Estimating the costs of reducing forest emissions A review of methods

CIFOR CIFOR International Institute for Environment and Development Center for International Forestry Research (CIFOR) CIFOR advances human wellbeing, environmental conservation, and equity by conducting research to inform policies and practices that a ect forests in developing countries. CIFOR is one of 15 centres within the Consultative Group on International Agricultural Research (CGIAR). CIFOR’s headquarters are in Bogor, Indonesia. It also has o ces in Asia, Africa and South America. CIFOR works in over 30 countries worldwide and has links with researchers in 50 international, regional and national organisations.


4.1
Alternative This paper reviews main approaches to estimate the costs of REDD, with a focus on the opportunity costs.These can be classified into local-empirical, global-empirical and global simulation models.In local-empirical models, per-area opportunity cost estimates are derived from detailed studies (surveys) and carbon density estimates -both specific to the particular area studied.Global-empirical models use local-empirical estimates, aggregate these to global per-area costs of reducing deforestation, and use uniform values of carbon density (ton/ha) to obtain a single, global estimate of opportunity costs ($/tCO2eq).Global partial equilibrium models simulate the dynamics of the world economy to estimate supply of REDD services (or avoided deforestation), represented in supply curves.
One striking observation is that the cost of REDD differs substantially across model approaches: global simulation models yield far higher REDD prices than empirical models, including the Stern estimate.The 'true' cost estimate is most likely to lie somewhere in between the values provided by the local-empirical models on the one hand (lower end) and global simulation models on the other (higher end).
The following messages arise from this review: Alternative cost assessment methods suit different needs Each method has its own advantages and disadvantages: for an indication of total costs of REDD as a climate change mitigation option compared to alternatives, the use of global simulation models seems adequate.Local models are useful complements of global estimates, in particular for capturing sub-national or sub-regional variations in opportunity costs.

Need for systematic transaction cost assessments
Transaction costs are important components of the costs of REDD, but methods for systematic assessment remain limited.Investment in guidelines for systematic transaction cost assessments could be of added value.This could be done in a manner similar to the three-tier methodologies developed by the IPCC Guidelines for land use and forestry projects where alternative methods are proposed depending on existing resources and capacity.

REDD costs depend on the degree of emission abatement (all versus almost all)
The marginal costs of REDD are increasing which means the more emission abatements are sought, the more has to be paid for one additional unit of REDD until no more reductions can be made (i.e., the maximal level of reductions is reached).In turn, emissions from deforestation and forest degradation to nearly zero requires far higher costs than would be needed when only reaching for the 'low hanging fruits', i.e., those emission reductions that can be achieved at lower costs.REDD policies could consider this option to reduce the costs of emission abatements.

REDD costs also depend on payment design
The costs of REDD further differ depending on the ultimate REDD payment design.Uniform payments, where every landowner receives the same amount of payment for REDD, are more costly (up to 4 times as found in one case in the Brazilian Amazon) than differentiated payments where each landowner receives just his/her opportunity costs.Because alternative payment schemes have different cost implications, and provided differentiated landowner payments are considered feasible in REDD schemes1 , it could be useful to dedicate some efforts to cost-benefit analyses of alternative payment schemes.

