5 Preface Executive Summary

4.4 Operational recommendations for remote sensing in REDD projects

As illustrated by the projects above, remote sensing has a central role to play in the development of REDD. Analysis of satellite imagery is integral to both baseline setting and to the monitoring of changes in standing carbon stocks. The World Bank Biocarbon Fund RED Methodology¹¹⁷, which gives a general template for avoided deforestation projects to follow, requires the use of RS for all but a few of its nine steps. However, it gives few specific recommendations on which RS technologies projects should use, other than the recommendation that “medium resolution” technologies such as Landsat or SPOT should be the minimum.

The Biocarbon Fund methodology addresses deforestation exclusively. However, that does not preclude the application of its general structure to measure degradation as well. Addressing degradation can be done with more detailed classifications of land cover. The Biocarbon Fund methodology requires at a minimum that land cover be classified into 6 broad categories (forest, cropland, grassland, wetland, settled area, and other); however, a more detailed classification – of forests in particular – is also possible. Distinguishing between different forest types makes it possible to measure degradation if areas of forest change from a more carbon dense forest type to a less-dense one. The general method of measuring degradation is thus quite similar to the method for measuring deforestation. In both cases, maps are generated of forest and non-forest land cover classes, and changes between classes are monitored. Estimates of deforestation will look at the area of land that changed from a forest class to a non-forest class, while estimates of degradation will look at the area of land that is changing between forest classes.
Distinguishing between forest classes of different carbon densities is the greatest challenge for monitoring degradation. The spectral mixture analysis approach as described above may prove effective for many projects. This gives a single value (NDFI) that correlates well with forest degradation and that may be used to define forest classes based on areas that have NDFI values with certain ranges. Ground plots can then be used to generate statistical relationships between carbon content and area within each forest class. This technique can be used with Landsat imagery, although the precision using Landsat may not be sufficient to generate reliable descriptions of forest types.
All steps in the measurement and monitoring process involve a tradeoff between increasing cost – either by using higher-resolution imagery or more extensive ground plots - and increasing accuracy. One approach that has been taken in some project methodologies to address this tradeoff is to use inexpensive lower resolution imagery, generally Landsat, for most of the historical sampling years, but higher resolution imagery such as SPOT for the sampling year that is closest to the project start date. An alternate - or additional – strategy for reducing cost is to restrict the use of higher resolution data to specific project areas which are known to be changing more rapidly, or which have more variability in the data. The target areas for higher resolution sampling can be selected based on the lower resolution sampling and on project staff knowledge.
The combination in which medium resolution imagery and high resolution imagery is used will depend on the characteristics of a particular project. Two of the most important considerations in this regard are the scale of the project and the relative importance of degradation vs. deforestation in the area. These will both influence how cost-effective a given use of imagery is. Regarding project scale, it may be worthwhile for smaller projects to use higher-resolution imagery such as Quickbird or IKONOS. Although this is more expensive, it will allow the project to capture more carbon credits per unit area, which may be important for projects operating at small spatial scale. The usefulness of higher-resolution imagery will also depend to some degree on the relative importance of degradation as compared to deforestation in the project area. The larger typical forest clearings in an area are, the easier they are to detect with technology such as Landsat. In such cases, high resolution imagery may not be worth the investment. However, in projects where forest degradation is responsible for more of the forest carbon emissions than outright deforestation – this may often be the case in Vietnam, for example – there will have to be greater use of high resolution imagery.
At present, remote sensing in REDD projects is only used to measure area directly. Estimates of carbon stocks are established from carbon density estimates for particular forest types. Carbon density – that is, the average tonnage of carbon per area of a given forest type – needs to be established using ground plots, and its estimation will vary in precision depending on the number of plots and the overall variability of the forest type being sampled. As technology develops, it may become possible to directly measure biomass using images from laser sensors. For the time being, however, the only approach which if feasible at a broad scale is the area-based approach as described above.

4.5 National forest monitoring schemes

4.5.1 Lao PDR

Laos has twice completed a national inventory of forest cover – once in 1992 and again in 2002. The Department of Forests (DoF) in the Ministry of Agriculture and Forestry (MAF) has the responsibility of carrying out this inventory once every ten years. Both inventories to date used SPOT imagery for the assessment. In 2002, the DoF quantified forest cover change by comparing the two sets of SPOT images. This was done by a two-stage sampling approach whereby 999 10km x 10km primary sampling units (PSU) were randomly chosen over the surface of the country, and 81 secondary sampling units of 7ha each were chosen within each PSU. This process sampled 2.3% of the surface area of the country, and generated a statistically sound estimate of forest cover change.¹¹⁸

The DoF is in the process of establishing a set of permanent sample plots in forest areas across the country. There are already 200 plots established in each of three provinces, with 100-200 plots planned to be established in each of the remaining 14 provinces. These plots are the focus of ongoing national stand inventories, and allow long-term measurement of tree volume, height, and survival. This will be useful data for future evaluation of carbon stocks in Lao forests.¹¹⁹
Remote sensing has been used in Laos to successfully target and reduce illegal logging. The catchment area around the Nam Theun 2 hydroelectric development was identified at high risk for these activities, and so a survey regime was instituted in the area by GTZ with the support of the Lao government. This program used a combination of Landsat and SPOT images together with aerial photos to survey the area in both 2000 and 2002. Landsat was chosen because of its good resolution and low price, while SPOT was used as a backup in situations where cloud-free Landsat images could not be obtained. Aerial photos were used instead of high-resolution satellite images such as IKONOS because the latter was deemed too expensive for the required use.¹²⁰

4.5.2 Nepal

Of the three countries discussed here, Nepal has the most limited national forest monitoring program. Nepal’s national proposal, or R-PIN, to the World Bank’s FCPF identifies the improvement and extension of forest cover databases as a central priority for the development of REDD in the country.¹²¹ The Department of Forest Research and Survey (DFRS) is responsible for forest monitoring in Nepal. Although forest assessments have been completed, Nepal lacks the regular system of forest monitoring that Lao PDR and Vietnam subscribes.

