Uncertainties in deforestation emission baseline methodologies and implications for carbon markets

Carbon credits generated through jurisdictional-scale avoided deforestation projects require accurate estimates of deforestation emission baselines, but there are serious challenges to their robustness. We assessed the variability, accuracy, and uncertainty of baselining methods by applying sensitivity and variable importance analysis on a range of typically-used methods and parameters for 2,794 jurisdictions worldwide. The median jurisdiction’s deforestation emission baseline varied by 171% (90% range: 87%-440%) of its mean, with a median forecast error of 0.778 times (90% range: 0.548-3.56) the actual deforestation rate. Moreover, variable importance analysis emphasised the strong influence of the deforestation projection approach. For the median jurisdiction, 68.0% of possible methods (90% range: 61.1%-85.6%) exceeded 15% uncertainty. Tropical and polar biomes exhibited larger uncertainties in carbon estimations. The use of sensitivity analyses, multi-model, and multi-source ensemble approaches could reduce variabilities and biases. These findings provide a roadmap for improving baseline estimations to enhance carbon market integrity and trust.

No more than 15 years ○ Green Climate Fund i.
Preference for 10-15 year period (As cited in Nigeria FRL submission) ii.
Allows for reference period of 5 to 20 years (As cited in Mozambique FRL submission) 2. Number of historical reference years for deforestation risk mapping ○ VCS VM0015 i.
Start date of reference period to not exceed 10-15 years in the past ii.
End date to be as close as possible to the project start date Selection of variables tend to differ between countries due to national circumstances that affect availability of data.
Variables used in FRL/FREL constructions are similar, but it is unclear whether there are specific recommendations for all variables.The following variables and recommendations are based on the VM0015 methodology: 1. Carbon pools * Significant = constitutes >5% of total GHG benefits generated ○ AGB (tree) is the only mandatory carbon pool across all projects ○ Harvested wood products must be included when significant ○ following are only mandatory when requirements are met for specific projects: i. AGB (non-tree) ii.
Dead wood iii.
Litter ○ The following are recommended and not mandatory i. BGB ii.
Soil organic carbon 2. GHG sources ○ No mandatory sources, significance to be determined for specific projects 3. Forest classification ○ Minimum classes are "forest" and "non-forest" ○ Forest strata dependent on resolution of national data 4. Reference years ○ Long reference periods could result in the inclusion of historical patterns that do not reflect predicted future patterns ○ Short reference periods may be insufficient in capturing the true historical trend of emissions ○ For FRL/FREL: Availability of reliable national Activity Data (AD) and Emission Factors (EF), satellite imagery, occurrence of natural disasters (e.g.hurricane Maria affecting Dominica's FRL calculation) Note S3.Brief explanation of rationale and theoretical background for variables used in linear modelling.

Elevation and slope
Elevation is an important driver for biogeographical patterns and thus ecological and zones; human settlement dynamics and government policies in certain countries may also follow elevational zones (Bhattarai et al., 2009;Trigueiro et al., 2020).
Steep slopes may inhibit deforestation as these are unconducive for agriculture and may be less accessible (Carvalho Lima et al., 2018;Trigueiro et al., 2020).

Temperature and precipitation
Temperature and precipitation are climatic drivers which can affect fire dynamics in areas which are fire-prone, such as the ignition and ease of spread; hotter and drier climatic conditions may thus increase deforestation risk (Aragão et al., 2008;Laurance et al., 2002).Certain climates more conducive for agriculture may be attractive for agriculture-driven deforestation (Bax & Francesconi, 2018;Grau et al., 2005).

Gross Domestic Product (GDP) and Human Development Index
The environmental Kuznets curve suggests that a metric for environmental degradation (e.g.deforestation) rises at first with rising development due to greater demand for resources and consumption, but environmental degradation drops after a turning point as development brings technological improvements and greater demand for environmental amenities.Different relationships between GDP/development and deforestation have been suggested (Koop & Tole, 1999).

Nightlight intensity and population density
Settlement patterns, reflected in population density, may influence deforestation drivers such as agricultural expansion, infrastructure development, and resource extraction (Teo et al., 2019;Tritsch & Le Tourneau, 2016).This may come in the form of large-scale agriculture, export-driven agriculture, or subsistence agriculture and the extraction of firewood for domestic use (Fisher, 2010).
Nightlight intensity is frequently used as a proxy for population density and economic activity, and to complement such datasets (Dorji et al., 2019;Liu et al., 2021).

forest area and percentage agricultural area
The percentage of remaining forest area, and percentage of agricultural area, are key factors in forest transition theory (Mather & Needle, 1992).Forest transition theory describes the relation between the stages in development (reflected by agricultural land) and forest cover.

Percentage land area occupied by mining land uses
Mining is a key driver for deforestation in many areas (Alvarez-Berrios & Mitchell Aide, 2015;Ang et al., 2021;Ranjan, 2019).Deforestation pathways may include infrastructure development, urban expansion for workers' housing and services, as well as broader supply chains (Sonter et al., 2017).

Fig S1 .
Fig S1.Mean deforestation rates for each jurisdiction across all datasets and entire study period.
Note S2.Variables in ex-ante estimation of emission reductions.
3. Carbon pools for FRL/FREL submissions○ IPCC provides default values for dead wood, litter and soil organic carbon (As cited in Belize FRL submission, but unable to find the given values) 4. Tools and methods for deforestation risk mapping Any models and software can be selected, as long as they are peer-reviewed and prove