T
he costs associated with the reduction of greenhouse gas (GHG) emissions from deforestation and forest degradation (REDD) shape both the demand for, and supply of, REDD services.Investors and policymakers in particular are highly interested in REDD cost information -not only for budget reasons, but also to decide where to allocate investments (sector and country), and to assess how much benefit (in terms of reduced emissions) is received in return for investment.
The economic concept of the supply curve suggests that there is no single cost-value, but that alternative levels of REDD supply are associated with different costs.For example, while some landowners can reduce emissions inexpensively, others will require a higher price per ton of CO 2 reduced.This logic is reflected in the supply curve, which graphs the quantity supplied against the price paid (see Figure 1).In other words, there is no single "cost per ton" but rather a range of costs depending on the quantity -up to a point where no more reductions are possible (due to limited forestlands).
The costs of REDD fall into two categories: The first represents the 'opportunity costs', i.e., 1.
the forgone profits from alternative land uses such as cash or food crops (including revenues from timber sales), which correspond to the minimum price to be paid for REDD services.
The second kind represents the 'transaction 2.
costs', which encompass all costs associated with establishing and running the scheme (government transaction costs), and the costs individual landowners have to bear in order to participate in the programme (private transaction costs).
This paper reviews current methods to estimate the costs of REDD, with a focus on opportunity costs.It is organised as follows: Section 2 describes alternative approaches used to estimate the opportunity costs of REDD; Section 3 reports how transaction costs are estimated or used in REDD cost calculations; and Section 4 comprises concluding remarks.Global-empirical approaches.2.
Global simulation models.

3.
Sometimes these approaches are simply grouped into local and global models. 2

Opportunity cost calculation with local-empirical models
Estimates from local-empirical models are based on detailed studies (surveys) in a particular area.Both the per-area cost estimates ($/ha) and the carbon density estimates (ton/ha) are specific to the particular region studied, and the division of per-area opportunity costs by carbon density gives the opportunity costs on a perton basis (Boucher 2008).Since results are specific to a particular region, their extrapolation or generalisation can be problematic (Grieg-Gran 2006).
The literature on opportunity costs of tropical forestland is growing rapidly.Based on a review of 23 studies, Boucher (2008) finds that the resulting values (data points, not entire supply curves) of opportunity costs tend to be quite low with the mean being $2.51/ tCO 2 eq.To him, one possible explanation refers to the likely position of these data points in the early stages of the respective supply curves.
More sophisticated empirical studies giving whole supply curves are now emerging (see Box 1).These studies differ in several aspects including scope (deforestation, forest degradation), level of discount rate, degree of spatial disaggregation (accounting for geographic variation of key determinants, especially carbon densities), inclusion of transaction costs, practicability (replication of methods can be constrained if data or capacity requirements are very demanding) and accuracy.Table 1 summarises key characteristics of three selected local-empirical studies of opportunity cost estimation.Areas of improvement for these important approaches include more explicit integration of transaction costs, spatial heterogeneity in estimates and emissions from forest degradation.
Although approaches that give particular attention to the spatial heterogeneity of land-use change determinants (e.g., Nepstad et al. 2007)

Opportunity cost calculation with global-empirical models
Global-empirical models use local-empirical estimates and aggregate them to global per-area costs of reducing deforestation (Boucher 2008).The conversion of the global area-based costs ($/ha) to emission-based costs ($/tCO 2 eq) tend to be based on uniform values of carbon density (ton/ha), obtaining a single, global estimate of opportunity costs ($/tCO 2 eq).Although it essentially ignores carbon density variation across space, this permits data on per-area opportunity costs to be used for regions where no per-ton-carbon costs exist (Boucher 2008).This method was, for example, applied in the Stern Review (see Box 2).
To compare this with emission-based estimates on REDD costs obtained from other studies, Boucher (2008) converts Grieg-Gran's 2006 estimates to emission-based estimates using global carbon density values (538 tCO 2 eq/ha) from Houghton (2007).The resulting cost estimates correspond to a range of $1.74 -5.22/tCO 2 eq, with the midpoint ($3.48/tCO 2 eq) being 36% higher than the mean of his aforementioned review of local-empirical estimates.One reason therefore refers to the uncertainty in deforestation and carbon density estimates (Boucher 2008).

