What
Building methodology and outputs to generate forest cover maps, forest land cover classification for forest monitoring and forest detection changes.

Where
Mato Grosso, Para, Acre, Brazil. The state of Mato Grosso is located in the center-western region of Brazil, with an area of 903.357 square kilometers. Most of its territory is inserted in the Amazon Legal region. The Amazon biome comprises 47% of the state. The Mato Grosso econmy is largely based on ranching, mechanized agriculture (specially, soybean and cotton) and logging.

Keywords
LULUCF1, deforestation, forest degradation, biodiversity conservation, protected areas, forestry law enforcement.

Context
Deforestation is responsible for about 20% of worldwide greenhouse gas emissions … Amazon accounts for 50% of the deforestation worldwide. The protection of the primary forests of the Equatorial belt is then a major environmental issue at global scale. In Brazil, a number of actors (NGOs, public institutions, local governments) are engaged in this action: they are working to support the creation and implementation of protected areas in the Southern Amazon, and to generate and disseminate strategic spatial information for the territorial environmental management.
Among these actors, in particular Imazon and ICV (Instituto Centro de Vida) are very active in the development of public awareness through the publishing of bulletins, the setting up of local projects in different states of Brazil (Matto Grosso, Para) and the development of monitoring techniques based on remote sensing.
This project shows recent trends and current status of primary forest ressources in a part of the Amazone forest.

Partners/clients
Imazon, the Amazon Institute of people and the environment is a Non-profit research institution, aims to promote sustainable development in the Amazon through studies, support for public policy formulation, broad dissemination of information and capacity building R&D to develop methodologies aiming to monitor forest degradation. Carlos Souza, the director of Imazon is the co-author of GOFC-GOLD “Source Book” aiming to provide UNFCCC with a comprehensive view of existing technical capabilities to implement REDD mechanism.

Objectives
Creating forest cover maps to provide a baseline for future monitoring of forest ressources using available satellite resources. Providing an homogeneous view of the deforestation process of the area in focus and derive environmental impacts (e.g. carbon stocks ,…). To enforce valuable tools and methodologies to monitor forest legal reserves in the Amazon.

Methodology/Results
Methodology included the following steps:
• Processing of satellite images to biophysical information with the Overland tool
• Generation of NDFI² maps
• Monitoring of reserved areas
• Forest general classification


All multi-source and multi-scale satellite images were processed using to characterize the vegetation and generate the so-called biophysical parameters that give absolute, physical characterization of vegetation and soil in the image.

Shapefiles supporting the Mato Grosso rural cadastre database were also used in order to be able to monitor the forest legal reserves from this database.

In the context of the Amazon forest, there are two major challenges that were properly handled with Overland:
- one the one hand, the images often have degraded atmospheric conditions (cloudy images)
- on the other hand the discrimination to operate between soil and non-photosynthetic vegetation, in order to provide a correct assessment of vegetation canopy density and to identify the forest degraded areas.
The several layers output of the biophysical processing wer able to provide the keys for discrimination of the different target classes, using a decision tree.

The following steps have been demonstrated:
• Generation of biophysical maps to characterize the forest vegetation. Overland is a unique tool today to perform biophysical processing from different sensor data and to generate LC caracterization of natural areas.
• Generation of maps of a reference forest indicator – the NDFI (Normalized Difference Fraction Index2) – in a robust and automated way, with powerful mechanisms for cloud and cloud shadow masking and for correction of local atmospheric effects within the image (e.g. smoke fumes, local thin clouds, …).
• Automated processing of reference GIS data (i.e. the shape files of defined forest legal reserves) to output table file information to monitor these legal reserves.
• and finally the development of robust classification schemes to perform initial forest land cover where this information is not available.

This pilot project also illustrated the possible complementarities between different sensor sources; images from medium resolution sensors (e.g. MERIS or MODIS) can provide regular information at a larger scale (up to the global scale) to trigger acquisition of high resolution images when necessary.

Way ahead - The next step would be now to design and prepare the way for a solution that shall provide a state-wide capacity for deforestation monitoring and be further extendable to encompass a number of applications in the connected domains of overall state land management and carbon stock assessment.

Data used
6 Landsat images (2000-2004), One Spot scene (2007), 6 Formosat-2 images (2007).

Land-cover classification

The objective of forest classification is :
- to discriminate forested areas from non forest areas (agricultural areas, bare soil),
- to discrimininate intact forest and degraded forest, but also to identify if possible other intermediate states of the forest,
- to map the burnt areas



¹ Land Cover, Land Use Change & Forestry
² NDFI is the Normalized Difference Fraction Index was developped by Imazon for the detection of forest canopy damage and it synthesizes information from several component fraction images.