Using payment for ecosystem services to meet national reforestation commitments: impacts of 20+ years of forestry incentives in Guatemala

International environmental initiatives, such as the Bonn Challenge and the UN Decade on Restoration, have prompted countries to put the management and restoration of forest landscapes at the center of their land use and climate policies. To support these goals, many governments are promoting forest landscape restoration and management through financial forestry incentives, a form of payment for ecosystem services. Since 1996, Guatemala has implemented a series of forestry incentives that promote active forest landscape restoration and management on private and communal lands. These programs have been widely hailed as a success with nearly 600 000 ha enrolled since 1998. However, there has been no systematic assessment of the effectiveness of these programs on preserving and restoring Guatemalan forests. This study evaluates the impacts of over 16 000 individual PES projects funded through two incentive programs using a synthetic control counterfactual. Overall, a program for smallholders resulted in lower rates of forest loss, while a program for industrial timber owners led to greater gains in forest cover. Across policies, we found dramatically higher forest cover increases from restoration projects (15% forest cover increase) compared to plantation and agroforestry projects (3%–6% increase in forest cover). Projects that protected natural forest also showed a 6% reduction in forest loss. We found forest cover increases to be under 10% of total enrolled area, although positive local spillovers suggest this is an underestimate. Restoration projects show the most promise at promoting forest landscape restoration, but these benefits need to be weighed against priorities like resilience and rural development, which may be better served by other projects.


Introduction
For at least the past decade, forest restoration has been portrayed as an essential part of climate change mitigation efforts [1], as it could provide enormous nearterm benefits in carbon sequestration [2] and numerous benefits to ecosystems and local communities.Guatemala, like many countries around the world, has set ambitious targets for reducing deforestation and restoring forest cover.Under the Bonn Challenge, Guatemala has committed to restore 1200 000 ha of forest, around 11% of its total land area, by 2030.Guatemala has experienced rapid deforestation in recent decades [3] leading some to call into question the efficacy of many Bonn Challenge commitments [4], but forest restoration remains a key priority for Guatemala.

Guatemala's forestry incentives
To meet these ambitious goals, sound landscape governance is needed to engage stakeholders and to provide the infrastructure as well as capacity to effectively restore and sustainably manage forests [5].Guatemala's long-standing forestry incentive programs, which have operated since the 1990s, have provided critical infrastructure and expertise to implement forest landscape restoration.PINFOR (Programa de Incentivos Forestales) was instated by the 1996 forestry law and incentivized landowners to improve forest cover and kickstart the nascent timber industry in Guatemala [6].PINFOR primarily focused on productive systems, promoting planting and management of high-value trees.It proved highly popular, with enrolled acreage and projects increasing tenfold between 1998 and 2016 [7].However, nearly all smallholders were ineligible for the PINFOR program due to minimum size and land tenure requirements, so a parallel program called PINPEP (Programa de Incentivos Forestales para Poseedores de Pequeñas) was launched by the Dutch development bank in 2006 and funded by the Guatemalan Government in 2010, propelled by the grassroots organization of indigenous and smallholder groups [8].Smallholder priorities were built into PINPEP's design, with significant funding allocated to agroforestry and fire prevention, which were priorities for many of the indigenous and smallholder groups who championed it [8,9].Alongside these programs, the Guatemalan National Institute of Forests (INAB) created a national forest governance apparatus that introducted strict forest regulations, developed a network of forestry technicians, and built systems to track and disburse funds to participating landowners [6].Together, the PINPEP and PINFOR programs enrolled 588 276 ha between 1998 and 2019, approximately half of the area committed to forest restoration through the Bonn Challenge, and almost all of the restored forest area in Guatemala's national commitment comes from their forest incentive programs [7].Furthermore, Guatemala's Nationally Determined Contributions from the UNFCCC Paris Climate Agreement rely on these same programs for carbon sequestration via forest expansion [10].
Guatemala's forestry incentives have been hailed by politicians and policymakers as a success [8], and some studies have mentioned localized successes for PINFOR [11,12].However, there has not been a systematic analysis of how these programs have impacted forest cover in Guatemala.Meanwhile, several high-profile studies have questioned the efficacy of popular forest incentive programs.Just to the north, Mexico's billion-dollar Sembrando Vida may be accelerating forest loss [13].REDD+ continues to face challenges, with several recent studies finding mixed project success and overstated benefits [14][15][16].Without a greater understanding of impacts and trade-offs across projects that support forest landscape restoration and management, maintaining the momentum and funding needed to achieve Bonn Challenge and the UN Decade on Restoration goals will be challenging.
In this study, we investigate the outcomes and trade-offs of forestry incentive program approaches in Guatemala.We addressed the following questions: (1) to what degree did PINPEP and PINFOR contribute to forest cover changes in Guatemala; (2) how do different project types (e.g.natural forest management, plantations, restoration) impact forest cover; (3) do project impacts persist after unenrolment; and (4) are there any local spillovers, or effects of enrolment on surrounding areas?