Opportunity cost calculation based on global simulation models
Global partial equilibrium models simulate the dynamics of the world economy to estimate supply of REDD services, represented in supply curves.Generally, they include the forestry sector, the agricultural sector, and other important parts of the economy that affect land use.They also consider the energy sector and other parts of the economy that reflect fossil fuel use and affect carbon prices, but do not simulate them in detail to calculate economic equilibria.In turn, these models are termed dynamic 'partial equilibrium models' (Boucher 2008).Examples of where such models have been applied include the Intergovernmental Panel on Climate Change Report (IPCC 2007) and the Eliasch Review on REDD finance (Eliasch 2008).
3 sites in Indonesia, 1 in Peru and 1 in Cameroon (results n/a for Cameroon).
Brazilian Amazon region.

Scope
Deforestation only.Deforestation only.Deforestation and forest degradation.

Assumptions/ constraints
Only private landholdings considered n/a Zero deforestation after 10 years; baseline 20,000 km2 per year.Analysis level Municipal level.
Pixel-level (resolution n/a); results aggregated to province/regional level.

Spatial heterogeneity
Low (within-municipal variation not captured, linear transportation costs).
Medium (no spatial variation of prices).
High, across all variables (except transaction costs).

Data sources
Medium resolution remote sensing data; secondary data (from existing surveys).
Medium-resolution remote sensing data; own survey data (extensive field work).
Remote sensing data; secondary data.

Transaction costs
Not included, but discussed.Not included.
Partly included.

Accuracy
Probably lower (use of aggregate data).
Probably high (bottom-up modelling).Strong assumption of zero deforestation after 10 years.High data/capacity needs for land use models.

Replicability of analysis
Probably easy (due to use of secondary data sources).
Probably more difficult (due to high data needs).
Probably more difficult (due to complex model).
Table 1: Characteristics of local/regional-empirical models to estimate REDD costs * While most international data sources provide emissions data in terms of CO2 or CO2E, many U.S. sources provide emissions data in units of carbon (C) or carbon equivalents (CE).Emissions reported in CO2 or CO2E units are 3.67 times emissions reported in C or CE units (www.pewclimate.org/global-warming-basics/facts_and_figures/gwp.cfm).† The choke price is the price that allows compensation of all avoided deforestation.† † Intergovernmental Panel on Climate Change.
In simulations on the cost of REDD, such models take into account the fact that the cost of reducing emissions depends on how deep those reductions are (i.e., the costs start out low and increase with the amount of reductions made).In addition, the supply curves vary over time.In the earlier years, more reduction will be achievable at lower costs than in later years because opportunity costs will rise in response to greater land productivity upon increasing agricultural intensification induced, inter alia, by REDD policies.
There are currently three main global partial equilibrium models that have been used for this purpose, known as GTM, DIMA and GCOMAP (see Box 3).The three models differ in various details including the sectors included, how their dynamics are simulated, how they divide up the globe spatially, the interest rates used internally for calculating equilibria, and data sets used (Boucher 2008).
Recently, the developers of the three models have collaborated to compare their models by generating a series of runs -using similar initial conditions -on the potential contribution of avoided deforestation activities to reducing GHG emissions (Kindermann et al. 2008).To this end, a baseline of future deforestation was specified under a zero price for REDD.As each model's baseline embeds model-specific assumptions The authors estimated municipal-level opportunity cost (REDD supply) curves for private landholdings in two Brazilian states in the Amazon region: Amazonas and Mato Grosso.Combining remote sensing data on deforestation from the Brazilian Institute for Space Research (INPE) with municipal-level survey data from the Brazilian Institute of Geography and Statistics (IBGE), the authors extrapolated municipal-level deforestation rates from 2000-2006 to 2007-2016.For each municipality, the share of private landholdings, forestland on private landholdings, land-use mixes and expansion of land-use categories were calculated.Gross per-hectare returns were calculated and combined with estimates of timber revenues and transport costs -the first implying an increase, the latter a decrease of net returns of up to 20 per cent, depending on the distance to the state capital.A 10 per cent discount rate was used in the calculation of net present value.To obtain opportunity costs per ton CO 2 , carbon biomass values were used from Houghton et al. ( 2001) lowest estimates (to remain conservative).Transaction costs are discussed, but not explicitly integrated into the analysis framework.The authors conducted a retrospective, spatially explicit analysis of the tradeoffs between carbon and economic returns in five sites across the tropics -three in Indonesia, and one each in Peru and Cameroon.The opportunity cost analysis comprises five main steps for each site: i) clarification and description of major land uses, ii) calculation of time-averaged carbon stocks (t-ave C) for the major land uses, iii) calculation of the private and social profitability of the land uses in terms of discounted net present value (NPV) using a 20-25 year time horizon with a 10 per cent private and a 3 per cent social discount rate, iv) land-use characterisation and landuse change analysis (since 1990) using medium resolution remote sensing data and extensive ground-truthing, and v) identification of sequestering and emitting land-use changes based on a pixel-by-pixel analysis of change in NPV and t-ave C, measured in terms of units CO 2 equivalent (CO 2 eq).The authors developed a conceptual framework to estimate the costs to tropical nations of implementing REDD programmes and applied this framework to the Brazilian Amazon.A bottom-up approach was used to estimate opportunity costs.Spatially explicit rent models for high-carbon (timber) and low-carbon (agriculture, ranching) uses of the Brazilian Amazon forest were derived from biophysical, climatic and infrastructure constraints to agriculture and livestock expansion in tropical forest regions.For each analysed pixel (4 km 2 ), rents for each competing land use (soy, cattle, timber) were calculated for 30 years, assuming a 5 per cent discount rate and a pre-determined schedule of highway paving.The maximum opportunity costs of forgone profits from nonforest land uses were used as upper limit benchmark for the level of REDD payments to landowners.The cost calculation assumes close to zero deforestation after the first 10 years of the programme.To obtain the costs per ton carbon (tC), pixel-level opportunity costs are divided by the carbon stock for each forested pixel using the forest carbon map developed by Saatchi et al. (2007).One important additional component of this study (compared to others) is that in addition to opportunity costs of REDD, the costs of implementing an REDD scheme (i.e., establishing and running three relevant funds) are considered.about future changes of biological and economic conditions (including population, technology, and trade), the three models yield different deforestation and carbon-emission projections.To determine the marginal costs of carbon storage resulting from avoided deforestation, additional simulations were conducted with the three models and assuming constant carbon

