Despite myriad challenges, clinicians see room for progress.
Image credit: Shutterstock/David Tadevosian.
New Research In
Physical Sciences
Featured Portals
Articles by Topic
Biological Sciences
Featured Portals
Articles by Topic
Edited by Anthony J. Bebbington, Clark University, Worcester, MA, and approved October 10, 2019 (received for review March 7, 2019)
Significance
REDD+ is the main international policy to reduce CO2 emissions from deforestation in tropical countries. However, there are no empirical studies on the impact of REDD+ implemented at the country level. Here, we evaluated a nationwide REDD+ program implemented in Guyana. We apply synthetic matching to estimate tree cover loss that would have occurred in the absence of the national REDD+ program (the counterfactual scenario). We found evidence that the program reduced tree cover loss by 35%, equivalent to 12.8 million tons of avoided carbon emissions. We also found evidence of accelerated tree cover loss at the end of the program. A multinational REDD+ approach implemented in a region that includes continuous forest protection payments will improve national REDD+ outcomes.
Abstract
Reducing emissions from deforestation and forest degradation (REDD+) is a climate change mitigation policy in which rich countries provide payments to developing countries for protecting their forests. In 2009, the countries of Norway and Guyana entered into one of the first bilateral REDD+ programs, with Norway offering to pay US$250 million to Guyana if annual deforestation rates remained below 0.056% from 2010 to 2015. To quantify the impact of this national REDD+ program, we construct a counterfactual times-series trajectory of annual tree cover loss using synthetic matching. This analytical approach allows us to quantify tree cover loss that would have occurred in the absence of the Norway-Guyana REDD+ program. We found that the Norway-Guyana REDD+ program reduced tree cover loss by 35% during the implementation period (2010 to 2015), equivalent to 12.8 million tons of avoided CO2 emissions. Our analysis indicates that national REDD+ payments attenuated the effect of increases in gold prices, an internationally traded commodity that is the primary deforestation driver in Guyana. Overall, we found strong evidence that the program met the additionality criteria of REDD+. However, we found that tree cover loss increased after the payments ended, and therefore, our results suggest that without continued payments, forest protection is not guaranteed. On the issue of leakage, which is complex and difficult to quantify, a multinational REDD+ program for a region could address leakage that results from differences in forest policies between neighboring countries.
In 2007, under the Bali Action Plan, the global community adopted a policy mechanism known as reducing emissions from deforestation and degradation (REDD+) as a climate mitigation strategy (1). REDD+ at its genesis was intended to incentivize and compensate developing countries (non-Annex I) for verified emissions reduction through payments from developed countries (Annex I) via a compliance carbon market (1). These financial transfers for the protection, sustainable management, and enhancement of vulnerable forest carbon stocks was developed to offset the opportunity costs associated with agricultural and other commodity production (e.g., palm oil, cattle, gold) that drive deforestation (2, 3). REDD+ was thus seen as a cost-effective means to help limit global warming to 2 °C and rectified a deficiency in its predecessor, the Kyoto protocol and subsequent United Nations Framework Convention on Climate Change (UNFCCC) Conference of the Parties meetings, that failed to tackle greenhouse gas emissions from deforestation.
In the years after the Bali Action Plan, REDD+ attracted global interest, garnering more than US $9.8 billion in financing commitments from 2006 to 2014, primarily through bilateral and multilateral funding (4). As of 2018, there are an estimated 467 REDD+ projects and programs located in 57 countries, of which 359 are classified as active interventions, 67 have been completed, and 42 either have not been initiated or have been discontinued (5). However, almost 2 decades after the conception of REDD+, less than one-half of all REDD+ finance (42%) has been allocated for ex-post results-based and verified emissions reduction payments (5). Furthermore, there is a paucity of REDD+ interventions implemented at the national jurisdictional level that use results-based outcomes for evaluation (6). The lack of impact evaluation of REDD+ initiatives at the project and the national levels has led to questions on its effectiveness, and has handicapped our ability to learn from REDD+ climate finance (6⇓–8). Understanding the outcomes of REDD+ is pivotal to inform current initiatives as well as the design of future iterations of results-based payments for climate mitigation (9).