Study area
Guatemala has a diverse range of ecosystems, with wet tropical lowlands in the north and along the coasts, pine-oak forests and a cooler, wet climate across the country's central highlands, and dry tropical regions in the Pacific Coast and Motagua Valley [17].As referenced earlier, Guatemala has experienced rapid deforestation in recent years, primarily due to agricultural expansion [18].Guatemala's deforestation rate hovered around 2% in the 1990s [19] but declined to around 1% in the late-2000s [20].Rapid deforestation in the 1990s was attributed to ranching and smallholder farming [19] but recent scholarship suggest narcotrafficking [21] and expansion of export crops [22] are driving much of the current forest loss.Some research also suggests that migration is driving agricultural expansion through the re-investment of remittances into smallholder farming [23,24] and that Guatemala's forestry incentive programs might be responsible for the declining rates of forest loss [24].
We analyzed forest cover trends for Guatemalan sites enrolled in PINFOR and PINPEP programs between 1999 and 2018 (figure 1).Data on enrolled sites was gathered via the INAB's GIS portal, which provided data for over 50 000 PINPEP and PINFOR projects.Sites were filtered to exclude uninitiated projects, duplicates, and those lacking project type information.Project area was estimated based on available metadata, which varied between PINPEP and PINFOR projects (SI appendix 1).Ownership types for assessed PINFOR is quite diverse, with projects spread across individuals, businesses, cooperatives, municipalities, NGOs, and government organizations (SI figure 1).PINPEP property owners were primarily individuals (SI figure 2).
PINPEP and PINFOR allow landowners to enroll in one of five types of projects, each with different goals, management, and payment schemes.PINPEP supports tree planting through agroforestry or plantations, and natural forest management for protection of natural resources (NFM Protection) or timber production (NFM Production).PINFOR offers the same projects with the exception of agroforestry, instead paying for forest restoration.Each project requires that a certified forestry technician design a management plan that is then approved by INAB.NFM Production projects were the least common (less than 200 projects for PINPEP or PINFOR), but Basic information on project sites included in this analysis, broken out by program and project type.We include a measure of mean years enrolled for projects that began before 2010 and ran to or near completion, as projects that began after 2010 may be ongoing.Total area and total USD spent were reported in Custodio De León [7] and cover all funded PINPEP and PINFOR projects from 1998 to 2019.
we included over a thousand sites for every other project type in the analysis.On average, PINPEP projects in the analysis lasted 1.8 years and PINFOR projects lasted 4.7 years, although this underestimates the actual enrolled period (table 1).PINPEP and PINFOR sites were on average wetter, at higher elevation, and initially had more forest cover than unenrolled areas (SI table 1).