Estimating the global costs of avoided deforestation for the Stern Review. Source: Grieg-Gran (2006)
A combination of local-level estimates of returns and more generic/average estimates was used to calculate opportunity costs of avoided deforestation for the purpose of the Stern Review.The analysis assumes 100% additionality, zero leakage, and is limited to the 8 main tropical forest nations (Brazil, Bolivia, Cameroon, Democratic Republic of Congo, Ghana, Indonesia, Malaysia and Papua New Guinea), which make up 46 per cent of global deforestation.Main inputs to the analysis are: (i) the return per hectare under different land uses and different conditions, and (ii) the size of the area to which the different cost estimates should be applied.Annual returns per ha are converted to net present values (NPV) per hectare, with a 10 per cent discount rate and a time horizon of 30 years.
Three sets of cost estimates were calculated: i) the first scenario assumes that no returns to one-off timber harvesting have to be compensated as part of land conversion, ii) the second assumes that timber is harvested over 100 per cent of the deforestation area, and iii) an intermediate scenario takes into account the practical limitations on timber harvesting in some of the countries concerned.
One important element in the analysis is information on how much of the area converted will be used for each land-use form (i.e., whether pasture, soybeans, food crops etc.).Because land-use patterns depend on a number of local factors (soil, climate, access to markets etc.), it is difficult to make robust predictions.Where available, predictions from the literature were used.In most of the cases, however, this information was derived from subjective assessments of %-breakdown from qualitative statements found in the literature.
The review yields total annual opportunity costs for 8 countries as: US$ 3 billion (scenario 1), US$ 6.5 billion (scenario 2) and about US$ 5 billion (scenario 3).*Transaction costs were estimated (based on a literature review) to be around US$4-15/ha/year.Important factors that affect the costs include i) the discount rate used to calculate NPV, ii) the time horizon, iii) the estimates of commodity prices, and iv) the assumption about the proportion of deforested area that will be in high-or low-value agricultural alternative use.Moreover, compensation costs are likely to be higher if the assumptions of 100 per cent additionality and zero leakage were weakened.
prices ranging from US$0/tCO 2 to US$100/tCO 2 .Marginal costs tend to rise over time because the lowest-cost opportunities are adopted first and rates of deforestation decline, while later the opportunity costs of land rise due to increasing agricultural productivity (see Figure 2).The resulting marginal cost curves differ across models for a number of aforementioned This model assesses land-use options in agriculture and forestry in 0.5 degree grid cells across the globe.It predicts deforestation in forests where land values are greater in agriculture than in forestry and, vice versa, afforestation of agricultural and grazing lands where forestry values exceed agricultural use.To this end, the model calculates the NPV of forestry and NPV of agriculture and compares the two.Main determinants for the NPV of forestry are: income from carbon sequestration, wood increment, rotation period length, discount rates, planning costs, and wood prices.Main drivers for the NPV of agriculture on current forestland are: population density, agricultural suitability, and risk adjusted discount rates.The two values are compared and deforestation is subsequently predicted to occur when the agricultural value exceeds the forest value by a certain margin.

