Monitoring of natural forests in Madagascar
About the Project
In less than a hundred years the forest cover of Madagascar, one of the world’s most important hotspots of biodiversity, has fallen dramatically. The remaining forests serve as the last local and regional refuge for numerous species and represent important natural resources for the local communities. These forests have become extremely fragmented, which makes them very sensitive to human pressure and the impacts of climate change. Due to poverty and high population growth, lack of fertile land and water, political instability, as well as difficult climate conditions it is a daily challenge for the people to get a decent income in the rural areas. Often the only way to sustain a livelihood is to get it from the forest. Loss of forest is, however, causing increasing problems with e.g. access to clean water and resilience to extreme weather conditions. Thus, it has become increasingly acknowledged that preserving forests is one important piece of the puzzle in efforts of reducing absolute poverty and supporting sustainable livelihoods.
Protection, restoration and monitoring of natural forests in Madagascar
The purpose of this project is to develop a method for the conservation of small forest fragments in the dry areas of Madagascar, while providing a means of sustainable development for the local communities. It also aims at creating a system for monitoring the forests that, based on satellite images freely available in the internet, could be easily adopted by local organizations working with limited resources. The main parties co-operating in the project are The Finnish Association for Nature Conservation (FANC) and the NGO Vakanala in Madagascar.
• Background
In less than a hundred years, the forest cover of Madagascar, one of the world’s most important areas of biodiversity, has fallen from 90% to 15%. The remaining forests serve as the last local and regional refuge for numerous species and represent important natural resources for the local communities. However, these forests are extremely fragmented, making them very sensitive to human pressure and the impact of climate change. Faced by the consequences of this critical situation – loss of arable land and of groundwater – the communities of Menamoty Iloto, a rural town in the southwestern part of Madagascar, have approached the local NGO Vakanala in an effort to preserve their forest.
Globally the project consists of two branches:
• Reforestation, aimed to small forest fragments conservation and restoration for the preservation of biodiversity, soil preservation and groundwater aquifer sustainability and to the development of sustainable agriculture model intending to replace the traditional slash and burn practice;
• Monitoring, aimed to developing a tool for any stakeholders to map and monitor the forests easily and without operational costs. Over 2012-2015, the monitoring branch is also aimed to creating a detailed map of the forests located in the western part of Madagascar.
Since 2011 Transparent World has participated into monitoring branch of the project. During this period our specialists have developed the idea and technical workflow of mapping process, took part in field trips, conducted seminars and GIS training for local partners, have created the map of forest and comprehensive system of gathering and sharing the data, including FTP-database and geoportals.
• Forest monitoring: general strategy and objectives
One of the central weaknesses in previous forest protection and reforestation attempts in Madagascar has been the monitoring of the forests, and this is what FANC especially intends to focus on in the project. The idea here is to apply an already proven method, based on satellite images, in the context of Madagascar, thus providing a tool for the Madagascar NGOs and other interested groups to map and monitor the forests easily, on-line and practically without cost. Over the last few years, advanced internet-based applications have made it possible for anyone to access high-quality geographic data that enables effective capacity building also in small-scale projects around the world. Compared with other already existing monitoring tools, the method chosen provides more detailed data that better enables the monitoring of forest fragmentation and transformation, when supplemented with manual mapping. Also, the original data as well as the results of the project will be freely available for everyone to benefit from. Most importantly, this approach provides an interactive mapping base, with open access for anyone to complete the data and learn from others observations.
The method has been developed and improved over the past decade in co-operation with Russian NGOs, who have successfully applied it in campaigning for the protection of old-growth forests in the borealis region. In Madagascar, the corresponding data could benefit NGOs, officials and academics alike, in their efforts to try and preserve the remnants of their natural forests and to monitor the success of reforestation programs. Connecting with these new, innovative methods will also enable the Madagascan NGOs to more efficiently participate as part of global networks of activists and specialists, such as Global Forest Watch.