In this article, we assess the impact of a REDD+ program implemented at the national jurisdictional level through a bilateral agreement between the Kingdom of Norway, the largest donor of global REDD+ climate finance, and Guyana, a high forest cover, low deforestation (HFLD*) country (10⇓–12). We apply an emerging policy evaluation approach called the synthetic control method, or synthetic matching, to quantify the impact of the REDD+ program on Guyana’s tree cover loss. Synthetic matching is an empirical approach that constructs a counterfactual time-series scenario, in our case “what tree cover loss would have occurred without the REDD+ program?” to evaluate the causal impact of policy interventions. As is the case in this paper, the method is particularly relevant when assessing outcomes of a policy implemented over large aggregate jurisdictions, such as a single country or state, where there are not identifiable comparisons for causal impact inference (13⇓⇓⇓–17).
Synthetic matching builds a counterfactual country based on the outcome of interest (in our case, the rate of tree cover loss), and observable covariates, or factors that drive that outcome (in our case, deforestation drivers such as agricultural expansion, and mining) collected for countries that can serve as appropriate comparisons to the treated country (Fig. 1). To reduce, and hopefully eliminate selection bias, the comparison countries in synthetic matching are weighted based on the similarity of their covariates to the treated country. In addition to matching on the covariates, the average outcome of interest before the intervention is matched based on different linear combinations to control for unobserved factors that influence the outcome and whose effect vary over time (14). Thus, synthetic matching explicitly accounts for unobserved drivers of the outcome (i.e., factors that influence the outcome that we do not know of or do not have empirical data for). This matching on outcomes is an improvement over other quasiexperimental impact evaluation methods, such as propensity score matching, that only account for observed and measured covariates to build the counterfactual scenario (18⇓–20).
Fig. 1.
Map with high forest cover, low deforestation (HFLD) countries that are in the pretransition phase based on the forest transition theory. Base map (green) is forest cover estimated in year 2000 applying a 30% forest cover threshold; red is measured tree cover loss, and blue is measured tree cover gain between 2001 and 2017 derived from the Global Forest Change dataset (22).
REDD+ Case Study
Guyana is a critical test case for national REDD+ programs† as it has one of the most intact tracts of old-growth tropical rainforest globally, with an estimated forest cover of 85% (18.4 million ha) and an annual deforestation rate of 0.01% for the period 1990 to 2000 (21). The Norway–Guyana REDD+ program is unique in that it is the first REDD+ ex-post results-based payment for emissions reduction to a HFLD country. This REDD+ initiative is implemented at the national jurisdictional level and employs a combined incentive-crediting baseline for emissions reductions (SI Appendix, 1). We construct a theory of change (ToC) of the Norway–Guyana REDD+ program based on the intended impact pathway for forest protection described in the agreement in order to evaluate this national REDD+ program (SI Appendix, 2).
We can quantify the impact of this national REDD+ program because the outcome of interest, annual tree cover loss, is available due to advances in remote sensing and computational power that has resulted in annual global forest cover change datasets at 30-m spatial resolution (22). We use these Global Forest Change data for several years prior (2000 to 2009), during (2010 to 2015), and after the Norway–Guyana REDD+ program (2016 to 2017), to construct a time-series counterfactual scenario of tree cover loss for Guyana applying the synthetic control method. Our methodological approach, which builds a time-series counterfactual trajectory for Guyana’s tree cover loss, directly quantifies the additionality criteria of national REDD+, i.e., did this REDD+ intervention reduce forest loss during the intervention period? We also use the tree cover loss time-series data and analytical outputs from the synthetic matching to explore key attributes that influence the effectiveness of REDD+: leakage, i.e., the displacement of deforestation to neighboring countries, and permanence, i.e., how long forests protected under the national REDD+ intervention remain intact after the program ended. Additionality, absence of leakage, and permanence are considered central to the success of REDD+ as a climate mitigation strategy (Fig. 2) (23⇓–25).
Fig. 2.
Conceptual diagram of the key attributes of REDD+ as an international climate mitigation strategy. Additionality is the reduction in deforestation and emissions achieved because of the REDD+ intervention that is site specific. The effectiveness of the REDD+ intervention is, however, mediated by displaced deforestation outside of the REDD+ jurisdiction (leakage), or increased deforestation after the REDD+ intervention has ended (permanence).
Guyana is also a unique national REDD+ case study because of its development status as a pretransition country characterized by high forest cover and low deforestation rates based on the forest transition theory (SI Appendix, 3) (12, 26). The central focus of our study is to determine whether the Norway–Guyana REDD+ program reduced forest loss. Results from our analysis will address a major concern at the conception of national REDD+, that is, whether monetary payments to countries to protect forests with historically high forest cover and low deforestation rates would provide a sufficient incentive to limit their deforestation (23, 27).
