A review of land-based greenhouse gas flux estimates in Indonesia

We reviewed six recent government-led inventories that estimated and reported historical GHG emissions and removals from LULUCF, including peat drainage and peat fires, between 2000–2012 (table 1). Different government agencies developed these estimates, often in collaboration with academic researchers and civil society experts. The inventories fulfilled a range of both domestic policy requirements and international reporting obligations, including the National Action Plan to Reduce GHG Emissions (RAN-GRK), biennial update reporting to the UNFCCC (UNFCCC 2012), and forest reference emission level (FRELs) developed for the Reducing Emissions from Deforestation and Forest Degradation (REDD+) program (UNFCCC 2014).

We systematically examined the input data and methods, assumptions, and definitions, used to construct each GHG inventory. This included assessing their approaches to: (1) delineating the geographic extent within which GHG fluxes are estimated; (2) selecting which GHG emitting or removing activities are included; 3) estimating fluxes from above- and below-ground biomass (AGB and BGB), necromass in litter and debris, and/or soil organic matter; (4) including greenhouse gases including carbon dioxide (CO₂), methane (CH₄), and/or nitrous oxide (N₂O); (5) incorporating legacy fluxes, or GHG fluxes due to historical land cover change activities that occurred prior to the first year of accounting but continue during the inventory timeframe (Ramankutty et al 2007); and (6) accounting for carbon sequestration in land cover following a deforestation event.

We conducted a sensitivity assessment to determine the impact of observed methodological differences on emissions estimates. To do this, we developed a simple reference GHG flux estimate for deforestation, peat drainage, and peat fire activities using average activity data over the 2000–2012 period. This reference case is not a refinement of previous GHG estimates, but rather a neutral inventory which we used as a tool for comparing alternative methods. Because we controlled the input data and underlying methods generating the GHG estimate, we could introduce individual changes while holding all other factors constant.

We provide details of the reference GHG flux estimates from deforestation, peat drainage, and peat fires in the supplementary information available at stacks.iop.org/ERL/13/055003/mmedia. Briefly, to estimate deforestation emissions we quantified the area of conversion among six forest classes, two plantation classes, and two non-forest classes from 2000–2003, 2003–2006, 2006–2009, and 2009–2012, using data from the Ministry of Environment and Forestry (MoEF 2015b) (SI table 1). We used nationally specific carbon stock values to estimate GHG fluxes due to each type of land cover conversion (Krisnawati et al 2015) (SI table 1). To estimate GHG fluxes due to peat drainage, we assumed deforestation on peat lands is a proxy for peat drainage, and used default emissions factors (IPCC 2013) (SI table 2). To estimate GHG emissions from peat fires, we used MODIS active fire data to estimate the area of peat soils impacted by fire (MCD14ML), nationally specific estimates of the mass of fuel burnt (Krisnawati et al 2015), and default emissions factors (IPCC 2013). We also estimated emissions from forest degradation and removals from forest restoration, but did not test the impacts of modifications to the estimation methods for these activities.

We compared our reference GHG flux estimate to the results using different input data and estimation methods, summarized in table 2 and described in detail in the supplementary information. We did not test the impact of all possible theoretical modifications to the reference estimate, but rather focused on the approaches that existing Indonesian inventories put in practice, to identify the causes of discrepancies among actual inventory estimates. For example, none of the inventories accounted for carbon transfers to harvested wood products, or differentiated natural from anthropogenic disturbances using a managed land proxy, and we therefore excluded these theoretical variations from our assessment of alternative methods. In addition, we did not test the impact of incorporating alternative datasets outside the scope of what previous inventories have already used. For example, we did not test the impact of using burned area data, rather than active fire data, although the recent release of Collection 6 MODIS burned area product could substantially improve estimates of the extent of peat lands impacted by fire (Giglio et al 2016).

