PoliMappers/mapping deforestation

Introduction

by Professor Maria Antonia Brovelli, Professor Ludovico Biagi, Eng. Gorica Bratic (research fellow at Politecnico di Milano) and Lorenzo Stucchi.

Premises:

1. Land Cover (LC) is a term that describes material that covers the Earth’s surface. The description can range from very general (e.g. forest, non-forest) to the very detailed one (i.e. Broad-leaved forest, Coniferous forest, Mixed forest, etc).
2. LC is an Essential Climate Variable, i.e. a variable which helps to study the problem of climate change
3. The number of LC maps, is increasing lately, which is an indicator of them being in high demand, which further implies usefulness of the information they provide. What is more important, several global high-resolution LC maps have been produced produced. These maps are, therefore, describing the whole globe with a higher level of detail. Thus, they allow us to check the areas of the world where there have been the most critical changes.
4. An example of high-resolution Land Cover map is the S2 Prototype Land Cover 20m Map Of Africa (A) provided by European Space Agency (ESA), another one is the map computed by the National Geomatics Center of China called GlobeLand30 (B). These global maps (with resolution 20 m and 30 m respectively) are characterized by several classes (10).

The former (A) has:

• no data
• Trees cover areas
• Shrubs cover area
• Grassland
• Cropland
• Vegetation aquatic or regularly flooded
• Lichen Mosses / Sparse vegetations
• Bare areas
• Builtup areas
• Snow and/or ice
• Open water

On the opposite the latter (B) has:

• Water bodies
• Wetland
• Artificial Surfaces
• Tundra
• Permanent snow and ice
• Grass lands
• Barren lands
• Cultivated land
• Shrub lands
• Forests

Availability of maps

The available Land Cover maps are the product of classification of satellite/aerial imagery. Classification outcome is affected by classification algorithm realization, quality of satellite imagery, training dataset completeness and quality, heterogeneity of LC classes, etc. For all these reasons, additional information about LC is valuable. In fact, except from the independent use, OSM data related to the LC can support LC maps production by having a role of training data or a role of reference data for validation (i.e. accuracy assessment). Deforestation could be a phenomenon to begin with in order to show the importance of LC. Deforestation affects climate, biodiversity, soil erosion, and it needs to be regulated. Information about LC can be useful to understand the phenomena and to make adequate policies that will keep the phenomena under control. To study the problem of deforestation, first of all we can check if the area covered by the Forest class is reducing over time. Furthermore, we can check if in the middle of the forest the following LC classes appear:

• Bareland (Bare areas in A or Barren lands in B)
• Plantation (Cropland in A or Cultivated land in B)
• Artificial (Built up areas in A or Artificial Surfaces in B)

and monitor the evolution of these classes in time as a proxy to deforestation cause. We decided to define these three classes because there are no international standards about LC classes. A proposal for these classes was already provided here. We wanted to define the classes in a broader sense in such a way not to be too demanding for photo-interpretation from satellite imagery (related to mapathons), and also to keep the classes as similar as possible to the available global high-resolution LC maps.

Tagging landcover in Amazonian Forest

Considering the previous aspect and to map an area near a forest where no data are present and not everywhere good resolution imageries are available our idea is considering some general class that can be after improved by satellite images with more resolution or local knowledge.

Bare Land

Bare land general class that can be mapped in 2 main different types:

• beach/sand area cover by sand natural=sand to not be considered to the Amazonian area
• area with rock and no vegetation natural=bare_rock.

This can be hard to identify in large area with not optimal images so the proposal is to create the tag landcover=barren.

Artificial

Easy to identify also from satellite images so possible to distinguish between the different subclasses:

• mining area that can be mapped as landuse=quarry.
• areas with houses landuse=residential.
• areas with industrial landuse=industrial.

This classification can be easily done in an area with imagery with good resolution, but in some area where the resolution can be bad we think that the tag landcover=artificial can be useful.

Plantation

The key crop=* is very used, but needed a local knowledge of the area, in combination with the tag landuse=farmland or the tag landuse=meadow, that also needed to have local knowledge of the area. So there is the proposal to use the tag landcover=cultivated.

Forest

Two types of classifications exist:

• natural=wood for not managed woods like for the main part of the Amazonia.
• landuse=forest for managed woods maintained by humans.

Also, this information partially needed a local knowledge so the more general tag used can be landcover=trees, also quite used as shown in taginfo.

Comments

Please comment on the discussion page.