The objectives of the monitoring branch of the project were to create:
• a detailed map database, created with the highest quality and the most recent satellite data available through the internet;
• open access for anyone to all the existing materials;
• an interactive function that enables updates by users;
• maps to be functionally modified also for the purposes of communication and campaigning.
The long term objective of the project is not only to encourage the participants in using and completing the data available on the internet, but also in developing the method further to match the conditions in other parts of Madagascar and applying it for various purposes, including monitoring and campaigning.
• Forest mapping: methods
Forest transformation/degradation map is required for better understanding the transformation reasons and making decisions for more sustainable land use; at once the existing maps do not directly show the degradation level. The measuring of the level of forest “intactness” / degradation is a technically challenging task as the “intactness” is not directly visible in satellite images and other spatial datasets.
Developing the idea of the degradation level mapping we took into account the next assumption:
• Forest degradation influences the forest canopy structure.
• More degraded forests usually have more simple canopy structure – while more intact forests have more complex canopy structure formed by trees of diverse sizes (at least this fact was proven for humid evergreen tropical forests and for a number of types of boreal and temperate forests of Eurasia).
• Intact and old-growth forests usually have more than 1 layer of closed canopy (Multi-layer – ML). Degraded or secondary forests (usually being regenerated after burn and slash cycle) have a single layer of closed canopy (Single-layer – SL). Intact forests may also have a simple canopy structure (single layer) in specific habitats or landscape positions (for instance, on top of the ridges). The third type of structure - a single layer accompanied with occasional big trees (SL with BT) – may represent selective logged pristine forests of some types of intact forests.
• The forest canopy structure is reflected in satellite images as texture characteristics and a spectral response.
Pic. 1. Classification scheme for Manombo region. 7 forest classes from intact to disturbed compared to canopy structure classification
Therefore, we use canopy structure approach to define the level of forest degradation from space. It means mapping tree crown shadows marking irregularities in the canopy (a kind of “reversed” single trees crown mapping method), calculating the density of shadows and classifying forest stands polygons on this base. SPOT-5 panchromatic channel (2.5 m ground resolution) is detailed enough to detect single big trees crowns and canopy irregularities. The core of the algorithm is what we define the density of shadows by calculating local minimum reflections pixels density. Firstly we got the polygons by segmentation of SPOT images spectral channels (spatial resolution – 10 m). Than we defined pixels with local minimum reflection in SPOT panchromatic channel (resolution – 2.5 m) within 5x5 pixels window and selected canopy shadows of different size represented by local minimum pixels with the reflection below 60-80 DN. We calculated density of shadows for each polygon/area (as shadow pixels number / area of a polygons * 100) and have got the maps of forest structure complexity based on density of shadows.
Pic. 2. Algorithm details. 1. Selecting canopy shadows – local minimum pixels with the reflection below 60-80 DN in SPOT panchromatic channel (resolution – 2.5 m). 2. Map of forest structure complexity based on density of shadows after counting of density of shadows. Polygons are result of segmentation by SPOT spectral channels (spatial resolution – 10 m).
According the canopy complexity measurement the area was divided into the next classes: non-forests (shadows density below 0.1), single layer (SL) forests (shadows density 0.1-0.2), single layer forests with big tress (SL BT) forests (shadows density 0.2-0.3) and multi-layer (ML) forests (shadows density over 0.3).
Both, Madagascarian (Malagasy colleagues from local associations and joint Malagasy-FANC-TW teams) and international experts concluded that canopy coverage is another important indicator. The field observations proved that the canopy complexity alone does not reflect the whole diversity of human-made and natural disturbances. So we have extended the classification and taken canopy coverage into account. The coverage has been visually identified with high-resolution images and four classes of canopy coverage (high - over 70%, medium – 40-70%, low 20-40% and very low – below 20%; percentages are indicative) were included into classification scheme.
Intercrossing of canopy coverage and canopy structure complexity classes results in forest degradation classes scheme:
Pic. 3. Forest degradation classes scheme...
Pic. 4. ...and what real forests are in these classes
Other characteristics like the altitude and the location in the relief also should be taken into account. For instance, sparse single-layer forests are natural in tops of ridges at high altitude and indicate a high level of human transformation in valleys and lower slopes.
