Can timber provision from Amazonian production forests be sustainable?

Inventory data

Spatial data

Environmental data for both study plots and Amazon-scale extrapolation were extracted from WorldClim 2.0 [15] (precipitation, seasonality of precipitation and solar radiation) and SoilGrids [16] (bulk density, CEC, soil depth, proportion of clay, of sand and of coarse fragment) at a 1 km resolution. When extrapolating to make regional predictions, spatial data was aggregated to a 1° grid by averaging values of all 1 km pixels inside areas available for logging in each 1° cell.

Annual stem mortality rates (as a proportion of live stems), estimated with the metadata from Johnson and colleagues [17], were extracted from the ForestPlots database [18] and interpolated with the R package gstat [19] on a 1° resolution grid. The climax volume, and the gross volume productivity at climax were estimated with the individual-tree-based gap model FORMIND [20]. Climax volume was calculated as the volume of all trees ≥50 cm DBH (per ha) in an old-growth forest; climax gross volume productivity (GVP in figure 1) was calculated as the gross volume gain from photosynthesis (before accounting for respiration losses) of trees ≥50 cm DBH in an old-growth forest. Raster maps of climax volume and climax gross volume productivity were created at 1 km² resolution, and values were then aggregated to a 1° resolution grid.

Zoom In Zoom Out Reset image size

The map of areas available for logging (figure S4) was constructed as the intersection of 3 maps: a buffer of 25 km around all roads and motorable tracks from the OpenStreetMap database [21]; the map of areas outside of protected areas from the World Database on Protected Areas [22] (except the category VI of the IUCN classification, i.e. areas with sustainable use of natural resources that are included in the analysis); and pixels with >90% forest cover from the map developed by Hansen and colleagues [23].

Proportion of timber species

The locally-defined pool of harvested species (used in figure 2) is the per site list of timber species actually harvested. At larger scales though, timber species preferences are subject to much variation in both time and space. We thus decided to use an extended pool of all species that have been recorded as commercial at least once anywhere in Amazonia. The list was derived from (i) a working list of commercial timbers [24]; (ii) commercial species lists provided by national forest services [25–27]; and, (iii) timber species identified by TmFO site principal investigators (personal communications). The potential timber species list is provided in the supplementary material S6. The proportions of potential timber volume in the total forest volume were then interpolated from the 2646 RadamBrasil plots and the pre-logging in TmFO plots. We used the R package gstat [19] to produce an Amazonian map of (pre-logging) timber proportion on a 1° resolution grid (figure S3).

Zoom In Zoom Out Reset image size

The VDDE model

The VDDE model [10] focuses on the volume of live trees with DBH ≥50 cm (the standard minimum cutting size in the Amazon Basin), hereinafter referred to simply as volume. The model was calibrated with volume dynamics data (volumes, volume gain, volume mortality, post-logging volume loss; supplementary material S3) from permanent sample plots. Calibration was carried out using an adapted form of the Hamiltonian Monte Carlo using Stan's programming language [28], and was developed in R [29] (table S1 provides parameters prior and posterior; and presents a convergence diagnostic of the Markov chains).

Three parameters of the VDDE model were expressed as a function of spatially-explicit variables: (i) the climax gross volume productivity α_G, i.e. the annual gross volume increment from trees >50 cm DBH (without volume losses from respiration and mortality) in an old-growth forest; (ii) the potential volume vmax (i.e. the maximum volume that an old-growth forest could reach), and (iii) the pre-logging forest maturity τ₀, which is reflective of the site's pre-logging disturbance level [30]. Other model parameters (β_G, β_M and θ: see supplementary material S5 for a description of the VDDE model [10]) were assumed to be constant across Amazonia.

The climax gross volume productivity α_G and the climax volume Vclimax were extracted from the map obtained with FORMIND [20] (see paragraph 'Spatial data'). Because the potential volume vmax is expected to vary with soil and topography, we allowed it to vary between plots among and within site. The potential volume of plot p in site s was modelled as:

$$\begin{eqnarray}&&v{{\max }}_{p,s}\sim { \mathcal N }({Vcli}{{\max }}_{s},{\sigma }_{v{\max }}),\end{eqnarray} \tag{ 1 }$$

where σ_vmax is the standard deviation, and Vclimax_s is the climax volume predicted with FORMIND [20] at site s.