Data description
In order to account for factors relevant to forest landscape restoration success, we identified gridded datasets to track terrain, climate, and human influence.Terrain features comprising slope, elevation, and aspect, were derived from SRTM v4 elevation data [25], with slope and aspect calculated using a 3 × 3 moving window with bilinear resampling.We included yearly population density from WorldPop, which provides population estimates at approximately 100 m resolution [26] and travel time to cities with a population larger than 20 000 [27] to estimate human influence.We used distance to rivers and streams, derived from the HydroSHEDS dataset [28], as a proxy for water availability and accessibility.To control for climate, we included historical measures of yearly precipitation, precipitation of the driest month, and mean annual temperature from WorldClim [29].Coordinate location was also recorded for each site to measure geographic spread of projects.
Additionally, we gathered data on forest extent, loss and structure to estimate project impacts on forest extent and density.To calculate a pre-treatment baseline, we used above and belowground biomass in the year 2010 [30], 30 m resolution forest cover maps for 2001, 2006 and 2010 provided by partners at Universidad del Valle de Guatemala (UVG) [20], percent forest cover in 2000, yearly forest loss events from 2001-2010 and cumulative forest gain 1 Guatemalan Quetzals converted to USD using an exchange rate of 0.1275.Reported without inflation adjustment.from 2000-2012 from Global Forest Watch (GFW) [31], and forest extent and height in 2000 from the Global Land Analysis and Discovery lab (GLAD) [32].We also included distance to intact forest using the 2010 UVG forest cover data, with intact forest defined as 6 or more contiguous pixels of forest.To measure outcomes, we used yearly forest loss events from 2011-2020 from GFW [31] and forest height and extent change between 2000 and 2020 from GLAD [32].We used GFW forest loss events to better isolate the impacts of natural forest projects, which work to conserve and manage existing forests, and used GLAD forest extent and height to evaluate planted forest projects that may experience periodic harvesting.

Program impact estimation
We estimated the impact of PINPEP and PINFOR on forest cover and forest loss using synthetic controls.For a valid impact analysis, we need to isolate the effect of enrollment from other factors that could influence forest cover.Accurately estimating impact can be complicated, as programs are rarely unbiased in their assignment of treatments, unless they are randomized.There are a few standard methods to retrospectively isolate the impacts of treatment on outcomes like forest cover, but counterfactuals are generally seen as more accurate and conservative than traditional analyses [33].We estimate impacts by generating a counterfactual using the synthetic controls methodology developed by Robbins et al [34].Synthetic controls were first introduced by Abadie & Gardeazabal [35] and Abadie et al [36], and have rapidly grown in popularity in the intervening decades because of their transparency and interpretability.Robbins et al [34] extend the synthetic control methodology to interventions with many treatment units by creating a single pooled synthetic control for a large treatment group.This provides a flexible framework for impact evaluation that can be easily applied to large programs like PINPEP and PINFOR.There are alternative specifications for this estimation, including matching with difference-in-differences, which we compare with the synthetic control results in SI appendix 2.
Using synthetic controls requires a large group of untreated sites which should be generally comparable to the treated sites.In this case, treated sites were projects that were enrolled in either PINPEP or PINFOR.Untreated sites were created by randomly generating 100 000 3.1 ha areas across Guatemala and filtering for overlap with enrolled sites (figure 1).PINPEP and PINFOR projects that began before 2010 were removed so that all projects began between 2010 and 2018.We then created the synthetic control by assigning a weight to each untreated unit, such that the weighted untreated units matched the treated units in all covariates and baseline variables.This allowed us to measure the effect as the percent difference between the treated group and synthetic control.Standard errors were calculated using survey methods to account for the weights of the synthetic controls [34].

Project type impact estimation
We estimated the impact of the five project types within PINPEP or PINFOR using the microsynth R package provided by Robbins and Davenport [37] to replicate their method for applying synthetic controls to multiple treatment units.In this case, a separate synthetic control was generated for each program and project type combination, for eight synthetic control runs in all.For PINFOR sites enrolled in Restoration projects, we included additional control sites within a high plateau region in Western Guatemala called the Sierra de los Chuchumatanes, where many of the projects were located, to get a more representative set of untreated sites (SI appendix 2).Importantly, the relevant outcomes vary across project types.We used forest cover loss as an outcome for projects that manage natural forest, as these projects should be maintaining existing forest relative to the control.Alternately, we use changes in forest cover and canopy height as outcomes for tree planting projects, as these seek to expand forest extent.

