Accuracy Data on Gain and Loss of Canopy on GlobalForestWatch.org

Does Pugu Forest Recover after all development efforts have failed and the forest seems forgotten? Or does the remote sensing technology create illusions?

The interactive Global Forest Watch Interactive Map (GFW Map) provides blue pixels for Gain and Purple pixels for areas with canopy Loss.

A second Layer containing polygons indicating conservation areas can be added to the map and a layer indicating the highest points of the canopy.

If accurately representing the status and health of the forests on global level this map is a fantatsic tool. In order to check the accuracy of the provided data on the GFW Map a transparent overlay of the data (Gain/Loss of Canopy and Tree Height Layers) is made and compared with the historical Google Earth imagery which shows visible Forest Loss and Gain while individual trees can be identified on the images. The first assessment of the data do suggest that there is still much room for improvement as is discussed below.

1. In analysing the GFW data a first issue is georeferencing. Often the Google Earth (GE) images are not accurately located on the (virtual) Earth. This can be checked by looking at a recognisable feature on GE like a road crossing or building and upload various historical images of the same feature which may jump to another location for each image. The location (georeferencing) can introduce inaccuracy. Similarily stretching or squeezing of the projected images may result in displacement of objects on the image. The displacement can also happen with the layers obtained from the Global Forest Watch and the inaccuracy will contribute to inaccurate or wrong conclusions if based on two inaccurate layers (GFW and GE) from different data sets. To reduce the inaccuracy, GE has been selected as reference image and the GFW layers are adjusted to GE (in this case the GE image of Pugu Forest area dated 09/07/2014 is selected). The result of combining GE and GFW Gain/Loss layers is shown on the image on the right. (click on image to enlarge)

The blueish transparent area are "conservation areas" on the GFW Map, in this case Pugu Forest Reserve, Kazimzumbwi Forest Reserve and Ruvu South Forest Reserve. While we are interested in the Purple (Loss) and Bleu (Gain) of the forest canopy we have corrected (on the basis of the GE imagery) the GFW Gain/Loss layer which can be achieved by fitting the GFW overlay with the features in GE. A good reference feature is the expanding quarries which match well with some of the larger Purple Areas (Loss) in the GFW map. The boundary of the conservation area (Forest Reserve) is not accurate and this seems to be a georeference issue on the GFW map but not important at this stage.

ForestReservesDar-es-Salaam
   

2. A specific section in the Forest Reserve close to Kisarawe Town (on the left) is selected to check the individual Gain and Loss pixels. The image on the right is self explanatory. Areas with purple do correspond with for instance a quarry which seems to have been expanded during the period between 2001 - 2012 (Purple Circle in upper right corner). The successful matching (georeference) of the GFW layer with the GE layer is demonstrated by the greyish track in the GFW layer which corresponds well with the same road on the GE image. (click on image to enlarge)

Some of the Bleu pixels do seem to make less sense. while most fall in areas were trees have been removed and biomass reduced compared to other areas. The White Circles in the image indicate undisturbed forest areas were Gain (of trees over 5.m length) would be expected. It will be interesting to learn what the bleu pixels, which originate from LANDSAT ETM+ data, reflect. According to the information provided, trees of more than 5m which were not there at the beginning of the period (2001). It is expected that trees planted on cleared ground can be remotely measured after reaching some biomass (5 meter length??). This can well be gardens or farmland or even invasion of alien trees in a cleared forest sections, rather than Forest Recovery.

It may be more difficult to measure the undisturbed or recovering sections of the Forest which will already have a higher biomass (being undisturbed or less disturbed) but that information is of interest perhaps even more than almost cleared areas in the forest which recover (increase of tree over 5m) which like in the case of Pugu Forest will be alien invaders in most locations.

GFW-Loss-Gain-Pugu-Kisarawe-Section
   

3. The third GFW Map layer which is considered is the "tree height" layer which according to the provided information is based on Lidar (Laser) data measuring the highest point in the canopy. It is assumed that the darker green pixel of 1km2 represent areas with higher trees. The darkest pixel in the investigated section of the Pugu Forest starts just above the white elipse in the upper left corner of the previous image. Considering the necessary inaccuracy of the data on a global scale the data seem to match well with this visible stand (GE) of taller trees in the forest. A second green pixel more to the south indicates a similar displacement of this layer with a displacement of about one kilometer to the North while shifting these pixels to the south would correspond well with a dense forest areas.

What is not understood from the "tree height" layer are the numerous green pixels (1km2) which are located in the high density unplanned settlements of Dar es Salaam which are seen as whitish areas at the right. The white pixels are unpainted galvanised corrigated sheets used for roofing in the high density outskirts of Town leaving little room for trees. (for detailed view click on image).

Overlay-GFW-TreeHeight-Layer-Pugu