5 Preface Executive Summary

4.2 Forest cover, biomass, and degradation

Much traditional forest monitoring – both remote sensing-based and otherwise – has focused on the total extent of forest cover, with many forest statistics given in terms of percentage area under forest cover, number of hectares deforested, or similar measures. Total cover is perhaps the most fundamental element in the accounting of above-ground carbon, but by itself it is insufficient. Different areas of forest can have greatly differing carbon densities, so by extension the emissions produced by deforestation will vary greatly depending on the initial forest conditions. In addition to the natural variety in emissions between areas, there is a spectrum of forest degrading activities from minor degradation to outright clearing. The necessities of carbon accounting therefore sets two challenges for the monitoring of above-ground carbon that are above and beyond a simple measurement of deforestation rates: 1) quantifying the carbon content per unit area of forest, and 2) measuring degradation of standing forests.

4.2.1 Measuring biomass

Attempts have been made to quantify biomass directly using medium resolution remotely sensed data. However, no general statistical relationships have been developed that can describe this relationship consistently.⁹⁸ Any biomass estimation needs to therefore be supported by ground level data that relates to a given region or project locality. The IPCC has described three tiers of accuracy that can be achieved in this estimation.⁹⁹ Tier 1 uses very general data on the carbon content of different biomes that the IPCC has compiled and provided. This level would estimate carbon released from deforestation by calculating the area of forest, and then multiplying that by an IPCC emissions factor for the relevant biome, for example Southeast Asian tropical forests. A tier 2 estimate would incorporate more nationally-specific data. So instead of using a general value for an entire biome as would have been done in tier 1, it would base its estimate on some previous work in the country on the carbon content of specific forest types. These two levels of accuracy may both be insufficient for REDD, as it has been suggested that REDD will need to operate at the tier 3 level of accuracy in order to be viable.¹⁰⁰ Tier 3 estimates involve locally specific forest inventories and measurement. In the case of REDD, this would likely involve measurements of canopy height, tree diameter, tree density, and species composition in a series of plots in the project area. In the future, images from laser sensors may be able to provide some of these measurements remotely. However, for the time being, direct fieldwork is the only viable option. Allometric equations are used to relate the measurements of tree density and dimensions to carbon content.

4.2.2. Assessing degradation

The ability to assess forest degradation¹⁰¹ is of central importance to the development of REDD. This is especially true in Vietnam, where degradation outstrips deforestation as the primary source of forest carbon emissions.¹⁰² However, assessing degradation is a major challenge for most remote sensing analysis. Medium-resolution remotely sensed imagery such as SPOT or Landsat can detect variation within forested areas; however, as was mentioned above, no general statistical relationships exist that can relate this variation with changes in biomass. There is potential to develop relationships of this type specifically for a given region or project area, although the accuracy and reliability of estimations generated in this way would likely vary among forest types. This approach would involve stratifying forest areas into different levels of growth or degradation (e.g. mature, logged, burnt, young secondary, and older secondary) and doing ground surveys within each of these groups.¹⁰³ It may then be possible to generate relationships between these different categories of forest and their spectral characteristics on satellite images, thus enabling detection of transition between the categories and an estimation of associated emissions.

One method that may provide more straightforward detection of degradation in satellite images is Spectral Mixture Analysis.¹⁰⁴ This combines multiple spectral bands of Landsat images together into a single band called the Normalized Difference Fraction Index (NDFI). In many situations, the NDFI has been shown to be effective at detecting degrees of degradation (Figure 2). However, use of this method does not replace the need for ground measurements to determine the relationship between NDFI and carbon content in the project area.
These methods, although promising, will likely be associated with significant measurement uncertainty. Forest degradation will thus remain one of the most difficult aspects of accurate carbon accounting in any REDD project. Current REDD activities address this issue in a variety of ways, with some use of Spectral Mixture Analysis (COMIFAC in Cameroon) and some relatively intensive ground survey work (particularly NK-CAP). Specific project methodologies will be discussed further in the next section.

Figure 4-2: NDFI values (shown in color bar below each image) and their illustration of varying degrees of degradation in a scene from the Brazilian Amazon.¹⁰⁵

4.3 Project methodologies

Although discussion of REDD at the international level is moving towards a focus on national-level REDD programs, most REDD activity to date has been initiated by smaller projects. These projects – and the verification standards they have been held to – have driven much of the development of REDD methodology.

4.3.1 Noel Kempff Mercado Climate Action Project – Santa Cruz, Bolivia

The Noel Kempff Mercado Climate Action Project (NK-CAP) was the first avoided deforestation project to have emissions reductions verified by a third party. In 2005, Société Générale de Surveillance (SGS) validated emissions reductions from NK-CAP totaling 5.8 million tonnes of CO₂ between 1997 and 2026, of which 989,000 tonnes had already been reduced by 2005. These reductions and the baseline against which they were established were determined through a combination of ground-based monitoring and remote sensing.¹⁰⁶ The baseline was calculated from three related sets of information that will be discussed in the following paragraphs:

• 
  A linear extrapolation from historical data of deforestation rates in the project area;
  
• 
  The location of future deforestation as predicted by GEOMOD land use change modeling; and,
  
• 
  The ratio between the deforestation rate in the project area and the rate in a reference area.
  