Results
Tree Cover Loss in Guyana and Other HFLD Countries.
In 2000, forest cover in Guyana was 18.9 million hectares, covering ∼88% of the country. The average annual tree cover loss in Guyana during the preimplementation period (2001 to 2009) was 0.036% y−1 (6,787 ha⋅y−1) and increased to 0.056% y−1 (10,652 ha⋅y−1) during the Norway–Guyana REDD+ program (2010 to 2015; Fig. 3A). During the implementation of the Norway–Guyana REDD+ program (2010 to 2015), annual tree cover loss did not increase above the 0.1% threshold at which REDD+ payments would have ceased. There were 3 y where deforestation was higher than the 0.056% baseline deforestation rate, against which payment deductions occurred (Fig. 3B) (28). In the 2 y after the Norway–Guyana REDD+ program ended (2016 to 2017), tree cover loss more than doubled to 0.122% y−1 (22,985 ha⋅y−1; Fig. 3A).
Fig. 3.
Annual tree cover loss rates (percentage) extracted from the Global Forest Change dataset for 2001 to 2017 and synthetic matching outputs. (A) Tree cover loss rates for Guyana and comparison countries classified as HFLD countries. (B) Observed trends in gross tree cover loss for Guyana compared to the synthetic counterfactual for Guyana. (C) The difference between the observed rate of tree cover loss for Guyana and its synthetic counterfactual. The Norway–Guyana REDD+ program was implemented from 2010 to 2015 (blue panel). The agreement stipulated that REDD+ payments would decline above deforestation rates of 0.056%, and no payments would be made if deforestation reaches 0.10%.
For the HFLD comparison countries used to build the synthetic counterfactual for Guyana, all had forest cover >60% of their land area in 2000. Average tree cover loss across these countries was 0.136% y−1 (136,721 ha⋅y−1) in 2001 to 2009, 0.214% y−1 (216,441 ha⋅y−1) in 2010 to 2015, and 0.362% y−1 (365,190 ha⋅y−1) in 2016 to 2017 (Fig. 3A). Gabon and Suriname had the most similar tree cover loss rates and trajectory to that of Guyana in the pretreatment period (2000 to 2009). Tree cover loss rates in Gabon and Suriname, respectively, were 0.065% y−1 (16,096 ha⋅y−1) and 0.034% y−1 (4,758 ha⋅y−1) from 2001 to 2009; 0.112% y−1 (27,485 ha⋅y−1) and 0.082% y−1 (11, 507 ha⋅y−1) from 2010 to 2015; and 0.145% y−1 (35,402 ha⋅y−1) and 0.126% y−1 (17,458 ha⋅y−1) from 2016 to 2017. Notably, Guyana’s tree cover loss was most similar to that of Suriname, before and after the Norway–Guyana REDD+ program, but Guyana’s tree cover loss was 32% lower than Suriname’s during the program implementation period (2010 to 2015).
Additionality: Effect of the Norway–Guyana REDD+ Program.
Our estimate of the effect of the Norway–Guyana REDD+ program on tree cover loss in Guyana is given by the difference in deforestation rates between Guyana and its synthetic counterfactual for the 2010 to 2015 period (SI Appendix, 4). Estimated annual tree cover loss in the synthetic counterfactual for Guyana was higher (0.087% y−1) than observed tree cover loss in Guyana (0.056% y−1). The difference between the counterfactual and observed tree cover loss is equivalent to a 0.031% y−1 reduction in the tree cover loss rate for Guyana over the 5 y, with the largest reduction occurring in 2012 to 2014 (Fig. 3C). Note that because there are only 4 countries with valid synthetic counterfactuals (Guyana, Gabon, Republic of Congo, and Suriname; SI Appendix, 5), we would expect the probability of observing a difference as large as that for Guyana and its synthetic counterfactual for the 2010 to 2015 period to be 60% by chance alone. The annual treatment effects show that the Norway–Guyana REDD+ program had the strongest effect in 2014 and 2015 (Fig. 3C), with a 25% probability of observing a reduction in tree cover loss as large as the one observed for Guyana by chance alone compared to the other 3 in-space placebo comparison countries. The bootstrapped uncertainty estimates (95% confidence intervals), which includes all countries in the donor pool, also indicate a significant decline in Guyana’s tree cover loss during the Norway–Guyana REDD+ program relative to the counterfactual scenarios (SI Appendix, 5).