The Generalized Comprehensive Mitigation Assessment Process Model (GCOMAP)
This is a dynamic partial equilibrium model that analyses afforestation in short-and long-run species, and reductions in deforestation in 10 world regions.Specifically, the model simulates the response of the forestry sector to changes in future carbon prices.It uses data on country-specific activity, demand, and costs of mitigation options and land-use change by region.The model allows explicit analysis of the carbon benefits of reducing deforestation in tropical countries.However, it does not consider the impact of increasing CO2 concentration (CO2 fertilisation) on changes in the carbon cycle, and its effect on biomass growth.The model's objective is to estimate the land area that land users would plant above the reference case or prevent from being deforested, in response to carbon prices.

The Global Timber Model (GTM)
This is a dynamic optimisation model that optimises the land area, age class distribution, and management of forestlands in at least 146 timber types globally.Specifically, it maximises the NPV of consumers' plus producers' surplus in the timber markets.Because forestland competes with agricultural land, it models the interaction between the two markets via land supply functions that account for the costs of renting forestland.These land supply functions are specified for each timber supply region in the model.They are either constant or shift over time, depending on the assumptions about future development of agriculture in the region.
reasons, including the input data sets and modelling methodologies and assumptions.As depicted in Figure 2, the GTM model tends to predict the greatest emissions reductions for a given cost, whereas the position of the GCOMAP and DIMA estimates vary over time (Kindermann et al. 2008).
Weaknesses of these models include: often simplifying assumptions about potential rents from agriculture and livestock on tropical forestlands (Nepstad et al. 2007), the use of empirical estimates from the literature that in part may come from far away (Sohngen 2008), and the exclusion of transaction costs and other institutional barriers that raise the costs in practice (Kindermann et al. 2008).On the other hand, they all share a common approach as they are all based on the opportunity costs of different land uses (Boucher 2008), and are able to capture within-sector and crosssector interactions that can have important implications for REDD costs (Sohngen 2008;Kindermann et al. 2008).
Transaction costs exist as the result of limited knowledge and information and refer thus to the costs incurred in making an economic exchange.Transaction cost research yields various empirical approaches, each with separate focus and methodology -which further differ whether government agency cost or only individual costs are considered (Antinori and Sathaye 2007). 3According to Milne (1999) important categories of transaction costs include: information and procurement, scheme design and negotiation, 3 See Antinori and Sathaye (2007) or Milne (1999) for a review of transaction cost concepts and applications in carbon project assessments.

