Spatial priorities for conserving the most intact biodiverse forests within Central Africa

We restricted the analysis to the remaining forests of Central Africa as defined by our forest cover data at a resolution of 500 m (Hansen et al 2013, shown in figure 1). The forest ecosystems map was based on (Shapiro et al 2020) who used a combination of existing forest ecosystem maps, satellite remote-sensing data, and ground surveys, which resulted in 64 forest ecosystem classes (table S1).

Zoom In Zoom Out Reset image size

Biodiversity patterns were indexed by the geographic distribution of large mammals, for which data are available at the appropriate scale, and which indicate overall biotic patterns. Large mammal species are often the first lost from intact forest as human impacts increase, because they are preferential targets for hunting (Plumptre et al 2019). We therefore incorporated models of great apes and forest elephants in our analysis. Five models were previously developed to predict distribution, density or relative abundance across their range, including (figure 1): forest elephant (Loxodonta africana) (Maisels et al 2013); western lowland gorilla (Gorilla gorilla gorilla) and central chimpanzee (Pan troglodytes troglodytes) (Strindberg et al 2018); Grauer's gorilla (Gorilla beringei graueri) (Plumptre et al 2016); eastern chimpanzee (Pan troglodytes schweinfurthii) (Plumptre et al 2010); Bonobo (Pan paniscus) (Hickey et al 2013).

Two sets of data were used as indicators of forest intactness. The first was the Intact Forest Landscapes (IFL) data, which depicted IFLs for 2013 based on Potapov (2017).

In addition to IFLs, we also created the Forest Intactness Index (FII) given that IFLs are limited by being a binary measure of forest intactness (i.e. forests are intact or not). We constructed the FII using a combination of: 1) Forest Condition Index (FCI), and 2) Forest Pressure Index (FPI). The FCI was based on Shapiro et al (2020) who applied a multi-resolution spatial pattern analysis (MSPA) to classify the forest into four different fragmentation types based on its pattern of edge effects (Soille & Vogt 2009). For each change in fragmentation type, they used a measure of potential biomass of core forest, against predicted biomass loss within the edge type to construct an index (see supplementary materials for details). Values ranged from 0–100 and were rescaled to 0–1 to be combined with FPI. The FPI was developed based on the potential anthropogenic pressure on forests, which incorporated the combination of location of human settlements, population density, and accessibility (see supplementary materials). The FII was then created by the multiplication of FCI and FPI to place an emphasis on identifying the most intact areas, i.e. those with low fragmentation and degradation, and low anthropogenic pressure.

The protected areas data were compiled from the World Database on Protected areas, and data sources from WRI and WWF (Pélissier et al 2019). From this we calculated the protected forest area based on the overlap of these protected areas and the forest cover layer (figure 1).

We compared four different scenarios that incorporated different combinations of biodiversity, forest intactness, and the inclusion of protected forests (table 1). We used the conservation planning tool Zonation to prioritize all forested areas at a resolution of 500 m using an additive benefit function (Moilanen 2007). The algorithm starts by removing the least important cell based on maximizing the prioritization objectives until it reaches the highest ranked cells. The output is an importance ranking of each cell across the region in meeting the spatial prioritization objectives.

The biodiversity representation objective was to maximize the inclusion of each of the 64 ecosystem types and five species within the highly ranked cells. Recognizing that lowland and terra firme forest types are targeted by industrial activity, especially logging (Kleinschroth et al 2019b), and thus more vulnerable to loss, we weighted these forest types ten times higher than the higher elevation and swamp forest ecosystem types within the conservation feature weightings system in Zonation.

One forest objective was to select forests with the highest forest intactness based on the FII. Another was to prioritize IFLs, which are the largest remaining patches of intact forest, but did not necessarily exclude all other forests outside of these landscapes. We included both indices due to IFLs being able to preference the spatial prioritization to the largest blocks of forests, and the FII to ensure at the pixel level (500 m) that forest intactness was generally being maximized.

The first scenario (S1) 'biodiversity representation and forest intactness objectives (with protected forests)' focused on the potential expansion of conservation outside of existing protected areas (e.g. set-asides within forestry concessions, new protected areas, community-based conservation). When protected areas are considered, Zonation maximizes the complementarity of what is already protected when identifying new priority areas.