Leakage: Spillover Effects of the REDD+ Program.
To test for leakage, we conducted a geospatial analysis that employed the assumption that transboundary leakage of deforestation (negative spillover effects) would occur in forests closer to Guyana due to proximity (SI Appendix, 6). If there is elevated tree cover loss in the border region of Suriname, relative to its interior during the Norway–Guyana REDD+ program (2010 to 2015) compared to the period prior to the Norway–Guyana REDD+ program (2001 to 2009), this could be interpreted as evidence of deforestation leakage from Guyana. In our analysis, we found that tree cover loss along the border region of Suriname before the Norway–Guyana REDD+ program (2001 to 2009) was 0.020% y−1 (SE: 0.015) compared to 0.003% y−1 (SE: 0.001) in the interior of Suriname. Tree cover loss along the border region and the interior of Suriname during the Norway–Guyana REDD+ agreement (2010 to 2015) was 0.032% (SE: 0.023) y−1 and 0.011% (SE: 0.003) y−1, respectively (SI Appendix, 6). We found that tree cover loss along the border region with Guyana and the interior region of Suriname did increase during the Norway–Guyana REDD+ program, but coincided with a 250% increase in the price for an ounce of gold (SI Appendix, Fig. S9).
Permanence: Tree Cover Loss after the Norway–Guyana REDD+ Program Ended.
At the end of the Norway–Guyana REDD+ program in 2015, observed tree cover loss for Guyana increased by 200% from 0.069 to 0.140% in 2016. This increase in tree cover loss for Guyana is equivalent to the level of its synthetic counterfactual after the Norway–Guyana REDD+ program ended (Fig. 3B). In the 2 y after the REDD+ program ended (2016 and 2017), tree cover loss in Guyana also increased and remained above the 0.1% deforestation benchmark that would have triggered no REDD+ payments under the Norway–Guyana REDD+ agreement (SI Appendix, 1).
Discussion
Norway–Guyana REDD+ Additionality.
We found that tree cover loss increased in our focal country, Guyana, and all of the other comparison countries, classified as pretransition forest economies from 2001 to 2017 (Fig. 3A). However, we identified a 35% reduction in tree cover loss in Guyana (5,800 ha⋅y−1) from 2010 to 2015 during the Norway–Guyana REDD+ program (Fig. 3B). This reduction in forest loss is equivalent to 12.8 million tons of CO2 not emitted to the atmosphere over the program period, achieving the intended impact of national REDD+.
The sliding scale crediting baseline that penalized Guyana for increasing deforestation was effective at keeping Guyana’s tree cover loss below the 0.10% threshold, at which point Guyana would receive no REDD+ payments. Specifically, the results-based payments under the Norway–Guyana REDD+ program seemed to have attenuated the deforestation pressure associated with the international price of gold. This finding is particularly important to the Guiana Shield, where the production of gold from artisanal mining, which manifests in small-scale (<5-ha) deforestation patches, is the dominant driver of forest loss across the region (29⇓–31).
In the year 2012 when gold prices peaked at US $1,669/oz, tree cover loss increased by 71% in Suriname, a country that shares similar biophysical characteristics and deforestation drivers as Guyana, but only increased by 36% in Guyana (SI Appendix, Fig. S9). The Government of Guyana, as the owner of all subsurface mineral rights, directly controls the rate of deforestation by leases and incentives for commercial forest use activities such as mining. REDD+ financing, at a US$5 carbon price, surpasses the Government yields from gold, and as such, with REDD+ financing, the government is incentivized to reduce forest loss to capture more revenue for the national budget (32). We discuss the motivation of the Guyanese government to engage in REDD+ programs in more detail in SI Appendix, 2: REDD+ Theory of Change as a Climate Mitigation Strategy.
