from herbaceous plants, lianas and pioneer species. These targeted silvicultural interventions also make it possible to manage the forest stand in such a way as to favour species that are useful to local populations.
• In the South, when the local context is unfavourable, succession is so slow that more active interventions by planting local species should be considered if the objective is to reconstitute a forest cover within a few decades. With this in mind, we recommend reforesting with a mixture of local species (Hérault et al 2020) in order to minimise the ecological, health and commercial risks (Messier et al 2022) and using Taungya-type planting techniques which, during the first few years of planting, allow the rural population to farm while maintaining the young plants (Kalame et al 2011).

Beyond that, the technical management options for secondary forests proposed above are only broad guidelines to orient public policies, technical services and to optimise investment and training efforts according to the restoration strategies that are most likely to be implemented in a given region. However, these broad guidelines cannot be used alone in a top-down planning of actions and must imperatively be accompanied by a field diagnosis in the target abandoned fields. This diagnostic stage will certainly be biophysical in order to estimate the real potential for forest restoration based on the current state of the secondary succession, but also socio-economic in order to integrate the aspirations and room for manoeuvre of smallholders in the choice of technical itineraries and species to be promoted.