The rate of historical deforestation in the project area was calculated from forest cover data at three time points: 1986, 1992, and 1996/97. Forest cover at these times was generated from images produced by the Landsat TM sensor using bands 3, 4 and 5. Measurements were taken in terms of total hectares of deforestation within the project area. By extrapolating linearly – that is, projecting deforestation into the future on the assumption that it will occur at an unchanging rate – the NK-CAP established a future deforestation rate to serve as a baseline.¹⁰⁷

The NK-CAP then used this projected deforestation rate and the maps of forest cover change that were used to generate it as input into land use modeling package GEOMOD, which is part of remote sensing software IDRISI Kilimanjaro. GEOMOD takes a set of inputs (e.g., the location of roads and rivers) and can estimate areas of probable deforestation, given the deforestation rate that was used as input. It is important to realize that GEOMOD is not used to estimate the extent of deforestation; rather, it takes a fixed total amount of deforestation and then approximates where that deforestation will occur on a landscape. This has important implications for carbon balance, as parts of the landscape will be more carbon dense than others. Estimates of carbon in the project area are made using a locally specific 1996 forest carbon inventory.



Figure 4-3: Map of Noel Kempff Mercado Climate Action Project area, buffer zone, and reference area.
An innovative aspect of the NK-CAP carbon balance monitoring methodology is their use of a reference area as a correction factor for the project baseline. Using a simple linear extrapolation of historical data, as many projects do, is problematic as it does not account for the potential for changing contexts to have an effect on deforestation rates. Using a reference area provides a mechanism whereby changes in local deforestation rates can be observed independently of project activities. The reference area is monitored using the same remote sensing techniques as the project area. One potential danger of this technique is that any leakage of forest degrading activities from the project area to the reference area may artificially inflate the project’s estimated emission reductions by exaggerating the difference between the project area and the ‘business as usual’ reference area. The NK-CAP addresses this by having a 15km-wide buffer zone where they monitor leakage. However, it is possible that leakage will occur at a greater distance from the project area than 15km.

4.3.2 Juma Sustainable Development Reserve – Amazonas, Brazil

The Juma Reserve Project does not process any independent remotely sensed data that is unique to the project. However, it draws heavily on PRODES, the Brazilian government’s extensive forest monitoring program.¹⁰⁸ PRODES has conducted a complete assessment of forest cover in the Brazilian Amazon – and the associated mapping of deforestation – every year since 1988, and has made all of the data publicly available.¹⁰⁹ PRODES – and the Juma Reserve Project by extension – uses Landsat TM and ETM+ imagery.

The Juma Reserve Project has established its baseline level of deforestation using PRODES maps and a land-use change modeling program called ‘SimAmazonia’.¹¹⁰ Project proponents using SimAmazonia predict that up to 62% of forest area in the Juma Reserve Project area would be deforested by 2050 under the ‘business as usual’ scenario, so any improvement on that steep baseline may be eligible for crediting. Carbon flux will be monitored by a continued use of PRODES for forest cover, as well as more involved measurements of biomass. According to the most recent project design document (PDD), the latter methodology is still being developed.¹¹¹

4.3.3 Ulu Masen – Aceh, Indonesia

The Ulu Masen project has not produced a baseline deforestation rate from historic data as most other REDD projects do. The project developers argue that Aceh has been so unstable in recent years that a historic rate would not be representative. Instead, the baseline was presented to auditors in a narrative form with a worksheet identifying which areas would likely be the most at threat from various pressures. The scenario thus developed was spatially explicit, with areas being mapped as low, medium, or high risk of deforestation. This approach led to a projected annual deforestation rate of 1.3%, which is fairly conservative when compared to rates on other Indonesian islands. Remote sensing was not used to establish the baseline, but Landsat imagery in 2000, 2002, and 2006, as well as SPOT 5 imagery from 2006 will be used to assist future monitoring.

Carbon stocks were determined using general values from the IPCC regarding average carbon volumes for the appropriate forest type. This is a very low-accuracy approach (tier 1) when compared with projects that develop more locally specific information. However, Ulu Masen project developers argue that with such a large-scale endevour, a rough estimation is the most appropriate approach at the project development stage. The methodology described in the PDD allows for measurement accuracy to gradually increase (to tier 2 and 3) as capacity builds and the project progresses.

4.3.4 Central African Forests Commission (COMIFAC) Pilot Project - Cameroon

This initiative is not a REDD project itself, but rather an experiment in methodologies that intends to lay a groundwork for future activities. It has been orchestrated by the Global Monitoring for Environment and Security (GMES) initiative, a program established jointly between the EU and the European Space Agency (ESA). The COMIFAC initiative also operates at a national scale, so its goal is less to support private sector projects as it is to support future national-level REDD programs.