Estimating the transaction costs of rEdd
implementation, monitoring, enforcement and protection, and verification and certification.
Information on the transaction costs of REDD schemes remains limited.Most of the discussions tend to build upon insights and anecdotal evidence from other experiences, such as payments for environmental services (PES) schemes or forestry Clean Development Mechanism (CDM) projects (see Table 2).Transaction costs are reported either in the form of total costs or as percent-share of the entire budget.In the national-level PES schemes in Mexico and Costa Rica, administrative costs are ceiled by law at 4 per cent (Mexico) and 7 per cent (Costa Rica) of the respective budget, but also  High, due to pilot nature of scheme (65,000 -100,000).
slightly under-estimated since some of the transaction costs were covered outside the legally designated agencies (Muñoz-Piña et al. 2008;Pagiola 2008).In Costa Rica for example, private transaction costs (i.e., those borne by landowners entering into a PES contract) are reported to be in the range of 12-18 per cent of the environmental service payments (Wünscher et al. 2008).Table 3 provides further information on transaction costs associated with PES schemes (generally referring to the costs of setting up and running a scheme, not private transaction costs).
No consistent methodology for collecting data on transaction costs could be found.Transaction costs are often intangible and therefore difficult to measure directly.As mentioned before, the examples found in the literature differ in the categories of transaction costs that are analysed, and the way transaction costs are incorporated into an overall cost analysis.This makes it difficult to compare values across projects or regions.Quantitative analyses of transaction costs associated with carbon projects are emerging (e.g., Antinori and Sathaye 2007).
Insights from PES and CDM schemes suggest that transaction costs tend to be particularly high in early stages of a scheme (startup costs), and when the size of the scheme is small.Cacho et al. (2005), for example, show that for four carbon projects in Indonesia the startup costs can be quite large whereas running costs tend to be more manageable.Similar evidence can be found in the local PES schemes in Ecuador (see Table 3).Evidence for the economies of scale pattern of transaction costs includes Antinori and Sathaye's (2007) assessment of 28 GHG projects, and the comparative study of 15 PES schemes by Wunder et al. (2008).
One general observation is that existing information on transaction costs is limited.Good estimates are rarely available; their generally intangible nature makes them hard to estimate.In addition, transaction costs can vary substantially across governance contexts.Realising projects or PES-like arrangements in weak governance contexts is likely to imply higher transaction costs than in settings where institutions and rights are well-defined and wellfunctioning.At a practical level, this suggests a need for careful selection of projects.Nevertheless, transaction costs are in most cases not prohibitive.This review of methods reveals that various approaches to estimate REDD opportunity costs exist.The suitability of these approaches depends on the objective of the analysis.One striking observation is that the cost of REDD differs substantially across model approaches: global simulation models yield far higher REDD prices than empirical models, including the Stern estimate.One explanation is that global simulation models not only consider the opportunity costs, but also the costs arising from interrelations with other sectors and from the fact that for practical reasons land users are likely to be paid a uniform price, not differentiated according to their opportunity costs (Eliasch 2008).However, as global models can be criticised for their use of aggregated data and other simplifications, the 'true' cost estimate lies more likely somewhere in between the values provided by the local-empirical models on the one hand (lower end) and global simulation models on the other (higher end).