We observe several similarities among Indonesia's LULUCF GHG inventories, and highlight the overall correspondence among the inventories' input activity data. All of the inventories use the same definition of forest based on canopy cover > 30% (MoF 2004), and estimate land cover change using maps derived from the interpretation of satellite imagery (MoEF 2015b). In addition, all of the inventories use a map of peat land extent from the Ministry of Agriculture (MoA 2011), and assume that land cover change provides a proxy for peat drainage. Of the inventories which report emissions from peat fire, all base their estimates of the area of peat lands burnt on MODIS active fire data. Estimates of the area of each activity are therefore similar except for the INCAS 2015 estimate, which is based on land cover change data from the National Institute of Aeronautics and Space (LAPAN 2015) (figure 2(a)). The INCAS 2015 inventory also defines forest degradation as any deforestation events followed by forest gain and as timber harvesting in areas without canopy cover disturbance, resulting in a substantially higher estimate of forest degradation than other inventories.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Additionally, the inventories largely base their calculations of GHG fluxes on the same carbon stock estimates from the National Forest Inventory (MoEF 2015a) (SI table 1). They also use the same Tier 1 estimates of emissions from peat drainage (SI table 2) and peat fire (IPCC 2013). The inventories that estimate emissions from below-ground biomass, necromass, and non-CO₂ gases (SI table 3) use Tier 1 estimates (IPCC 2006a, 2006b, 2013). The exception is FREL 2014, which uses carbon stock data and peat emissions factor data based on the average of several national studies (SI tables 1 and 2). All the inventories assume that emissions from forest change and peat fires occur in the year of conversion (committed emissions), but assume that emissions from peat drainage occur in equal increments annually. The exception is INCAS 2015, which models GHG fluxes due to forest change over time.

Based on the reference approach, we estimated average annual GHG emissions of 0.47 Gt CO₂e yr⁻¹ over the 2000–2012 period, including 0.23 Gt CO₂e yr⁻¹ due to deforestation, 0.07 Gt CO₂e yr⁻¹ due to peat drainage, and 0.16 Gt CO₂e yr⁻¹ due to peat fires (figure 2(b)). Our reference emissions estimates fall within the range of the government-led inventories, except for our peat drainage estimate, which is lower. This may be because all reported peat drainage emissions estimates include at least one modification to the reference approach, such as the inclusion of methane emissions or legacy fluxes.

The largest changes to the reference deforestation GHG flux estimate occurred when we included emissions from belowground biomass and necromass pools (producing a 61% increase relative to reference), used alternative emissions factors (43% increase), and accounted for carbon sequestration due to forest regrowth after a deforestation event (31% decrease) (figure 3(a)). Including emissions due to plantation forest loss, which impacted an average of 62 kha yr⁻¹, resulted in an increase of 7% relative to the reference.

Our reference peat drainage GHG flux estimate increased substantially when we included legacy emissions due to deforestation that occurred prior to the year 2000 (figure 3(b)). There are two ways that the inventories in this assessment account for legacy emissions: (1) all 5.8 Mha of non-forest peat lands in the year 2000 continue to emit GHGs due to a prior deforestation event, producing an estimate 151% higher than the reference, and (2) only the 2.6 Mha of peat lands that were deforested and drained over the period 1990–2000 continue to emit GHGs, producing an estimate 94% higher relative to the reference. Including non-CO₂ emissions due to peat drainage, and using forest degradation as a proxy for peat drainage, increased emissions by just 4% each. When we used alternative emissions factors from the FREL 2014, our estimate of peat drainage emissions was 23% lower than the reference.

The largest changes to the reference peat fire GHG flux estimate resulted from modifying the assumptions of the mass of fuel burnt in a peat fire event (figure 3(c)). Increasing our estimate of the depth to which peat soils are burnt, and increasing the estimate of the bulk density of peat soils, resulted in emissions estimates which were 88% and 27% larger, respectively. When we accounted for methane emissions, our estimate of peat fire GHG emissions increased by 35% relative to the reference case. On the other hand, an average of just 19 kha of peat fires occurred in persistent forest lands each year, and therefore assuming these fires did not ignite peat soils reduced emissions by just 3% compared to the reference.

Our reference estimate of LULUCF emissions, of 0.47 Gt CO₂e yr⁻¹, almost doubled when we included all the modifications identified in our assessment, to 0.95 Gt CO₂e yr⁻¹. This difference is more than half of the observed 0.7 Gt CO₂e yr⁻¹ range between the lowest and highest inventory emissions estimates. The remaining disagreement can be explained in part by differences in input data, and resulting differences in estimates of the area impacted by each activity and emissions associated with each activity, and potentially by methodological differences that we were unable to identify in our review.