Durability analysis
We evaluated the persistence of forest outcomes after project completion by conducting event study and simple regression analyses on project unenrollment.Program impacts can be short-lived if landowners clear the forest or stop beneficial practices after the project ends.Testing the impact of unenrollment allowed us to determine if clearing is occurring soon after payments stop.We used an event study analysis, which estimates the impact of a treatment on a timevariant outcome [38], to determine if clearing occurs shortly after payments stop.Event studies provide a good estimation of treatment effects on outcome trends in short time windows, with increasingly difficult assumptions over longer timespans.The model specification is described in SI appendix 3. We also regressed project end date on forest outcomes to see if there is an impact of unenrollment date on forest outcomes.This allowed us to test if projects that ended earlier are seeing less improvements in forest cover than projects that ended later.Both analyses control for baseline forest conditions, climate, and population and included year, region, and project type fixed effects.

Spillover analysis
We investigated local spillovers by examining changes in forest outcomes in the areas surrounding projects over the study period.Spillovers are effects that are seen outside of a project's boundaries due to projectinduced changes in markets, landowner behavior, or other factors, and often lead to unaccounted benefits or harms [39].We estimated the impacts of enrollment on forest extent and height in areas directly adjacent to enrolled sites (local land use spillovers).Adjacent areas were generated by taking a buffer around each treated site equal to that site's radius.To estimate the impact of treatment on the non-treated buffer area, we regressed forest cover change and forest height change on site type (PINPEP, PINFOR, or respective buffers) controlling for initial forest conditions, climate, and 2010 population.This approach allowed us to identify spillovers at the local level, amongst landowners and neighbors, but program spillovers that occur at a larger scale, such as changes in wood product markets or the creation of a national forestry technician network, were not assessed in this analysis.

Robustness checks
We assessed the robustness of the synthetic control results with untreated 'placebo' synthetic control runs to assess the variation in potential non-treated forest trajectories and provide a confidence interval for our data.Placebos were generated using random selections of non-treated units.Each of the randomly selected groups was weighted to match the treated group using the same method that created the initial synthetic control.Then, the outcomes were compared across the treatment period to get a measure of the variation in non-treated outcomes.Because of the number of sites included in this analysis, we limited the number of placebo runs to 200, but we were still able to find near-perfect fits for all placebo runs.We used the placebo runs to generate 95% confidence intervals, which we compared to the synthetic control standard errors.

Program-wide results and comparison
Using synthetic control methods, we estimated the impact of enrollment in either the PINPEP or PINFOR program on percent change in GLAD forest cover and forest height from 2000 to 2020 and GFW forest loss in the years after enrollment.Forest cover increased at both PINFOR and PINPEP sites relative to the control, with 8.7% and 3.2% increases, respectively (table 2).PINFOR sites saw a 5.7% increase in forest height relative to the control, while PINPEP had no significant effect.Forest loss showed diverging trends, with PINFOR sites experiencing a 1.6% increase in percent forest cover loss and PINPEP sites seeing a 3.4% decline.

Impacts by project type
We evaluated the impacts of the five different project types under PINPEP and PINFOR on forest cover, height and loss using the same counterfactual approach.PINFOR Restoration projects showed the largest gains in forest cover, with forest cover increasing 15% as compared to the control (table 2).Tree planting projects (Agroforestry, Plantation and Restoration) all experienced increases in forest cover relative to the control, with PINFOR outperforming PINPEP.The only tree planting project type to see a change in forest height was PINFOR Restoration, which increased in height 12.5% relative to the control.Tree planting projects generally had slight increases in forest loss relative to the control.We found diverging trends for natural forest management projects, with projects focused on forest protection seeing decreases in forest loss and projects focused on production experiencing increases in forest loss relative to the control, although this trend was not significant for PINFOR NFM Protection when using standard errors (figure 2).