The second scenario (S2) 'biodiversity representation and forest intactness objectives (without existing protected areas)' differed from S1 by ignoring the existing protected areas to eliminate potential biases arising from inclusion of the existing protected areas (Blake et al 2009, Campos-Arceiz and Blake 2011). Comparing S1 and S2 provided an indication of the efficacy of the current protected forests against these objectives.

The third scenario (S3) 'biodiversity representation objectives only' was the same as S1, except we did not consider the FII or IFLs. Comparison with other scenarios provides an indication of the benefits of including forest intactness measures.

The fourth scenario (S4) 'forest intactness index only' did not use Zonation but rather identified priority areas only using the FII. For this we ranked the cells based on their forest intactness score with cells of higher forest intactness being more important. Optimization of multiple objectives including biodiversity, IFLs, and spatial configuration were not considered. This allowed a comparison of results when using only forest intactness compared to the baseline prioritization incorporating multiple objectives (S1).

All scenarios using Zonation (S1-3) maximized connectivity and patch size of highly ranked areas using a Zonation function called the Boundary Length Penalty that aims to reduce the length of the boundary of the highly ranked areas using a weighting of 0.05 after exploring numerous values to find one that balances the amount of clumping with the efficiency of the results maximizing other objectives.

To compare scenarios, we either identified the top 10% most highly ranked cells outside of protected forests or the top 25% of most highly ranked cells overall. Around 15% of forests are currently protected within the region, therefore 10% was a reasonable level of conservation expansion outside of the existing protected areas, yielding up to a total of 25% as the threshold, and allowing a reasonable comparison of scenarios. We measure the results for the region and make comparisons for each country.

Within the region forests cover around 213 million ha of which ∼15% are found within protected forests, although there is variation across countries (table 2). We identified spatial priorities to expand conservation outside of the protected area network for both biodiversity representation and forest intactness objectives (S1). Measuring the top 10% ranked cells outside of the existing protected area network, we found over half of priority areas (56%) are within DRC (table 3, figure 2). This might be partly explained by the amount of forests there, with 61% of forests in the region within DRC (table 2, figure 2). The spatial priorities are concentrated in the eastern part of DRC and surround existing protected areas (figure 2). We also found spatial priorities were particularly high in Gabon (20%) and Republic of Congo (ROC) (15%) based on the top 10% ranking and proportion of country (table 3). Similarly, this could be accounted for by forest areas there containing the next highest forest coverage in the region with ROC and Gabon both having 11% of total forests, and by the fact that these countries have higher average forest intactness index scores (Gabon = 0.85 and ROC = 0.84) (table 3). Spatial priorities in Equatorial Guinea are less significant, probably resulting from its size, it already has extensive protected areas, and does not contain as much forest compared to other countries in the region. It also has the lowest average forest intactness score (0.45) (table 3).

Zoom In Zoom Out Reset image size

Spatial priorities identified by S2, which ignored existing protected areas, did not differ significantly from those identified by S1. DRC still contained the largest area of spatial priorities when comparing the top 25% ranked areas, but there was a slight increase in high priority areas within ROC and Gabon (both 19%) (table 3). While there was not a large overlap (52%) when measuring the top 25% priority areas between S1 and S2 (table 4, figure 3), we did find the same level of average forest ecosystem representation within priority areas, both averaging 55% (figure 4). This indicates that the existing protected areas are reasonably well placed to represent the different forest ecosystems compared to that of the optimal solution. However, we found the average FII score of the prioritized areas was much higher when we removed existing protected areas from the analysis (average score 0.82 in S2 compared with 0.72 for S1) (figure 4). This indicates that the placement of existing protected areas is skewed towards areas with lower FII values compared to that of the optimal intactness solution (S2).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To understand the influence of including forest intactness objectives when identifying spatial priorities for biodiversity, we compared S1 with a scenario that did not include the forest intactness objectives (S3). We found that when comparing the top 10% of priority areas outside of protected forests, less than half of the priority cells were overlapping (47%) (table 4. Figure 3). Ignoring forest intactness objectives resulted in the top 25% of high priority areas (including the protected forests) increasing in ecosystem representation (based on average proportional area across ecosystems within priority areas) by 7%, with an average representation of 62% for S3 compared with 55% for S1 (table 3, figure 4). This increase in ecosystem representation benefits was associated with a 10% decrease in the average FII score from 0.72 (S1) to 0.62 (S3).