The pre-logging maturity τ_0,s in site s, was modelled as:

$$\begin{eqnarray}&&{\tau }_{0,s}={\left(\displaystyle \frac{1}{{{mort}}_{s}}\right)}^{\lambda }\cdot (1-\delta {i}_{s}),\end{eqnarray} \tag{ 2 }$$

where mort_s is the annual stem turnover rate (%) [17], and λ > 0 is a power parameter to the relationship between the maturity and the stem turnover rate. In our study area, Western Amazonian forests grow on nutrient-rich but unstable soils [31] and are thus more prone to natural disturbances like big blow-downs [32] than northeastern Amazonian forests. Frequent disturbances and high resource availability favour fast-growing species with high turnover rates. For this reason we chose the stem turnover rate as a proxy of the disturbance regime. Because in some sites there were human disturbances prior to the logging experiment, we added a parameter δi_s that represents the gap between the estimated and the expected pre-logging maturity at site s (table S1 provides parameters prior and posterior).

Accounting for defective stems

A significant part of large trees in natural forests have hollows or other defects that make them unsuitable for timber uses [33]. The proportion of commercial volume with defects unacceptable for sawmills ranges 20%–50% in the Brazilian Amazon [34–36]; an extensive data collection in forest concessions in French Guiana reported that on average 20% of harvestable stems had hollows and were not harvested (ONF: personnal communication; [26]). We thus multiplied all timber volumes in our simulations by a factor (1-Pdef), with Pdef the proportion of defective volume modelled as:

$$\begin{eqnarray}&&{Pdef}\sim { \mathcal B }{eta}(6,14),\end{eqnarray} \tag{ 3 }$$

where ${ \mathcal B }{eta}(6,14)$ is the beta distribution of shape parameters α = 6 and β = 14. The mean value of Pdef is 30%; to reflect the uncertainty on this value, we chose a distribution with a large 95% credible interval (16%–49%).

Simulations were carried for every pixel of a 1° grid. We simulated five scenarios: (1) standard logging rules (logging intensity Vext = 20 m³ ha⁻¹, cutting cycle 30 years); (2) low logging intensity (10 ${{\rm{m}}}^{3}\,{{\rm{ha}}}^{-1}$) with a standard cutting cycle (30 years); (3) high logging intensity (30 ${{\rm{m}}}^{3}\,{{\rm{ha}}}^{-1}$) with a standard cutting cycle; (4) short cutting cycle (15 years) with a standard logging intensity (20 ${{\rm{m}}}^{3}\,{{\rm{ha}}}^{-1}$); (5) long cutting cycle (65 years) with a standard logging intensity (20 ${{\rm{m}}}^{3}\,{{\rm{ha}}}^{-1}$).

In each scenario, we consider that, each year, $\tfrac{1}{{trot}}$ of the area available for logging is actually logged (where trot is the cutting cycle), so that an area is logged every trot years and the total area logged each year is constant. Due mostly to slope restrictions and riparian reserves, but also heavy forest degradation in some parts of the Amazon, the area logged typically represent 60% of the total area allocated for logging [37, 38]. We multiplied the annual harvested area by a coefficient $\pi \sim { \mathcal B }{eta}(8.2,\ 5.9)$, where ${ \mathcal B }{eta}(8.2,\ 5.9)$ is the beta distribution calibrated with data from logging concessions in French Guiana (reported in figure S11).

To propagate errors on results, the following steps were repeatedly taken:

• 1.  
  At each location, map values (α_G, Vclimax, and mort in equations (1), (2)) are drawn from their distribution (error estimation is described in supplementary material S4).
• 2.  
  At each location, model parameters are drawn from their posterior distribution (see table S1). *Timber volumes (per ha) are calculated as:

  $$\begin{eqnarray}&&{V}_{t,{pix},l}={Vol}({\tau }_{t,{pix},l})\cdot {\omega }_{t,{pix},l}\cdot (1-{Pdef}),\end{eqnarray} \tag{ 4 }$$

  where ${V}_{t,{pix},l}$ is the predicted timber volume t years after the first harvest in pixel pix in scenario $l,{\tau }_{t,{pix},l}$ is the predicted maturity, ${Vol}({\tau }_{t,{pix},l})$ is the volume of all trees ≥50 cm DBH according to equation (15), ${\omega }_{t,{pix},l}$ is the proportion of timber volume and Pdef is the proportion of defective volume;
• 3.  
  For each pixel, each time step t ∈ [1, 300] and each scenario 1 ≤ l ≤ 5, the total timber volume is calculated as:

  $$\begin{eqnarray}&&{{Vtot}}_{t,l}=\displaystyle \sum _{{pix}}\left[\left({V}_{t,{pix},l}\right)\cdot {{area}}_{{pix}}\cdot \pi \right],\end{eqnarray} \tag{ 5 }$$

