Peter Holmgren welcomed the participants in the meeting and outlined the objectives of the meeting. He indicated that the FRA 2000 was now complete and the programme FRA at Rome is now considering alternative approaches for the next Global Forest Resources Assessments (GFRA) due in 2005 and 2010. Further that this meeting builds on a conceptual Forest Information Framework developed last year in collaboration with UNEP. He also reminded that the meeting was informal so the participants were attending personally and not on behalf of one's organisation or institution.
Participants went through the Provisional agenda and recommended the order in which participants would like to make presentations. Most of the participants suggested that the discussions should focus on the development of the framework rather than the technological details.
Peter Holmgren introduced the Forest Resources Assessment (FRA) programme and explained his ideas on how the Information Framework fits in with the FRA programme (see Annex 2, p21).
He described the two major components of the FRA: Global Forest Resources Assessments (GFRA), and Support to National Forest Assessments (NFA). The participants were informed about the recommendations of the expert consultation on FRA, held in Kotka in July 2002 (Kotka IV) including holistic and broad assessment of forest resources using the framework of "Criteria" for sustainable forest management. Details were also given on the approach developed to support National Forest Assessments: a low-intensity systematic sampling based on the latitude-longitude grid is used to capture remotely sensed information and to conduct field inventories to generate new information, at national level, on forest and tree resources, as well as on their uses and management.
The initial discussion following the presentation spanned a wide range of important issues relating to development of the proposed framework including the following:
a) Manner the scientists programs could contribute to national and international needs;
b) Complementarity of sources of information and methods. The importance of integrating different approaches for collecting primary information, such as remote sensing and field inventory (observations, measurements and interviews) was underlined;
c) Ubiquity of lack or weakness of information about forestry and many different allied fields such as water runoff;
d) Current technology to produce national information and real problems in its implementation;
e) Application of the information framework to other goals and other monitoring efforts than the FRA, in particular to meet reporting needs of the Millennium Development Goals.
f) Effectiveness of a sampling approach.
Anne Branthomme from FAO/FRA presented the methods and results of the Remote Sensing Survey carried out for the last Forest Resources Assessments to lay the basis of discussion for the proposed design and expected information content based on the remotely sensed information (see Annex 2, p23). She informed that the Remote Sensing Survey for FRA 2000 used a stratified random sampling design with each sample representing a Landsat frame over the tropics, with a 10% sampling intensity. The survey was based on three date time series of high resolution satellite data (1980, 1990 and 2000 data sets). Transition matrices showing land cover changes over the periods 1980-1990 and 1990-2000 constituted one of the major outputs of the survey. Estimates on forest area and forest area changes were also calculated at different levels (tropical, regional, ecological zones). Trend assessment was carried out by comparing the two studied periods. The main findings of the survey were also given. Some of potential fields of improvement of the survey were outlined, including the identified weaknesses of last remote sensing survey like the lack of intensive and systematic field checking and the inefficiency of the sampling design with respect to the estimation of some variables due in particular to a stratification that was neutral for some surveyed characteristics, and to the high variation between the sampling units.
Ashbindu Singh, from UNEP, delivered a presentation entitled "United Nations Environment Programme and FRA" (see Annex 2, p26). He drew attention to the indicators of the Millennium Development Goals (MDGs), in particular to the Indicator 25 (Land area) and Indicator 26 (Land area protected to maintain biological diversity) that address the Target 9 (Integrate the principles of sustainable development into countries policies and programmes and reverse the loss of environmental resources) of the MDG 7 (Ensure environment sustainability). He informed that there is a lack of information to monitor effectiveness of protected areas like spatial coordinates and boundaries of the protected areas, evidence of change in land use and cover. Examples from Iguazu National Park and Wolong Nature Preserved were exposed to demonstrate this problem and show the necessity for further assessment of protected areas. Issues concerning the lack of monitoring data to detect and quantify land use and cover changes within Biodiversity Hot Spots were also addressed. Further that the current focus of UNEP is reporting on the state and change in environment.