Differences in reported GHG fluxes over the study period are largely driven by varying degrees of inventory completeness, in terms of the scope of activities considered, the approach to define each activity, and the comprehensiveness of the GHG fluxes accounted for. The INCAS 2015 inventory is the most complete, because it incorporates fluxes from all activities, uses the broadest definition of each activity (e.g. includes peat fires even in persistent forests), and is the most comprehensive in accounting for component GHG fluxes (table 2). As a result, it reports the highest estimate of average GHG fluxes over the study period, of 1.1 Gt CO₂e yr⁻¹. The FREL 2014 inventory is intermediate in terms of completeness, as it excludes forest degradation and forest restoration activities, but accounts for fluxes from BGB and legacy emissions, resulting in an estimate of average GHG fluxes of 0.7 Gt CO₂e yr⁻¹. The FREL 2016 and BP REDD+ 2015 are the most conservative of the assessed inventories. The former does not consider peat fire activity, or non-CO₂ gases and non-AGB pools, although it does account for legacy emissions from peat degradation, resulting in an estimate of average GHG fluxes of 0.5 Gt CO₂e yr⁻¹. On the other hand, BP REDD+ 2015 includes peat fire emissions and forest degradation emissions, but omits all other modifications described in table 2, resulting in the lowest average estimate of 0.4 Gt CO₂e yr⁻¹.

Unfortunately, the NDC submission (NDC 2015) and the recent biennial update report (BUR 2015) provide limited information on input data and methods. These inventories estimate GHG fluxes across all sectors of the economy, and do not therefore elaborate the LULUCF portion of the inventory to the same degree as the other inventories which focus specifically on LULUCF fluxes. The NDC 2015 inventory excludes forest degradation and forest restoration activities, but reports the highest emissions from deforestation and peat fires, possibly because of a relatively comprehensive treatment of all component GHG fluxes. On the other hand, BP REDD+ 2015 includes all five activities, but reports the lowest estimates of emissions from deforestation and forest degradation, possibly due to the omission of key GHG fluxes. As a result, the NDC 2015 inventory reports average GHG fluxes over the study period of 0.9 Gt CO₂e yr⁻¹, while the BP REDD+ 2015 inventory reports 0.6 Gt CO₂e yr⁻¹.

Although we focus on the underlying reasons for differences in average reported emissions over the 2000–2012 period, there are also clear differences in inter-annual variability reported by the studies (figure 1), driven by the inclusion of peat fire emissions. Peat fires correspond closely to differences in seasonal precipitation (Field et al 2009), and are therefore the most variable source of GHG emissions over time. The BUR 2015, INCAS 2015, and NDC 2015 inventories report similar temporal trends in emissions because they all estimate emissions from peat fires annually. This differs from FREL 2016, which does not consider peat fires, and BP REDD+ 2015, which estimates very small emissions from peat fires, and therefore report relatively stable emissions over the study period. The FREL 2014 reports peat fire emissions averaged over each three year period 2000–2012 and therefore reports a moderately variable trend.

We observe that Indonesia's LULUCF GHG flux inventories frequently omit key methodological detail, and fail to disaggregate reported GHG fluxes by activity, land cover type, carbon pool, greenhouse gas, or legacy source. Providing separate estimates of component GHG fluxes will allow direct apples-to-apples comparison between inventories. For example, comparability would be substantially improved if all inventories separately reported estimates of gross CO₂ emissions from AGB pools in the event of natural tree cover loss. Inventories which include additional fluxes, for example from BGB pools, methane fluxes, or legacy fluxes, should report these separately. This type of disaggregated reporting would support differentiation of disagreement produced by variations in estimation methods, from differences attributable to data uncertainties.

The IPCC guidance for national GHG inventories affords countries substantial methodological flexibility to account for a broad range of unique country circumstances. A clear requirement to ensure comparability is therefore a standardized reporting framework that facilitates clear articulation of underlying methods and assumptions, and separate reporting of partial fluxes (e.g. by activity, land cover type, carbon pool, greenhouse gas, or legacy source). The ongoing refinement of the 2006 IPCC guidelines for national GHG inventories is one opportunity to encourage enhanced reporting transparency, and specify appropriate levels of data aggregation (IPCC 2016). Another opportunity to encourage inventory harmonization and reporting detail is via the UNFCCC technical assessment process for FRELS, in which evaluators review and provide detailed feedback on submitted FRELs (UNFCCC 2014). This technical assessment process could include recommendations for countries with concurrent inventory efforts to identify underlying reasons for any observed differences.