Durability
When investigating the impact of project unenrollment on forest loss events to determine whether project benefits last past a project's end date, we found that forest loss events decreased in the years following unenrollment for PINPEP sites but that unenrollment otherwise had no effect on forest loss (SI figures 3 and 4).Our results also show that longer-running PINFOR projects saw greater gains in forest height from 2000 to 2020 (SI table 3) than PINPEP projects and that PINFOR and PINPEP projects that ended more recently showed less change in forest height by 2020.PINFOR projects that ended more recently also had less change in forest cover by 2020 (SI table 4).

Local spillovers
To investigate local spillovers from projects, we compared forest outcomes for enrolled sites, buffer areas, and the untreated Guatemalan average.The buffer areas of PINPEP and PINFOR sites experienced much higher forest cover and forest height change than the Guatemalan average, but slightly lower change than the treated sites.We observed increases in forest cover in both project and buffer areas, but forest height increases were only detected in PINFOR sites and buffer areas.In a regression analysis, areas around PINPEP sites showed higher gains in forest cover and height compared to those around PINFOR sites, which experienced significant but more modest increases (SI table 5).

Overall program effectiveness
The evaluated incentive programs consistently increased forest cover, though absolute changes have been modest.The size of these impacts is comparable to other studies of financial forest incentive programs that have used similar counterfactual approaches [40][41][42][43].Additionally, we found substantial local spillover effects that had large, positive impacts on forest height and forest cover (SI table 5), suggesting that our synthetic control estimates underestimate total project benefits.
We also found that, while PINPEP led to smaller forest cover gains, enrollment in the smallholderaligned program showed a decline in forest loss while forest loss increased in PINFOR sites.PINFOR generally supported productive forests, so while overall forest extent may have increased these areas are likely experiencing harvest cycles that produce a signal in the yearly forest loss data.PINPEP aligned heavily on smallholder priorities, meaning that many of the projects have a mix of useful trees which may be left longer than plantation rotations, and would be harvested on different cycles, leading to little to no deforestation signal.

Project type effectiveness
When we compared the outcomes of project types across programs, restoration projects, which were restricted to PINFOR sites, were clearly the most effective at increasing forest cover and forest height.Restoration had a far greater impact on forest cover than plantations and agroforestry, which also seek to increase forest cover through tree planting.Incentives that promote agroforestry or plantation projects may not be creating as much additional forest cover because the sites were already mostly forested before enrollment (46% for Restoration vs 78% for PINFOR Plantation), suggesting that restoration projects resulted in substantial additional forest cover by targeting non-forested areas and successfully restoring forest cover in these areas.NFM Protection sites showed declines in forest loss of around 6%, although this change was only shown to be significant for PINPEP projects.These reductions in forest loss suggest that PINPEP's smallholder projects are better at increasing the stability of natural forest cover.Additionally, PINFOR and PINPEP NFM Production sites experienced increases in forest loss after enrollment, likely because these projects are supporting the active management (i.e.harvests) in natural forests.

Project durability
When analyzing the durability of project benefits, we found no sign that programs were losing forest cover after unenrollment.Our event study results show that PINPEP sites experience less forest loss events after unenrollment, while our analysis of postproject impacts suggested that PINFOR sites continue to grow in forest cover after unenrollment.These results indicate that projects benefits are generally maintained for at least a few years after payments end.
While these results suggest that tree cover benefits continue for the first few years after unenrollment, durability is an ongoing concern for forests in Central America.Reid et al [44] studied longterm forest change in an area of southern Costa Rica and found that 50% of secondary forests were re-cleared in 20 years and 85% were re-cleared in 54 years.Rates of forest re-clearance in Guatemala have not been estimated, but clearance and regrowth has been a longstanding feature of forest landscapes in the country [45,46].Promoting projects that align with landowner objectives, such as planting productive tree species, improving water quality, and reducing fire risk could help improve the longterm stability of forests restored or managed through PES programs.

