Methods .. MODIS Classification

The initial vegetation classification of Madagascar uses MODIS MOD43B4 imagery to separate vegetation classes with a single date surface reflectance image and an entire year of 2002 vegetation greenness (NDVI) data. MODIS imagery has coarse, 1 km, spatial resolution to minimize data volume and enable a broad vegetation stratification of the entire country. MODIS also has a daily overpass so images can be composited to eliminate cloud cover. This is especially useful for the Northeast highland region of Madagascar that is almost permanently cloudy. The spectral resolution of MODIS bands 3,4,1,2,6,7 is identical to Landsat bands 1,2,3,4,5,7. This allows for the integration monthly MODIS data with single date Landsat scenes to incorporate seasonality in the vegetation classification.

	MODIS Classification for NW Madagascar click for full image.

MOD43B4 BRDF adjusted reflectance 16-day composites provides Nadir BRDF-Adjusted Reflectance (NBAR) at the mean solar zenith angle of each 16-day period for MODIS bands 1-7. This reduces differences in specular reflection and shadowing due to changing sun angle. MOD43B4 was acquired for all of 2002 for MODIS tiles H22V10 and H22V11. These two tiles cover the entire country of Madagascar except for a small slice of the southwest section of the country. The MOD43B4 16-day composites were made into monthly composites and NDVI was calculated for each month.

The monthly NDVI was added to a September composite image of surface reflectance. The month of September was chosen because the image has the least cloud cover. The resulting eighteen layer image has six MODIS surface reflectance bands and twelve months of NDVI. A supervised maximum likelihood classification was run on this image with training polygons created using the Royal Botanical Gardens Kew Madagascar Remaining Primary Vegetation map (1995), Conservation International CABS forest cover map (2002), and visual interpretation of the imagery. The initial classification separated 13 land cover classes. These included:

• western dry forest
• limestone
• fragmented forest/agriculture
• southwestern dry forest
• lowland humid evergreen forest
• high altitude savanna
• seasonally dry forest southeast
• mangroves
• seasonally dry forest south
• humid escarpment forest
• savanna
• evergreen woodland
• northwestern dry forest

The second iteration of the classification divides the savanna class into several subsets. These additional classes added to the classification are:

• High altitude savannah
• Savanna southeast
• Savanna east
• Savanna north

The theory of starting with broad vegetation classes gives an idea of what vegetation types can definitely be separated with satellite imagery and what classes can not be detected with MODIS imagery but might be possible with higher resolution Landsat imagery. One example of this is the narrow coastal forests in the SouthEast, which could not be detected with the coarse resolution MODIS imagery.

Class names for this initial stratification describe forests based on location and not species content or structure. A uniform projection is used (UTM zone 38 North, spheroid/ datum WGS 84).

• To download this map
• View this map
• View metadata for this map