This initiative used Landsat data from 1990 and 2000 followed by DMC data from 2005. The switch in sensors for the final monitoring year may be a result of the breakdown of the scan line corrector (SLC) on the Landsat ETM after 2003. Images were used to create a forest change map for all of Cameroon. After initial processing to try and remove cloud cover, two different approaches were used to train the classification of the images. In some cases, sample areas were selected for field study to verify land cover type, while in others Quickbird satellite imagery was used to fulfill the same role.¹¹²
One of the most useful contributions of the COMIFAC pilot project is its provision in a REDD context a technique that can monitor forest degradation with standard optical remote sensing techniques. This is the ‘Spectral Mixture Analysis’ ¹¹³ technique, which was discussed earlier in this chapter. In the case of Cameroon, this technique was useful in detecting areas of degraded forests.

4.3.5 ARBCP Lam Dong, Vietnam

The Asia Regional Biodiversity Conseration Program is the most advanced REDD project in any of the three focus countries of this report.¹¹⁴ In 2008, it completed field surveys for its carbon accounting and developed a projection for business-as-usual emissions from deforestation over the next 30 years. This indicated that the expected area of deforestation in the project area would be greater than 14 000 ha (140 km² ) over 30 years, and would result in emissions of 8 million tonnes of CO₂.

This baseline was developed using Quickbird imagery from 2001 and 2008. The location of future deforestation is predicted using a GEOMOD model that also incorporates data on distance from population centres and from waterways, slope, forest protection status, and forest type.

4.3.6 General lessons

It is clear from an examination of project methodologies that there is no single best practice; all methods need to be appropriate to the context of a given project. For example, the absence of a historical baseline for the Ulu Masen project – although completely at odds with most approaches – may be appropriate in that situation given the long-term armed conflict (and lack of commercial development of palm oil to date) in Aceh. However, there are general lessons that can be taken from these projects.

Project	Project area (km^2)	Sensors used	Years of historical data	Ground plots	Baseline determination and monitoring
Bolivia: NK-CAP	6400	Landsat ETM+ (bands 3,4,5); MODIS to estimate fire risk.	1986, 1992, and 1996/97	609 plots first assessed in 1997	GEOMOD modeling. Correction introduced by monitoring deforestation rate in reference area.
Brazil: Juma	5900	Landsat TM and ETM+ data from Brazilian PRODIS program.	Multi-year trends.	Ground plots monitored by local communities. Number not given in PDD.	Multi-year PRODES data incorporated into SimAmazonia model to determine baseline. Monitoring with PRODES data and other methodologies still in development.
Indonesia: Ulu Massen	7500	None to date. Have acquired Landsat images for future use.	N/A	Will be used in future to improve accuracy.	Qualitative analysis decided on baseline of 1.3% deforestation rate. Only rough monitoring in place now, but carbon inventories will become more accurate as project progresses.
Cameroon: COMIFAC	475 000 (Entire country area)	Landsat and DMC for wall-to-wall survey; Quickbird for image training.	1990 and 2000 (Landsat) 2005 (DMC)	Ground plots used for image training (in conjunction with VHR imagery).	GEOMOD modeling; not yet completed. Uses Spectral Mixture Analysis^115 to measure forest degradation.
Vietnam: ARBCO in Lam Dong		Quickbird	2001 and 2008	Ground plots measured in June 2008.	GEOMOD modeling used to generate a 30-year business-as-usual emissions scenario.

Table 4-2: Summary of remote sensing use by projects.

The best approach is one that contains internal checks and balances. A land cover change model such as GEOMOD is useful for locating deforestation. However – like all models – it is only as good as the parameter estimates it is calibrated with, and may become problematic if it is tuned too closely to historical pattern.¹¹⁶ A good methodology should therefore have a method to provide a crosscheck to the land cover model. In the case of NK-CAP, monitoring of the reference area provides this crosscheck to ensure that the deforestation rate used to calibrate the model remains realistic. Similarly, this reference area approach is not problem-free. It can also provide misleading estimates of emissions reductions if significant leakage is occurring. The reference area approach should in turn be crosschecked with qualitative research in the area to monitor leakage directly.

From the point of view of the imagery itself, certain characteristics are common among projects. Landsat, for example, is used almost universally because of its low cost and its good historical coverage. As a result of its universality, expertise with Landsat images is further developed than with other sensors; this has led to the development of useful techniques such as spectral mixture analysis, which was discussed above. The limited use of the very high-resolution optical sensors such as IKONOS and Quickbird, as well as the limited use of radar and laser sensors, indicates that the most technologically advanced approach is not always the most appropriate. In most cases this is simply an indicator that the extra expense of these technologies does not generate sufficient returns in terms of carbon crediting. The higher-resolution sensors do have their place, however – e.g., IKONOS and Quickbird can be used to ground-truth lower-resolution images in cases where fieldwork would be costly or impractical. Given its potential to pierce clouds and smoke, radar may become more common in coming years as individuals gain more experience with its use.