He described in brief the UNEP work plan for monitoring changes including provision of Landsat 1992 and 2000 data sets to Land Degradation Assessment in drylands project (LADA), FRA and other users using a geographic database; the collection of current and historic satellite data, ground photographs and stories for 130 sites around the world, and the development of a interactive map server, in the framework of "Satellite Images of our Changing Environment". Moreover, the UNEP is also interested in monitoring land cover changes in targeted areas around the world. Examples of monitoring in Indonesia and Lake Chad were given. Country capacity building was also mentioned as a priority activity under different UNEP programmes. He envisioned great potential of joint partnership between UNEP and FRA at FAO including satisfaction of MDGs and provision of information on the status, trends, causes, consequences and solutions of forest extents and changes, country by country.
Jeff Tschirley, from FAO/SDRN, introduced the Global Terrestrial Observing System (GTOS) which is sponsored by FAO, ICSU, UNEP, UNESCO and WMO (see Annex 2, p.29). The mission of GTOS is to facilitate the access to information on terrestrial ecosystems, detect and understand large-scale change, build links between research and policy-making and work with countries to develop global and continental or regional products. Its focus is more global and regional rather than national monitoring even though much of the data has its origin in national programmes.
GTOS seeks to define terrestrial observational requirements, design and implement observation and information systems that have measurements of known accuracy and quality, and to enhance the collection and use of data. Its programme is focused on facilitating the development of datasets in areas such as net primary productivity, habitat fragmentation, fire, land which are building blocks for global change studies and assist the modelling community in developing improved estimates of the magnitude and rates of change.
GTOS consists of 3 technical panels: Terrestrial Climate Observations, Terrestrial Carbon Observations, Forest and Land Cover Dynamics; an informal panel has been formed on Terrestrial Coastal Observations. GTOS is focusing on two regions: southern Africa and central and eastern Europe. The Terrestrial Ecosystem Monitoring Sites (TEMS) database is a global directory of sites that carry out long-term terrestrial ecosystem monitoring of 110 variables. The data from the sites studied in the framework of the FRA 2000 Remote Sensing Survey are also accessible through TEMS. The meeting was informed that a major challenge facing GTOS is the need for building stronger networks to support terrestrial observation systems based on satellite and in situ data.
John Townshend outlined main features of GOFC/GOLD (Annex 2, p32). One important goals of GOFC/GOLD is to make data sets and products available by linking with space agencies. He mentioned that currently it was more focussed towards satellite data than on in situ data. Efforts are structured into two main components: land cover (forest and land monitoring) and fire implementation teams.
David Skole then presented the role of the GOFC/GOLD in global and regional forest monitoring (see Annexe 2, p.34). He informed that one major goal of the GOFC is to improve access to remote sensing data, through in particular CEOS agencies contributions. It may also contribute to improve pre-processing of remote sensing data (atmospheric corrections and geo-registration) and to develop methods for change detection. Other objectives include improving the global land cover products, especially the development of global/regional forest maps based on high-resolution remote sensing data, and land cover change monitoring, by using multi level monitoring systems, and coupled remote sensing and in situ systems (kNN and neural techniques). Activities comprise forest cover and carbon monitoring, ecosystem assessment, forest resource assessment and management, linking carbon and global change community needs with international needs. The approach adopted consists in the acquisition of primary observations and the application of algorithms, establishing a framework for product development, and product packaging.
The GOFC program has capabilities to use and coordinate programs and facilities. Space agencies and data providers are directly involved in the initiative. NASA, for instance, contribute to centres such as the Tropical Rain Forest Information Centre (TRFIC). Through GOFC, global Landsat 2000 data set, produced by NASA, was given for free to the UN system. Partnerships with existing programs, such as FRA, Africover, are important in order to identify what are the needs. GOFC supports existing programs, for instance GLCN, to which it provides primary observations.
GOFC has produced a Global Land Cover 1992-1993 map, based on AVHRR data. A map based on MODIS data (250 m resolution) is being developed currently. Regional multi-date analyses were also carried out in Cambodia and Viet Nam. These were using simple classification Forest/Non forest. Currently fractional forest cover (fc) is being tested and used. GOFC established regional network in Asia, SEARIN. Moreover a web-based geographic information system was also developed.