SPOT spectral channels can also help separating native trees forests from introduced tree species planted, as well as various forest types. Single-layer closed forests may represent forests by native species, pine or eucalyptus plantation; single-layer forests with big trees - mountain forests in Andasibe, Ranomafana areas, low-land forests in Manombo or littoral forests.
Therefore, final classification scheme and list of classes (included 23 classes, 15 forest and 8 non-forest ones; 3 of them are not represented through pilot areas) were developed and can be downloaded here.
CLASSIFICATION SCHEME AND LIST OF CLASSES
Finally, the Random Forests algorithm was employed for classifying the area into 25 classes (23 listed above plus water and rice fields) by all their parameters measured by DEM and SPOT images. The classification workflow consists of the next steps:
1. Segmentation of SPOT images by spectral channels, 10m resolution (GRASS GIS module i.segment with the following parameters: min area – 10 pixels, similarity threshold = 0.3 for spectral channels).
2. Adding main statistical parameters from spectral image channels and DEM data to each polygon: mean, standard deviation, minimum, maximum, range to each polygon; as well as canopy complexity data.
3. Creating training dataset, using high-resolution image and field observation data as reference.
4. Creating model using the advanced decision trees algorithm - R:randomForest model of classification, data statistics for each class, estimate model quality and uncertainties.
5. Applying model to whole set of data and export result map (QGIS).
Forest mapping: results
Pic. 5. Result: overview map for pilot areas
Pic. 6. Result: map of Ranomafana pilot area
Comparing with global data shows good agreements between our results and IFL mapping.
Our maps are currently based on canopy structure features only. It should be possible later to connect certain canopy structure with a certain disturbance level but it will require more field observations. Не знаю, надо ли это на сайте отражать. Думаю,нет?
• Fieldwork
In 2011-2014 Transparent World participated in the join field trip to 8 regions various types of humid and semi-dry forests and human disturbances have been surveyed, marked in GPS, and documented. A lot of georeferenced photos, more than 3000 GPS points (about 2000 of them – from the forest stands) were collected. During the trips digital field forms collecting tool was tested on the PC-tablets.
In conversation with local partners the possible system of forest classes for mapping has been discussed and improved, some important information about landuse methods and history has been gathered.
• Gathering and sharing the data and experience
All our results and materials are available for our partners and local stakeholders (if they had a special requests), except datasets with restrictions of sharing (Spot5 from SEAS-OI). During last year, we share our forest maps and satellite images (more than 200 GB of data) for our local partners. Transparent World supports the FTP-server where we placed the spatial data in GIS format we received for Madagascar from various sources. The information about FTP had been sent to all project partners. Today the FTP-server mainly contains satellite data, available topographic maps, and thematic datasets.
Space coverage containing Landsat, Spot, other sources (for example, Aster, Corona) were placed on the FTP-folder. Transparent World has selected, downloaded and placed into FTP-server all available Landsat 5 images for Madagascar from USGS archive. Spot 5 images received from our partners was obtained and preparing for classification. More when 300 GB of very high resolution images from Openlandscape.info was gathered and preparing for analyze. Current image coverage includes 70 scenes of SPOT 5 for 2009-2014 (SEAS-OI,CNT) and more than 900 scenes of WorldView-1,2 for 2011-2013 (SCANEX).
New SRTM data with resolution 30 meters per pixel were downloaded and merged for the whole Madagascar. New GWF Forest change maps for Madagascar were downloaded and analyzed.
GPS data (points from the field trips and from our partners) were obtained and joined into GIS layers.
Transparent World also participated in the trainings for local activists in 2011-2012. Our role in trainings mainly focused on GIS software using for simple GIS operations, satellite images visual interpretation, transferring data from GPS to GIS systems, etc. In 2014 we shared our approach and classification algorithms, especially QGIS-GRASS-R using for image classification, for our Malagasy partners.
See Manondroala.wordpress.com
Manondroala On Facebook:
Creating a forest map is not a walk in the park
• Partners
|
