Detecting changes in levels of vegetation for the Brazilian Atlantic Rainforest using Kauth-Thomas transformation and Change Vector Analysis (CVA)

Manuela Rayner and Lily Ray (2006)

INTRODUCTION

Remotely sensed imagery is widely used to monitor changes in land surface conditions, particularly in highly vegetated environments.  Monitoring change in vegetation between two time periods can assess the health and vigor of forest and plant species, assess vegetation growth and regrowth or quantify forest loss caused by deforestation from large-scale subsistence farming. One of these new methods is the change vector analysis (CVA), which summarizes the change in imagery on a pixel-by-pixel basis, according to differences in reflectance on a number of spectral bands.

AIMS

The aim of the project was to explore sophisticated methods in monitoring degrees of change in vegetation in the Brazilian Atlantic Rainforest.

METHODOLOGY

Change detection was quantified by applying the Change Vector Analysis (CVA) to Kauth-Thomas transformed data.  Scenes of Landsat TM and ETM+ (1993 and 2000) data were acquired for this project and covered an area where  Iracambi is located. Imagery used consisted of a Landsat TM scene of the area around Iracambi from June 1993, and a Landsat ETM+ scene from August 2000 of the same area.  The two Landsat scenes were atmospherically corrected and the 1993 image was then georegistered to the 2000 image using bilinear resampling.

RESULTS

The direction image allowed for estimation of areas that had experienced deforestation, reforestation, or persistent levels of vegetation, and the magnitude image distinguished between areas of low, medium, high and extreme change. These two images were then combined with the CROSSTAB module in IDRISI to estimate the degree to which each area had experienced a particular type of change.

CONCLUSIONS

Our analysis show that there is significant vegetative change occurring in the Brazilian Atlantic Rainforest.  The almost 12% deforestation rate is very high, considering that the changes were measured over a relatively short period of seven years.  The nearly 29% reforestation rate would appear to be a positive sign, but since this is only representing seven years of change it can only represent small secondary growth.  It may also represent agricultural plots that are in a fallow period.
It is unclear at this point whether a change vector analysis using broad categories such as this one, which are determined only by general direction (increase, decrease) and not by vector angles, would accurately report the conversion of primary forest to secondary growth as persistence or as deforestation.  In order to assess the accuracy of the analysis, ground truth data or aerial photography is necessary, and so at this point it is impossible to determine the strength of MKT change vector analysis in distinguishing vegetation change in the Brazilian Atlantic Rainforest.