To date, much of the remote sensing work in Nepal mapping forest cover has been done using aerial photos. Forest assessments based on aerial photos were released in 1964 and 1979; however neither of these covered the entire country. The first assessment with complete coverage was not finished until 1998, and included a mix of data from 1990, 1991, 1992, and 1996. The data assembled for this assessment was from different sources. Most of the country was still mapped using aerial photos at a 1:50,000 scale. The only region of the country where satellite imagery was used was in the Terai districts, relatively flat areas below the Himalayan foothills. In the case of the Terai, Landsat TM images were used from 1990 and 1991. An analysis of forest cover change was undertaken in 2005 as a partnership between the government of Nepal, SNV, WWF, DFID, and SDC. This compared Landsat images between 1991 and 2001; however, the focus was again only on the Terai.
Nepal faces one major challenge in its forest monitoring: the shadows cast by its mountains. Many satellites fly over Nepal in the morning, meaning mountain shadows are often long.¹²² Although in some cases this may be correctable with image processing, in most cases this means a significant loss of data. This fact may mean that in some areas the use of aerial photographs for forest monitoring will remain more practical than traditional optical sensors such as Landsat.

4.5.3 Vietnam

Remote sensing techniques have been used to monitor forest cover in northern Vietnam since the 1970s. This early work was completed using 1:25,000 aerial photos, with forest cover maps created at the same scale. Following the launch of Soviet satellites and the American Landsat series, Vietnam began using satellite imagery to create forest cover maps. As of December 2007, Vietnam uses a ground receiving station that receives images from French and ESA satellites. This station has the capability to both receive and process images. The GoV provides the first five years of this project, after which point it is planned that the station will be self-sufficient.

The GoV has mandated that the National Forest Inventory, Monitoring and Assessment Program (NFIMAP) take place every five years. This inventory was initiated in 1991 and is presently in the fourth cycle: 2006-2010. In all cases, the inventory has involved ‘wall-to-wall’ sampling; that is, the entire forested area of Vietnam has been sampled. Each round of Vietnam’s NFIMAP has used a different remote sensing product with later cycles using products with progressively better spatial and/or spectral resolution.¹²³

Years	Sensor	Resolution	Data	Number of ground plots
1991-1995	Landsat TM	30m	Hard copy only	4200
1996-2000	SPOT	20m	Digital	4200
2001-2005	Landsat ETM	30m	Digital	4200
2006-2010	SPOT 5	10m	Digital	2100

Table 4-3: Vietnam’s four rounds of forest inventory.
The MARD Forest Inventory and Planning Institute (FIPI) is responsible for NFIMAP. FIPI manages a set of permanent sampling plots that are measured every five years. Different ecological zones are sampled in different years to make efficient use of labour; however, a particular zone will always be sampled in a single year, and sampling events for the zone will be five years apart. A permanent set of 4,200 plots was sampled for the first three cycles of NFIMAP; however, with the higher resolution SPOT 5 imagery being used in the current cycle, FIPI decided to scale down the number of ground plots by half.
Donors such as the World Bank, the ADB and GTZ have undertaken some forest monitoring projects in Vietnam. Some of these have used higher resolution data than the NFIMAP (e.g. IKONOS or Quickbird). These ODA projects support various Vietnamese governmental counterparts, which unfortunately are not under the same ministry/department; hence efficiencies are lost whilst redundancies and data hoarding occur.
In Vietnam, the Governments of Japan, Finland and United States have established forest monitoring projects that aim to boost the capacity of the GVN to engage in REDD initiatives. The Forest Agency of Japan has funded a research project on testing the potential applications of the Japanese Advanced Land Observing Satellite (ALOS)/PALSAR data to establish forest cover maps and to estimate forest carbon stock in two provinces; this activity is currently being implemented by a Japan-based consulting firm. The Finnish-funded Forest Management Information System (FOMIS) project (€4 million) is planned to begin in May 2009 to develop a more reliable forest inventory information platform and establish a more accurate forest stock baseline. In southern Vietnam, USAID Asia Regional Biodiversity Conservation Program supports Winrock International to develop a forest protection project ($6 million from 2005-2009), includes a project-based REDD initiative that uses Quickbird technology to estimate the forest carbon modeling of a 80,000 ha watershed forest in the Da Nhim basin of Lam Dong province; it is anticipated that this project will be the first operational REDD project in Vietnam.
Although monitoring of forest cover is quite advanced in Vietnam, monitoring of carbon stocks (as in many countries) is at a fairly basic stage. The national government has no spatially explicit information on carbon stocks. The only emissions data regarding land cover change comes from the scaling up of gross deforestation values by standard IPCC values estimating tonnes of CO₂ per hectare of forest. These values are generalized across many landscapes and types of forest, and so for any given area they are very approximate. The uncertainty in these estimates is very high.