Data limitations and program challenges
Data for this analysis were collated from many sources, and it is important to keep in mind the limitations of these data in estimating program impacts.Project areas were estimated using the best available metadata, but direct comparison of this dataset with publicly visible polygon data showed that these area estimations were inaccurate for irregularly shaped projects (SI figure 1).PINFOR estimated areas are overall less accurate and therefore may muddle treatment signals with spillover signals, based on the variations of size, shape and location of projects.Additionally, two of our major outcome variables only have measurements for 2000 and 2020.We used a number of datasets to control for baseline forest change from 2000 to 2010, and found that forest cover in 2010 correlated well with GLAD forest cover data.However, this may not have controlled for all changes in this data across the pre-treatment period.If our counterfactual had a bias in low-canopy forest cover change between 2000 and 2010 compared to the treatment, our forest cover and height results may be biased, as we were not able to account for this in our baseline data.
Researchers have noted the difficulty of running large incentive programs at a national scale.Political turnover, ongoing financial obligations, and administrative overhead all challenge program longevity [47,48], and these programs are no exception.Over the years, several issues in program management and administration have emerged.For example, PINPEP, which funds many small projects, has struggled to develop a robust infrastructure for project and landowner tracking.Many program participants report not receiving payments from the government, and many have had to travel to the capital to protest before receiving their payments [8].Additionally, PINFOR was designed primarily for forest production, focusing on high-value tree species and less on ecological services [8].Another program called PROBOSQUE that focuses more explicitly on ecosystem services replaced PINFOR in 2016 after an updated version of the forestry law was passed, although many PINFOR participants have continued receiving benefits under the new program.

Conclusion
Forest incentive programs underlie ambitious goals of large-scale landscape transformation but require clear evaluation to understand whether these goals can be met.While PINPEP and PINFOR combined have treated nearly 600 000 ha, the realistic additional forest cover these programs delivered is a small fraction of that amount.However, these programs still demonstrate clear success when compared to similar financial forest incentives.Restoration projects showed by far the most success at adding additional forest cover to the landscape, in part because these projects best targeted areas without existing forest cover.Ensuring that projects are sited on previously deforested or degraded lands is critical to ensuring the success of programs in providing additional forest cover.However, incentive programs aimed at achieving multiple benefits need to weigh the trade-offs between forest cover expansion and other projects that support local economies and ecosystems with more modest forest cover benefits.Overall, this analysis shows that large-scale forest incentives provide real benefits to forest cover, but that governments need to better target projects, dramatically increase their acreage, and more seriously consider land management trade-offs if they are to reach the ambitious goals they have set for forest landscape restoration.are shareholders in the company and thus stand to benefit financially from forest management targeted at climate change mitigation

Figure 1 .
Figure 1.Guatemalan forestry incentive locations.The distribution of PINPEP (red) and PINFOR (blue) projects across Guatemala.Non-treated sites used to calculate the counterfactual are shown in grey.
Outputs of the counterfactual analysis are displayed as the percent difference between the treatment and control, followed by the p-value in parentheses.Values in bold have a p-value of less than 0.05.Forest loss is measured as cumulative yearly difference from the control after treatment begins, whereas forest cover and forest height change are measured as the difference between 2000 and 2020 controlling for all forest covariates between 2000 and 2010.Values found insignificant using the confidence interval generated via the placebo runs are marked with ǂ.

Figure 2 .
Figure 2.The impact of natural forest management vs tree planting projects on GFW forest loss.Cumulative percentage of project area forest loss (5% means forest loss was recorded across an area equivalent to 5% of all sites) relative to the control from 2010-2020 for all project types.Graphs are broken up by natural forest management projects and tree planting projects.NFM Protection and PINPEP Plantation sites saw the greatest reductions in forest loss, while NFM Production and PINFOR plantation sites experienced increases in forest loss, relative to the control.

Table 1 .
Program statistics for PINPEP and PINFOR project types.

Table 2 .
Percent change between treated and counterfactual by program and project type.