Aside from results-based payments, the Norway–Guyana REDD+ program also emphasized performance-enabling REDD+ activities (33). Specific emphasis and allocation of Norway–Guyana REDD+ payments were made to support institutional and technical capacity to achieve the Norway–Guyana REDD+ outcome of reduced emissions. There is evidence that this emphasis on these Norway–Guyana REDD+ enabling activities, classified as outputs in the REDD+ ToC (SI Appendix, 2), resulted in both regulatory and technological additionality that improved overall forest governance. For example, prior to 2010, Guyana was ranked the lowest for its national forest monitoring capabilities, but by 2015, Guyana was reporting at tier 3 and higher for carbon emissions assessment under the Intergovernmental Panel on Climate Change guidelines and had strong in-country remote sensing capacity for monitoring forest cover change (34). Concerning the legal forestry framework, Guyana was highly ranked for key policy features that are considered necessary for national REDD+ outcomes (35). Clearly defined tenure and use rights were included in Guyana’s forest legislation, policy, and governance framework. Guyana also had forest management regulations in place and enforced, relative to the other donor pool countries (SI Appendix, Table S2). The advancements in both regulatory forest governance framework and technological capacity to measure, monitor, and verify emissions reductions seem to have been important elements achieved under the Norway–Guyana REDD+ program. Improvements in forest governance quality such as those quantified for Guyana are seen as pivotal for the reduction in forest loss, biodiversity protection, and prevention of environmental degradation (36).
These results show that a national REDD+ program reduces forest loss. Such a rigorous national-level counterfactual analysis, to our knowledge, has not yet been conducted due to methodological challenges of building the counterfactual scenario (6). We acknowledge that the REDD+ policy arena at the international and country level is highly dynamic, with the REDD+ concept itself evolving since its inception (9). The availability of national REDD+ finance to countries in the donor pool could have influenced our comparative analysis and impact evaluation inference. To account for country-level dynamics around national REDD+ implementation, we report on forest governance indicators and funding support for national REDD+ programs, which draws off the ToC (SI Appendixes, 2 and 3). In this qualitative comparison, we note that Suriname and Gabon, which constituted the majority of the weights in the synthetic matching to build the counterfactual Guyana, ranked poorly across all metrics for national REDD+ enabling activities, including national REDD+ finance. The more recent national and project-level REDD+ activities in Gabon and Suriname indicate that these 2 control countries were less affected by national REDD+ interventions during the Norway–Guyana REDD+ program.
Norway–Guyana REDD+ Leakage.
Due to its proximity, similar ecological and geological features, and reliance on gold mining, Suriname would have served as the best match for Guyana for a simple qualitative comparison to evaluate the impact of the Norway–Guyana REDD+ program. The synthetic matching validates Suriname’s position as an appropriate comparison country for Guyana, with the largest weight assigned to Suriname to construct Guyana’s counterfactual scenario of deforestation rates (SI Appendix, 4). The synthetic counterfactual for Guyana was, however, extremely sensitive to the exclusion of Suriname, with no combination of the other HFLD comparison countries able to recover a tree cover loss time-series trajectory that matched the tree cover loss observed for Guyana in the pretreatment period (SI Appendix, 5). A concern in causal impact inference and the synthetic control method is the potential existence of leakage or spillover effects from the treated country to the countries used in the comparison (14, 25). The issue of leakage is especially a valid concern as Guyana shares a particularly porous border with Suriname, and the countries have close socioeconomic links that include being members of an economic trading block (i.e., Caricom).
Our geospatial approach to explore leakage is deficient in that it does not address the specific drivers of deforestation, which are difficult to assess due to both their complexity and confounding factors (25, 37, 38). Further quantification of cross-border leakage associated with the Norway–Guyana REDD+ program would require more in-depth economic analysis that focuses on transboundary investment flows, labor, and market effects, especially those related to the demand and supply of gold and policies implemented by other countries (see SI Appendix, 6 for additional details on leakage). Such an approach is warranted, considering that we found increases in tree cover loss in both the border and interior region in Suriname, but which coincided with a significant increase in gold prices. In sum, accounting for both negative spillovers (e.g., displacement of deforestation) and positive spillovers (e.g., other countries begin to implement stricter forest policies) is needed to ensure there is no overestimation or underestimation of the REDD+ policy effectiveness.
Norway–Guyana REDD+ Permanence.
