Accuracy of Landsat derived deforestation maps

I have received a number of comments from colleagues working in Mexico and Central America regarding the accuracy, or lack thereof, of Hansen et al’s recently published deforestation map.

Many of the comments regard the timing of deforestation. Pixels shown as recently deforested were often regarded by those with ground based knowledge as having been deforested prior to 2000. If this is generally true then the results may exagerate recent deforestation and fail to pick up trends towards falling deforestation rates. The short paper published in Science does not include any details on the methodology used by Hansen’s group. It does contain a defintion of deforestation
“Forest loss was defined as a stand-replacement disturbance or the complete removal of tree cover canopy at the Landsat pixel scale. Forest gain was defined as the inverse of loss, or the establishment of tree canopy from a nonforest state.”
So one issue may concern how “tree cover canopy” was defined for the start of the period. If areas of highly disturbed, secondary forest and shrubland were included then there would be a tendency to pick up what are in fact “management” activities aimed at removing remnant trees and shrubs from land in the process of being cleared for pasture or the removal of unwanted regeneration and regrowth.

An element of deforestation that is often overlooked is the difficulty of achieving the conversion in many areas. Trees do not necessarily die after being cut down. This is particularly apparent in karstic areas such as the Yucatan. Tree roots penetrate many meters through cracks in the karst. The visible above ground tree may represent only a small proportion of the total woody biomass.

Tree roots 25m below the surface in an underground cenote in the Yucatan.

Tree roots 25m below the surface in an underground cenote in the Yucatan.

Slash and burn farmers in the Yucatan are in essence running a rotational “coppice” system that harvests the above ground component and the contribution made to soil fertility by accumualated organic matter. In fact Hansen’s map shows this very clearly. “Deforestation” in the Yucatan shown in red is compensated by reforestation (blue) or bidrectional change (pink).

Cover change to the south of Valladolid as classified by Hansen et al 2013.

Cover change to the south of Valladolid as classified by Hansen et al 2013.

A fascinating element of the portal provided by Google is the addition of street view. This is available for lareg areas of Mexico including the Yucatan. Thus deforested pixels at the edge of roads can be “visited” in order to inspect the process taking place.

Roads shown in blue are include in street view.

Roads shown in blue are include in street view.

Zooming to a deforested area of around 1 ha from within this region provides the view shown below.

Recently cut milpa in the Yucatan

Recently cut milpa in the Yucatan

Although the white karst gives the impression of severe degradation, this is not in fact deforestation. A 100m move down the road brings us to pixels that are classified as reforested. The vegetation consists of trees such Bursera sp. (simaruba?) of around 10 to 20 cm in diameter that have probably resprouted from rootstock left in the ground.

streetview2

As an aside, the map shows the tradition of maintaining forest reserves around the communities very clearly. Slash and burn begins around one km from most towns. This practice probably was a Mayan tradition, which was reinforced by Spanish colonial laws in some areas and has been further strengthened by recent moves to community forest management.

streetview3

This is clearly a managed landscape which cannot be regarded as part of the “agricultural frontier” in which uncontrolled deforestation is taking place. At the same time large areas around Cancun are clearly being permanently deforested by speculators for development value.

Large scale permanent deforestation to the west of Cancun.

Large scale permanent deforestation to the west of Cancun.

In other areas reprouting trees and shrubby regrowth may lead to a complex gradient of disturbed and degraded forest that cannot be cleanly classified even if visited on the ground.

So, to return to the issue of the accuracy of Hansen’s classification.

The fact that these maps have been placed on an interactive portal equipped with streetview and high resolution background imagery allows anyone to validate the results. Those with knowledge of processes taking place on the ground can provide detailed interpretation and explaination for the patterns shown. The maps may be “wrong” but noone should expect a global classifications at this scale to be completely reliable for any given site. The overall patterns shown are probably fairly reliable at a regional scale. Errors of comission on one side may be compensated by errors of omission. Errors will be specific to different regions and best detected by researchers with local knowledge.

There is a growing trend towards “citizzen science”. One way the web portal could be easily enhanced would be to provide tools that allow users to digitise and classify areas on the map in order to produce their own accuracy assessments and help others to interpret the processes taking place on the ground.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s