Alternative REDD cost assessment methods suit different needs
Despite the large difference in price results between local and global models, each method has its own advantages and disadvantages.For an indication of total costs of REDD as a climate change mitigation option compared to alternatives (i.e., for a worst-case cost scenario), the use of global simulation models seems adequate, although these models are often based on simplified assumptions and tend to exclude transaction costs elements.There remains room for improvement such as: including the consideration of avoided forest degradation, the use of more consistent parameters across models, and the use of more disaggregated data (especially as regards values of carbon density and deforestation rates).Furthermore, the important interrelation between deforestation abatement (REDD) and biofuel production remains little explored (R. Lubowski, pers.comm.).
Local models are useful complements of global estimates, in particular for capturing sub-national or sub-regional variations in opportunity costs.In particular, models that consider spatially varying determinants of land-use change (e.g.Nepstad et al. 2007;Swallow et al. 2007) seem useful in this context.However, the data and capacity needs required to develop and use such models may be prohibitive for replication in resource-constrained environments.Simpler approaches, such as the one by Börner and

concluding remarks
Wunder (2008), could be of greater operational use.Local models could be improved with the more systematic inclusion of private (landowners') transaction costs.

Need for systematic transaction cost assessments
Because opportunity costs vary across time and space, information on costs will continue to remain an important element in REDD design.This refers not only to opportunity costs, but also transaction costs -a component that has so far generally been neglected in cost models.Investment in creating standard procedures for opportunity cost and transaction cost estimations that yield quality estimates could therefore be of added value.When combined with guidelines for transaction cost assessments (both private and government costs), this could be done in a manner similar to the three-tier methodologies developed by the IPCC Guidelines for land use and forestry projects (Penman et al. 2003) where alternative methods are proposed depending on existing resources and capacity.

REDD costs depend on the degree of emission abatement (all vs. almost all)
The review of methods also underlines the increasing marginal costs of REDD.In other words, the further advanced along the REDD supply curve, the more has to be paid for one additional unit of REDD until no more reductions can be made (i.e., the maximal level of reductions is reached).This implies that reducing emissions from deforestation and forest degradation to nearly zero requires far higher costs than would be needed when only reaching for the 'low hanging fruits', i.e., those emission reductions that can be achieved at lower costs.It has been argued that REDD costs will be significantly lower if almost all emissions are abated, but not all of them.Boucher (2008) for example cites Nepstad et al. (2007) and their findings for Amazonia, where 1.49 US$/tCO 2 would have to be paid to abate 100% of emissions versus 0.76 US$/tCO 2 to abate 94 per cent of emissions.Similar examples can be found from other regions, suggesting that REDD cost assessments should consider this option.

REDD costs also depend on payment design
This raises important questions on the design of the payment mechanism, as the costs of REDD further differ depending on the ultimate payment design.An analysis of alternative payment mechanisms for REDD is beyond the scope of this paper, but it is important to note that their design can have substantial implications on the costs of REDD.Börner and Wunder (2008), for example, show that compared to differentiated payments (every landowner receives just his/her opportunity costs), uniform payments (every landowner receives the same amount of payment for REDD) increase the total costs of REDD by about 4 (from US$ 680 to US$ 2,745 million) and by nearly 2.5 (from US$ 143 to US$ 363 million) respectively in the Brazilian States of Mato Grosso and Amazonas for reducing total deforestation.Uniform payments predominate in current national PES schemes as they are easier to implement, but simple payment differentiations are also emerging in some national PES schemes (e.g., in Mexico), while differentiation predominates in private sector PES schemes worldwide (Wunder et al. 2008).Because alternative payment schemes have different cost implications, and provided differentiated landowner payments are considered feasible in REDD schemes 4 , it could be useful for the purpose of REDD to dedicate some efforts to costbenefit analyses of alternative REDD payment schemes 4 There are pros and cons regarding uniform/differentiated payments in the context of REDD, but more important is the political decision whether or not all countries will receive the same $/tCO 2 price.

Figure 1 :
Figure 1: Example of REDD supply curve

Box 1 :
Examples of local/regional-empirical approaches to estimating the costs of REDD*Example 1: Opportunity costs of RED in the Brazilian Amazon.Source: Börner and Wunder(2008)

Example 3 :
Costs of REDD in the Brazilian Amazon.Source: Nepstad et al. (2007)

Table 2 :
Non-exhaustive list of studies on transaction costs associated with REDD schemes