In Indonesia, the merging of the Ministry of Environment with the Ministry of Forestry in 2015 consolidated jurisdiction over GHG inventory reporting and land cover monitoring and management (Muridyarso 2014). However, this consolidation did not immediately clarify roles and responsibilities for LULUCF GHG inventory tasks: since 2014 at least four agencies and directorates reported different LULUCF GHG flux estimates at the national level (table 1). This shifting authority resulted in divergent methodologies and inefficiencies, as data collection, management, and reporting systems were rebuilt. A priority for harmonizing LULUCF GHG inventories is therefore a dedicated and permanent institution mandated with data collection, synthesis and reporting. Such an institution could also set nationally appropriate methodological standards for subnational GHG inventories, calculation of mitigation impacts, and recalculation of GHG flux estimates using newly available data, to ensure comparability and time series consistency. The nation's SIGN SMART system for GHG inventories is one tool that is currently operational, and has the potential to facilitate further institutional coordination.

A recent global assessment of national GHG inventories found that the omission of uncertain GHG fluxes, including from forest degradation and forest regrowth, from belowground and necromass pools, and of non CO₂ gases, is commonplace (Federici et al 2017). Indeed, several of the inventories assessed in this study, particularly those designed for FREL development, mention data uncertainty as the reason for excluding a flux component (although none specify the magnitude of uncertainty of these fluxes). Specifically, the FREL 2014 and the FREL 2016 omit GHG fluxes from an entire class of activity (forest degradation and peat fires, respectively) and non-CO₂ greenhouse gases, and BP REDD+ 2015 and FREL 2016 omit all carbon pools except for AGB. In the near term, completeness can be improved by including default (Tier 1) emissions factors until more detailed country-specific (Tier 2) estimates, or locally parameterized emissions models (Tier 3), are available (IPCC 2006a). In principle, GHG flux estimates should be consistent regardless of the data tier, but with larger uncertainty bounds for Tier 1 data than for Tier 3 data.

Over a longer time frame, an important strategy for encouraging inventory completeness is to increase confidence in estimates of GHG fluxes from key sources, by expanding the scope of data collection efforts. Others have noted the importance of improving the accuracy of estimates of LULUCF activity and emissions factor data across the tropics, and of peat land extent and peat carbon content specifically in Indonesia (e.g. Warren et al 2017). There is substantial ongoing research focused on estimating key GHG fluxes in Indonesia, including projects aimed at improving the accuracy of peat land extent and depth measurements (WRI 2018). This study highlights several additional research priorities, including estimation of GHG fluxes from belowground biomass and necromass pools, long term emissions due to peat drainage, the extent of peat drainage and peat fires, and the mass of fuel burnt in a peat fire event.

Only two of the six inventories report uncertainty; INCAS 2015 estimates uncertainty of deforestation emissions, but does not report a resulting confidence interval, while FREL 2016 estimates average uncertainty of ±16.1% from all activities. A priority for future inventories in Indonesia will be more systematic and comprehensive estimation of uncertainty, by propagating the uncertainty distributions for all input parameters through the inventory model (Griscom et al 2016). In addition to providing a gauge of the reliability of GHG inventories in general, clearly articulated uncertainty estimates are pre-requisite for identifying inventory components which should be priority targets for future scientific inquiry, underpinning confidence intervals for future emissions projections, determining conservative estimates of the success of mitigation actions (Grassi et al 2008), and potentially informing approaches for adjusting results-based payments for emissions reductions (Pelletier et al 2015).

This study highlights that a critical step for uncertainty assessments will be to isolate uncertainties associated with input data from differences due to alternative inventory model specifications, definitions, and assumptions. Using Panama as a case study, Pelletier et al 2013 demonstrated a method for doing this, by incorporating alternative inventory model assumptions into a Monte Carlo uncertainty assessment via binary inclusion/exclusion (Pelletier et al 2013). Given the magnitude of disagreement produced by alternative approaches observed in the present study, a similar approach to uncertainty quantification that enables comparison of uncertainty driven by input data versus model specification may be valuable in the context of Indonesia.