The last scenario (S4) identified spatial priorities based solely on FII. Comparing the overlap of top 10% ranked priorities outside of the current protected areas network between S1 and S4, we found that there was a 74% difference in priority areas between them (table 4, figure 3). Because biodiversity representation was ignored, we also found that while the priority areas in S4 have a much higher average forest intactness score (0.98) compared to (0.77) (table 2), the biodiversity benefits (measuring average forest ecosystem representation) were far less. When comparing the top 25% of priority areas, the average representation of ecosystems was 21% for S4 compared with 55% for S1 and a spatial overlap of 51% (table 4, figure 3). This demonstrates that the two approaches produce quite different results for ecosystem representation, with S4 resulting in much less ecosystem representation than S1.

We found that the inclusion of multiple objectives in the prioritization approach ensured that both biodiversity representation benefits and areas with the highest forest intactness could be achieved although neither objective was totally maximized. Setting different priorities, such as biodiversity representation only versus only prioritizing forest intactness, resulted in trade-offs (figure 4). When the prioritization did not include forest intactness (FII and IFLs; S3), biodiversity representation was maximized, but the areas prioritized were less intact. The existing protected areas represented ecosystems almost as well as the optimal solution (S2), but surprisingly existing protected areas were less intact than some areas outside of protected areas with similar biodiversity. Most importantly, we found that if we ignored biodiversity representation and only prioritized areas with the highest forest intactness (using FII), selected priority areas performed poorly in terms of representing ecosystems despite having the highest forest intactness of all scenarios (figure 4). This has important implications for using only forest intactness indices to prioritize biodiversity (e.g. Mccloskey and Spalding 1989, Potapov et al 2008) or human pressures or stressors (e.g. Theobald 2013, Venter et al 2016). For example, if only using IFLs to inform set asides for forest concession plans, it will likely perform poorly for biodiversity, which is why this method is being used to inform HCV planning in Central Africa (Supplementary Information).

Several areas of future research could improve our approach. First, improving our forest condition metric to appropriately capture ecosystem types that are naturally fragmented woodlands (e.g. the northern and southern regions of Central Africa that transition from more closed forests to open types). Second, refine the forest pressure index that currently makes assumptions about travel time, such as assuming human travel was related mainly to the location of settlements and roads and that there was higher forest pressure in places with higher population density. The quantified accessibility of (and hence pressure on) the ecosystems to humans may be inaccurate in places where persons may pursue alternative paths to access forest resources, including cross-border travel, or choose an alternate place to gather resources aside from the closest ecosystem (e.g. for cultural reasons), or where mapping was at too broad a resolution to quantify access paths and fine-scale population centers (e.g. artisinal mines, Maclin et al 2017). Our index could also have also overestimated human pressure in places where good conservation and resource management might have reduced pressures. Better mapping of where management is occurring, and evaluation of its effectiveness at restoring or maintaining biodiversity, is required to refine such overestimates. Third, further research to better understand patterns of defaunation are need as they are likely under-estimated here (Benítez-López et al 2019).

To ensure long term sustainability of conservation plans, it has become best practice within spatial conservation planning to include socio-economic factors in addition to biodiversity and other values when applying spatial prioritizations (Groves and Game 2016, Karimi et al 2017). This was not considered here in order to focus on the ecological values of the forests, but could be included within this approach in the future. Conservation actions usually benefit some groups more than others, and such social inequity is likely to affect the probability of achieving conservation objectives (Klein et al 2015). A range of techniques are available for including socio-economic factors into prioritization, including participatory mapping of cultural values and developing maps that reflect underlying uncertainty in conservation success (Mcbride et al 2007).

While the forests of Central Africa are extensive, their forest intactness varies and is continuing to decline. Even within the relatively intact forests, there are likely to be only small areas of true old-growth forests, or 'refugia', which date back to around 18 000 years (Maley 1996). Much of this region is now covered by more recent forests that have encroached into the savannahs in the last few hundred years, due to the post-Pleistocene forest expansion which continues today (Leal 2001, Mitchard and Flintrop 2013), with some forests on the edges less than 60 years old (Mitchard et al 2009). While being very difficult to detect using remote sensing methods, identifying and managing these refugia is likely to be particularly important.