The Norway–Guyana REDD+ program did not explicitly account for permanence, in that there is no requirement after the program for the areas that were not deforested to remain forested after the agreement ended (33). We observe that Guyana’s deforestation rate increased above the 0.10% threshold immediately after the REDD+ program ended in 2016 (Fig. 3B). Under the performance clauses in the contract, a deforestation rate of 0.10% would have stopped payments during the program, but because it occurred after the program timeline, payments were not altered. This outcome suggests that policymakers need stronger permanence clauses in national REDD+ contracts, especially to guard against socioeconomic shocks such as higher prices of commodities that drive deforestation. We illustrate this point by calculating the effective costs of the carbon benefit over the 2010 to 2015 period. Based on assumed payments of US$250 million, the cost of the 12.8 million tons of avoided CO2 emissions is US$19.53 per ton of CO2. If these emission reductions are permanent, then this (US $19.53) is the effective cost per ton of CO2. However, if this emission is only avoided for 5 y, then the rental rate at a 5% discount rate is US $4.50 per ton of CO2 per year, and the effective cost of carbon is around US$90 per ton of CO2.
Moving forward, we will have additional years of the forest-loss time-series data from the Global Forest Change dataset, which will enable us to quantify permanence better. A major advancement provided by our application of the synthetic matching is the construction of a counterfactual time-series trajectory, which is well suited to the task of assessing permanence outcomes from forest protection initiatives. Such long-term data will also provide information on how shocks, such as those related to commodity values, and other sociopolitical events, such as government elections, impact forest loss rates.
Synthetic Matching and Building a Time-Series Trajectory.
There is a tremendous challenge to quantify the impact of policies implemented at an aggregate level with a single treated unit, particularly when that treated unit is a country (14). Synthetic matching offers a systematic and transparent way to overcome this challenge because it builds credible estimates of counterfactuals in order to estimate the treatment effect and critically evaluates those results in a quantitative manner (13, 17). Synthetic matching removes the subjectivity of the analyst in selecting the best-suited comparison unit, and it enables us to overcome the limitations of other matching methods (18, 19). Through matching on the outcome of interest, it explicitly accounts for unobserved or unknown factors that might influence the outcome.
The use of annual time-series data in the synthetic matching process also improves upon before-after-control-impact analysis methods that rarely account for interannual variability. The counterfactual trajectory produced by synthetic matching enables us to assess thoroughly the quality of the counterfactual comparison at multiple points in time before, during, and after the intervention. To further improve transparency, synthetic matching provides explicit weights for both the comparison countries and covariates, thus allowing for a critical assessment of the counterfactual and the overall quality of the comparison based on expert knowledge.
Irrespective of these advances, finding appropriate counterfactuals for programs implemented at large aggregate scales like the country level is still a challenge, as seen by the limited number of donor pool countries, and the synthetic matching outputs (SI Appendixes, 3 and 4). The low number of appropriate comparative units affects the strength of inferences that can be made, as is evident by the 3 in-space placebo comparison countries in our analysis (Gabon, Republic of Congo, and Suriname), and the decreasing counterfactual fit with the exclusion of Suriname (SI Appendix, 5). To overcome this statistical sample size challenge, we employed a parametric bootstrap procedure that resamples the residuals of the counterfactuals to obtain the uncertainty estimates of the policy effect (SI Appendix, 5).
Conclusions
Targeted efforts at reducing deforestation in developing countries have received substantial support as evident by unprecedented funding commitments (4), country-level efforts to characterize deforestation drivers (12), and the large number of national and project-level REDD+ pilots being implemented (5). These initiatives are important because tropical forest loss has increased globally with the most significant loss in tropical forests recorded in 2016 and 2017, even in countries with historically low deforestation rates with the majority of intact forest landscapes (39, 40). Results such as these based off of rigorous impact evaluation of climate mitigation programs will help countries refine their climate action plans post-2020 (6).
Overall, we find strong evidence in our analysis that the Norway–Guyana REDD+ payments were effective in reducing tree cover loss in a country with historically low deforestation rates. This result is proof of emissions reduction that would not have occurred without the Norway–Guyana REDD+ program and thus satisfies the additionality pillar of REDD+. Our analysis also provides important insights into how carbon crediting baselines for national REDD+ can be effective, the establishment of which has stymied the progress of many other national REDD+ programs (23). The application of a sliding scale deforestation reference baseline, with increasing penalties as deforestation rises seems to have empowered both the buyer and provider of carbon credits. The use of a flexible reward scheme potentially bodes well for a forest carbon market that can respond to demand and supply relative to other commodities (32, 41).