  where ${V}_{t,{pix},l}$ is the timber volume (per ha) t years after the first harvest in pixel pix in scenario l; and ${{area}}_{{pix}}\cdot \pi$ is the area (ha) inside pixel pix that is available for logging.
• 4.  
  The real extracted volume (per ha) from each pixel pix is calculated as the minimum between the extracted volume in scenario l (i.e. the timber volume expected to be harvested) and the available timber volume at the time of logging. The total extracted volume at year t is the sum of the actual extracted volume from areas logged at t.
• 5.  
  Potential timber volume recovery (%) is calculated as the increase in potential timber volume over the first cutting cycle, divided by the total extracted volume (figure 3). The annual timber recovery is calculated as the increase in potential timber volume between two consecutive years (figure 4).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Steps 1–4 were repeated 100 times and summary statistics were calculated. Timber recovery and timber extraction were compared to the current and future demand for sawlogs. Current demand was assessed as the production of sawlogs in the Amazon region in 2004, 31 Mm³ according to the Imazon [12]. Future increase in demand was assumed to follow the trend of increase in sawnwood consumption in South America as projected with the Global Forest Products Model [39]. We thus computed the proportional increase predicted between 2006 and 2060 for four Intergovernmental Panel on Climate Change Scenarios (A1B, A2, B2 and A1B-Low Fuelwood) [39] and multiplied the current demand for sawlogs by this increase to get the future demand for sawlogs (figure 4).

Recovery of harvested species volume by the end of the first cutting cycle varied threefold across the experimental sites (figure 2(a)) and increased with the pre-logging proportion of the total volume shared by the local pool of harvested commercial species (Pearson coefficient ρ = 0.58). At the sites with the largest abundance of locally harvested species (INPA and La Chonta with >70% of stems ≥50 cm DBH), timber volumes are predicted to recover faster, because there is less competition with non-commercial species. This finding highlights the importance of expanding the local pool of species harvested in order to maintain timber stocks over time.

Regional variation in the rate of timber volume recovery of all potential timber species were consistent across logging intensities and cutting cycle lengths (figures 3(a)–(e)). Median timber recovery was highest in Western Amazonia (0.30 [0.18, 0.40] m³ ha⁻¹ yr⁻¹—numbers in [] represent the 95% credible interval) and lowest in the Guiana Shield (0.26 [0.16, 0.34] m³ ha⁻¹ yr⁻¹). These results resemble those from studies in old-growth forests that revealed higher rates of both wood production and forest demographic rates in western Amazonia than in the northeast [17, 31]. This pattern is potentially due to more frequent natural disturbances [32], or to spatial differences in seasonality and soil properties [31]. This result means that logging regulations need to reflect regional differences.

Lesser known Amazonian timber species compose a small share of the global tropical timber market, which remains heavily dominated by a few overexploited species [40, 41]. Selective logging in Amazonia usually targets a few high-value species such as mahogany (Swietenia macrophylla) and ipê (Handroanthus spp) [40] that typically represent <20% of the total volume in a particular site [41, 42]. When overexploited, these species' volume recovery is compromised within a typical 30 year cutting cycle [40]. Due to low recovery rates of prized timber species, what is available for second harvests is often species with low timber market values compared to the costs of extraction and transport [43].

Even if volume recovery scenarios include all the 348 lesser-known timber species, our simulations indicated that with a logging intensity of 20 m³ ha⁻¹ logged forests recover at most 70% of their pre-logging timber volumes (figure 3) within a typical 30 year cutting cycle. This result is consistent with a variety of local studies reporting that standard 30–40 year cutting cycles are insufficient for full recovery of timber stocks [3, 40]. This means that even with a substantial increase in the number of merchantable species, timber stocks will continue to decline in Amazonian production forests if current logging practices (extraction of around 20 m³ ha⁻¹ every 30 years) persist (figure 4).

Independently of logging intensity and cutting cycle length, median timber recovery from forest regrowth was <30 Mm³ yr⁻¹ (figure 4(c)). This over-harvesting results in a reduction in Amazon-wide timber stocks in all scenarios (figure 4(a)), meaning that natural forest regrowth will be insufficient to supply the commercial demand in the long-term. Moreover, the actual sawlog extraction could be higher than official numbers suggest: illegal logging is ubiquitous in the region, and is estimated to produce a volume of wood equivalent to 20%–60% of the legal timber markets [44, 45], further decreasing the likelihood of a sustainable timber supply from Amazonian production forests.