John Townshend presented the Global Land Cover Facility (GLCF) that allows downloading satellite data through the Internet (see Annex 2, p39). Currently about 1000 Landsat scenes are available freely for download. Other data products are also made accessible to users, in particular MODIS and AVHRR data sets, and Global Land Cover and deforestation maps products. Currently new MODIS reflectance products are being developed to produce NDVI. Percent forest cover maps based on MODIS data at 500 m resolution are being developed for the year 2000. The GLCF is a partnership between earth science and computer communities.
Frédéric Achard, from the Joint Research Centre (JRC) of the European Commission (EC), focused his presentation on the activities of the Institute for Environment and Sustainability and in particular of its Global Vegetation Monitoring Unit (GVM) and on the potential contribution of the JRC to the Information Framework (see Annex 2, p42).
Past activities of the GVM were first described, in particular the TREES and the GLC 2000 projects.
The TREES combined coarse and high resolution satellite data and a 6.5 % stratified sample covering the humid tropical domain, to study forest cover and forest cover changes. The stratification was based on deforestation hot-spots and vegetation. The method used in this survey was quite similar to the one applied in the FRA survey. Intensifying the sample in the hot spot areas proved to be a valuable approach but the allocation between strata should have been better balanced since the sampling units outside the hot-spot had a very high weight and contributed to increase in the variance. The quarter of Landsat scenes used in TREES, was a good compromise but even smaller samples would improve the sample design. The samples, available on TREES website, could be used to improve stratification efficiency for a future monitoring exercise.
The GLC 2000 project is finalising a new global land and forest cover map at regional level (1 km resolution from the Spot-5 Vegetation sensor) which will be ready at the end of March 2003. The comparison of forest area estimates from the TREES and the 1 km maps for the year 1990 shows that results were very close, demonstrating that low-resolution maps can provide good estimates on forest cover at national and regional levels. The new GLC 2000 map could be used to produce improved forest area estimates for the year 2000.
Frédéric Achard also informed about a few potentially relevant projects to be carried out in the near future (2003-2006) under the Global Monitoring for Environment and Security (GMES) umbrella:
David Skole presented some of the activities of the Tropical Rain Forest Information Center (TRFIC) including a study on selective logging in Brazilian Amazon using Landsat data from 1992, 1995 and 1999 (see Annex 2, p45). Former areas under logging were distinguished from new areas under logging and analyses of forest disturbance and degradation were also undertaken.
Discussions following the presentation covered issues like relative importance between logging and clearance by farmers in the Amazon, and large variations from one year to another in the deforestation and logging in the Amazon.
Kailash Govil presented the main features of sampling design proposed for the Information Framework (see Annex 2, p.48). The Global Forest Resources Assessment (GFRA) was first described, within the context of the forest related international processes. Information requirements for the GFRA 2005, to report on selected criteria were outlined. It was stressed that these requirements could be achieved only by combining remote sensing and ground inventories. The need of a common information framework was underscored as it would contribute to link or refer to, understand, improve efficiency, aggregate, coordinate, and decentralize the information.
The proposed design for the information framework is a systematic grid based sampling on the earth surface. A sample location is determined at the each crossing of the 1 degree by one degree latitude-longitude grid. Each site would cover 10 km by 10 km. It results in a 1 % sampling intensity of land surface, and about 13 000 location sites excluding the poles land. This design could be linked to the NFA sampling scheme, which is also based on the latitude-longitude grid. This sampling design was chosen using the assumption that smaller units, well distributed would fit better the information requirement than bigger units randomly selected. This was also confirmed by the analysis by Erkki Thomppo carried out for the Kotka IV.
After all the above presentation the group discussed in details the proposal of the information framework.
a) Sampling design and intensity
It was felt that that the sampling design indicating uniform distribution of samples over the space will lead also lead to cases where it is too less or too much to produce country estimates (case of small countries). It also raised the issue of need to have varying sampling intensity in the high latitudes because for a given land surface more sampling units are allocated in areas closed to the poles. The sample intensity should be reduced over certain latitude.
The group identified that the sample design of the information framework is a basis for designing other samplings, global, regional or local, and according to the application. A stratification can be applied to, for instance, have lower intensity in deserts and low forest cover areas, or higher in intensity in deforestation hot-spots.