Leakage and permanence are more challenging to quantify in the context of national REDD+ as an international climate mitigation strategy. The difficulty of accounting for leakage is a result of its complexity and finding direct causal links, especially as the drivers of deforestation are global in scope and associated with global commodity trade and investment flows (24, 38, 42). The fragmentary implementation of national and project-level REDD+ across forested countries leaves room for the displacement of deforestation from early adopters like Guyana that have developed rigorous forest carbon regulatory and governance systems to countries that have not engaged in REDD+ and that have weaker regulatory policies (43). A multinational REDD+ approach across all of the Guiana Shield biome countries that includes political cooperation and harmonizes forest governance and deforestation regulations would address the issue of leakage (Fig. 2) (24, 44, 45). The fact that the observed tree cover loss in Guyana climbs above the 0.10 threshold after the Norway–Guyana REDD+ program ended indicates that continued forest protection is dependent on sustained conditional national REDD+ payments, and stronger permanence clauses.
Materials and Methods
Deforestation Data.
We utilized the Global Forest Change dataset spanning 2000 to 2017, which measures percent tree cover in 2000, tree cover loss, and tree cover gain at 30-m spatial resolution based on earth observation data, primarily Landsat imagery (22). The percent tree cover in the 2000 layer represents the canopy cover of each 30-m pixel in the baseline year of 2000. Each pixel is classified from 0 to 100, with 0 being no canopy cover and 100 being 100% canopy cover. For this analysis, we considered a pixel ≥30% canopy cover as forested, which is consistent with Guyana’s and other national submissions to the UNFCCC REDD+ program (46). Tree cover loss is defined as a stand-replacing disturbance or a change from a forest to nonforest and is measured annually from 2001 to 2017. The Global Forest Change dataset provides us with a time-series trajectory of annual tree cover loss that includes 9 y of data preintervention (2001 to 2010), 5 y during program implementation (2010 to 2015), and 2 y after the end of the program (2015 to 2017).
The tree cover loss (percentage) was calculated as
Construction of the Counterfactual Time Series Using the Synthetic Control Method.
Evaluation of policy outcomes where a single aggregate unit like a country is exposed to an intervention and the policy target is determined for the country as a whole is challenging, in part because it is difficult to decide what the comparative units are (i.e., counterfactuals; what would have occurred in the absence of the intervention). In many cases where projects to reduce deforestation are examined such as the establishment of protected areas, the control group is easier to determine, particularly if it is derived from within the same country (17, 49, 50). In our case, the policy target is the rate of forest loss for an entire country, Guyana, and we choose as the control group other countries that are similar to Guyana based on the underlying structural processes related to forest loss and forest cover (SI Appendix, 2). We adopt the synthetic control method (hereafter “synthetic matching”) from the field of economics and political science for comparative case studies (13, 14, 17). Synthetic matching provides a systematic and quantitative means to construct counterfactuals and removes the subjectivity of identifying appropriate comparison units. As there may never be a perfect match to the treated unit, synthetic matching builds the counterfactual scenario based on a weighted combination of potential comparison units that approximates the characteristics of the treated unit. We apply the synthetic control method to reproduce the tree cover loss that would have been observed for Guyana in the absence of the Norway–Guyana REDD+ program.
In the selection of our donor pool countries (i.e., comparison countries used to build the counterfactual of Guyana without the national REDD+ intervention), it was important to use only those countries that share the same structural processes that influence forest loss and not subject to structural shocks during the program intervention (13). Thus, we restricted the donor pool to pretransition forest economies with HFLD countries identified by Hosonuma et al. (12) (see SI Appendix, 3: Identification of Donor Pool Countries). Based on these criteria, our donor pool includes 6 countries: Colombia, Republic of Congo (Congo), Democratic Republic of Congo, Gabon, Peru, and Suriname (10, 11). Pretransition economic status is based on the forest transition model, which posits that countries at different stages of economic development are subject to different drivers of deforestation (12, 26). All of our comparison countries had also submitted Readiness Preparation Proposals (R-PP), a preliminary requirement for the establishment of a national REDD+ financing mechanism under the World Bank’s Forest Carbon Partnership Facility. To account for the country level REDD+ dynamics in the donor pool countries, such as access to national REDD+ funding, we report on metrics related to the ToC for national REDD+ (SI Appendixes, 2 and 3).