We stress that our model estimates are based on optimal scenarios of the recovery potential of Amazonian production forests. (i) Our plots showed no signs of having suffered severe recent human disturbance prior to logging (e.g. fire, uncontrolled logging, or fragmentation) whereas this is not the case for an estimated one-third of Amazonian forests [46]. Such disturbances might reduce forest resilience to logging [47]. (ii) Reduced-impact logging techniques were employed in most of our experimental sites [11], but these recommended logging practices are seldom implemented in the tropics [4]. (iii) Not harvesting big defective stems and keeping them in the forest will increase their proportion in the next harvests. (iv) Our scenarios do not account for post-logging degradation (e.g. fires and illegal logging [48]) or deforestation [49], which are fairly ubiquitous in the region [47, 50]. Therefore, our results represent the maximum potential volume recovery of Amazonian production forests,which is unlikely to be attained in the real world.

While the fate of Amazonian production forests remains uncertain, several studies call attention to the rising impacts of human activities on the functioning and provision of ecosystem services [8, 51, 52]. Deforestation, forest degradation, and climate change will continue to affect the resilience of Amazonian forests to future disturbances including their ability to recover timber stocks after logging [47, 53]. Moreover, future trends in deforestation can be substantially affected by political choices (e.g. road building [48], law enforcement, agricultural subsidies, access to credit [54], and corruption [55]), which were not considered in our conservative scenarios with no deforestation. Climate change is also expected to decrease timber stocks and productivity through drier and hotter climate leading to higher mortality of large trees, which have longer hydraulic path, higher leaf area and crown exposure [56] (figure 5(a)). Increased frequency and intensity of disturbances are expected to decrease potential timber stocks while timber productivity is enhanced due to the decreased proportion of less productive old-growth forests (figure 5(b)).

Zoom In Zoom Out Reset image size

Our results show that with current cutting cycles and logging intensities, forest regrowth is too slow to recover timber stocks (figures 3, 4), highlighting the need to decrease the pressure on natural production forests by adopting longer cutting cycles, and reducing logging intensities and incidental damage to the stand through reduced-impact techniques [4]. Silvicultural interventions applied to increase the stocking, growth, and commercial yields from merchantable species (e.g. liana cutting, future crop tree liberation, and enrichment planting) could also help turning the tide of forest depletion, as well as providing several other benefits such as increased carbon storage and employment [57].

Enforcing longer cutting cycles, lower intensities, reduced-impact techniques or post-logging interventions will likely increase long-term forest recovery but decrease the short-term financial benefits from legal selective logging. Moreover, restricting legal timber extraction could potentially lead to an increase in illegal logging and forest conversion [58]. This means that parallel to adopting stronger logging regulations, additional efforts on law and forest tenure enforcement will be needed. These policies should include regional coordination to avoid illegal trade and leakages effects [58, 59]. Economic viability of tropical forest management will also increase with timber prices, which are currently low compared to production costs [43], and with sawmill efficiency, currently around 35% [60]. Efforts should also be done to change consumer preferences both in terms of species and size of logs, e.g. the potential use of branches (and not only trunks) could increase timber production without additional damage to the forest. Another opportunity to increase financial revenues is to develop economic mechanisms to value other goods and services provided by the forest such as carbon storage (e.g. REDD+), hydrology, biodiversity, ecotourism, and non-timber forest product management [61].

Changing logging practices may not be enough to meet rising demands for wood products. Additional sources of timber could come from various restoration systems: plantations of exotic or native species, enriched secondary or degraded forests [62], integrated crop-livestock-forestry systems and other agroforestry systems [63]. Increasing the area of timber plantations could significantly reduce the pressure on Amazonian natural forests [64]. Tree plantations have the potential to produce large quantities of timber on relatively small areas: timber plantations in Brazil, mostly fast-growing eucalyptus and pine, can produce 200–400 m³ ha⁻¹ of roundwood on 10–15 year cycles [65]. Such exotic species produce low-grade timber that is not directly equivalent to high-value wood currently extracted from Amazonian natural forests, but there is a potential to develop plantations of high-value native species [66, 67], even though technical alternatives are still scarce in Amazonia [67]. Moreover, it is likely that with ongoing technological advances, future high-grade timber demand will be gradually substituted by less-valuable fast-growing timbers transformed into highly-resistant materials [68].

The rising interest in tropical forest restoration, initiated by the Bonn challenge in 2011 [69], has led Brazil to commit to restore 12 Mha of forest by 2030 [70], and has motivated an initiative to restore 30 000 ha of forests in the Brazilian Amazon by 2023 [71]. These initiatives can provide opportunities to combine efforts to restore environmental values (e.g. carbon, biodiversity, and water cycle) including timber production [62], and fund applied research for both ecological restoration and timber production [72].