Concerning the proposed size (10 x 10 km2) of the site for the information framework, was observed that it would not be enough for spatial studies such as the ones on corridors, but is big enough for the observation of land use and land management, and for the analysis of forest biodiversity. It was also stressed that the context is very important for the interpretation of the data so a buffer zone could be applied. The issue of sites falling at the junction of two Landsat scenes shouldn't be too crucial since the dimension of the site is small. Overlaying the WRS2 grid with the sample location could allow estimating if many site areas would fall between two scenes.
b) Linkages with NFA
Further that it is a global initiative which is not focused on country estimates. However, efforts may be linked with National Forest Assessments. For instance, in Cameroon, the current National Forest Inventory could be complemented with an assessment based on remote sensing data. In a near future, the number of National Forest Assessments supported by FAO could increase up to 20 countries.
c) Application and neutrality
It was also stressed that the approach is global and should not make any difference between developing and developed countries. The idea to use a common and global framework would serve the neutrality.
It was highlighted that applications from the Information Framework may be decentralized and carried out locally with country involvement.
The Information Framework could become a verification tool for the Kyoto Protocol.
d) Data maintenance and access
The importance of maintaining the data sets over time and having them accessible for all was noted. It would be also suitable to release all the data as contribution to scientific and other communities, without presenting them as country data. There are a lot of potential interested users for the contents of the information framework: all Landsat scenes acquired could be made available to all users since there is no copyright for Landsat data.
It was noted that the amount (about 70 Gigabytes) of data required by FRA for the Information framework is not a lot (13 000 sites of 10 km by 10 km for three dates).
e) Data acquisition and processing
It was agreed that the Information Framework could start with existing data sets as the Landsat 2000 and Landsat 1992 sets are already available and global coverage would be suitable. David Skole proposed that GOFC contribute in the image extraction, selection and processing. Initially, 5 to 10 years interval may be appropriate for monitoring purposes.
Once the framework is set up and proves its efficiency, the process should start with new data sets. A procedure needs therefore to be established for the acquisition of these new data sets as well.
It was stressed that global coverage is needed to get the 13 000 samples plots. John Townshend mentioned that United States national archive may be insufficient so ground stations archives could be needed and that GOFC-GOLD will need arguments for asking a complete mid-decade (2004) global Landsat coverage orthorectified to NASA. 1992 and 2000 Landsat global coverage data sets were produced without a specific goal. It will be necessary to acquire the Landsat 2004 data set following a user-specified model. FRA needs to define the required specifications for acquisition of the imagery.
The request for a new full Landsat coverage should come quickly. Input from UNEP can contribute to facilitate the process. It was reminded that 2005-2008 will be an active acquisition campaign for Landsat.
The value of having other spatial agency than NASA coming into the system was underscored. For instance, IKONOS, IRS and European satellites should be involved in the information framework. Very high resolution satellite data such as Spot 5 could be very interesting within the information framework, as well as airborne data.
Due to the large number of scenes, the time for processing the data (extraction, etc) will be important. Therefore, it was decided simultaneously start work on the application as well.
f) Potential JRC contribution
Frédéric Achard stated that the JRC could possibly contribute in the implementation of the forest monitoring application in Europe, Africa or Russia. He reminded that the JRC will probably have the mandate to manage spatial information on forests over the territory of the European Union. The European Commission might be interested in providing financial support for the analysis of European sample locations. JRC could contribute more in the application side than in the data provision, and apply the same (or at least compatible) sampling than the FRA.
He also suggested that FAO could address a request to Geoland to develop automatic procedure for co-registration of the imagery.
g) Pilot study
The benefit of starting with a pilot study to test the methods and begin the work was outlined. Central Africa was proposed since NFA activities are currently undertaken in that subregion (Cameroon). This pilot study could use the 1992 and 2000 Landsat data sets available in the information framework to set up the prototype for image extraction (first phase) and their analysis for forest monitoring (second phase). Especially, it could be a way of verifying that the 10 km by 10 km size proposed for the sample sites is suitable.
It was suggested that someone in Ashbindu Singh's office (USGS) could start working soon on the image extraction for the pilot study.
Concerning funding, small costs will be involved in the first 6 months since no data have to be bought. The main cost will be the data interpretation for forest monitoring for the pilot study in Central Africa.