As covariates in the synthetic matching, we include annual time series (2001 to 2017) of proximate drivers of forest loss for pretransition economies identified by Hosonuma et al. (12) based on the country level R-PPs as well as other national-level socioeconomic pressures that relate to forest loss identified by other studies (29, 36, 41, 51). These include mining (mineral rents), forest utilization (forest rents), agricultural expansion, population growth, gross domestic product (GDP), governance quality, and land area under some form of protection (SI Appendix, 4). Annual data on agricultural expansion (percentage of land area), population growth (percentage), GDP growth (percentage), forest rents (percentage of GDP), and mineral rents (percentage of GDP) were extracted from the World Bank Statistical Database (52). Land area under some form of protection was extracted from the World Database on Protected Areas (53) and governance effectiveness from the World Governance dataset (54). In cases where there were gaps in the annual time series of the covariates, we impute these values using a structural modeling approach fitted by maximum likelihood (55).
To construct the synthetic counterfactual, synthetic matching assigns weights to 1) each comparison country and 2) each covariate, based on the similarity between both the outcome of interest and the covariates for the country that was exposed to the policy intervention and the comparison countries that were not exposed to the policy intervention (14). The synthetic counterfactual is achieved by minimizing the mean-squared prediction error between the outcome for the treated country and the synthetic counterfactual summed over all of the years in the pretreatment period (SI Appendix, 4). The counterfactual should thus match the observed pretreatment deforestation time-series trajectory as the treated unit. The effect of the policy intervention is then measured as the difference between the observed outcome from the treated unit and the outcome from the synthetic counterfactual for the period the project was implemented. We implement the synthetic control method with the Synth package in the open statistical software R (16). We interpret the synthetic counterfactual for the Norway–Guyana REDD+ in terms of additionality, leakage, and permanence. We report our results from the synthetic matching process (weights assigned to countries) as well as the placebo tests and sensitivity analysis conducted to test the quality of the synthetic matching process in SI Appendix, 5.
ToC.
As the Norway–Guyana REDD+ program did not have an explicit ToC to guide the impact evaluation, we adapt a generic ToC based on the international policy discourse for REDD+ as a climate mitigation strategy (56). We modify this generic REDD+ ToC based on the Norway–Guyana REDD+ agreement to understand why an intervention such as the Norway–Guyana REDD+ program is expected to contribute to climate mitigation using conditional financial payments for forest protection (57) (SI Appendix, 2). The ToC maps the path to achieve the outcome of emissions reductions under the national REDD+ program (e.g., capacity and institutional strengthening needed for a national REDD+ forest governance framework). We quantify progress related to these activities and associated outputs related to the ToC for Guyana and all other donor pool countries in SI Appendix, Table S2.
Acknowledgments
We express our gratitude to the researchers and data analysts who have compiled the multiple global datasets and statistics that we employed in our analysis. We also thank Vitor Possebom for providing comments on preliminary drafts of the manuscript and University of Florida REDD+ Working Group for comments on the REDD+ conceptual diagram.
Footnotes
- ↵1To whom correspondence may be addressed. Email: anandroopsind{at}boisestate.edu.
Author contributions: A.R. designed research; A.R. performed research; A.R. analyzed data; and A.R., B.S., and J.B. wrote the paper.
The authors declare no competing interest.
This article is a PNAS Direct Submission.
Data deposition: All datasets reported in this paper are open access and described in Methods.
↵*HFLD countries are classified as countries with more than 50% of historical forest cover remaining with deforestation rates less than the global average of 0.22% during the reference period of 1990–2000.
↵†Throughout the text, we refer to “national REDD+ program” as efforts to incentivize governments to change future trends in deforestation through implementation of country-wide programs. In contrast, “project-level REDD+” refers to efforts to change deforestation in specific projects that are undertaken in part of a country.
This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1904027116/-/DCSupplemental.
Published under the PNAS license.
References
Evidence that a national REDD+ program reduces tree cover loss and carbon emissions in a high forest cover, low deforestation country
Sign up for Article Alerts
Article Classifications
- Social Sciences
- Sustainability Science
Jump to section
You May Also be Interested in
Better explicating the strengths and shortcomings of these models will help refine projections and improve transparency in the years ahead.
Image credit: Witsawat.S.
Using CRISPR, researchers showed that a region some used to label “junk DNA” has a major role in a rare genetic disorder.
Image credit: Nathan Devery.
Mara Reed and Michael Manga explore why Yellowstone's Steamboat Geyser resumed erupting in 2018.
Some songbird parents might improve their own fitness by manipulating their offspring into leaving the nest early, at the cost of fledgling survival, a study finds.
Image credit: Gil Eckrich (photographer).