Discussion

The present study constitutes the most extensive analysis of deforestation and forest fragmentation ever conducted in Chile. The results demonstrate how changes in spatial patterns of the forested landscape may be assessed using multitemporal satellite data, as has been achieved recently in other areas of the world (Tommervik et al., 2003; Rees et al., 2003).

Assessment of methods and results

As classifications of satellite imagery into land cover types are never completely accurate, analyses of forest loss and landscape patterns of the present study are affected by errors in the classifications. According to the confusion matrices (Appendix 1), image accuracy tended to improve as the image date became more recent. This trend may be related to image quality, which is better in TM and ETM+ images. For the original image, the accuracy assessment showed that the native forest cover is slightly overestimated which can affect the analyses derived from the 1975 map. However, this might be related to the availability of concurrent land cover maps for the earliest images.

For instance, the accuracy of the 1975 map was assessed using aerial photograph-based maps for the year 1978. This difference in time may have caused a higher error commission than error omission, as an important area classified as native forest in 1975 had been converted to pine plantations or shrublands by 1978. This rapid conversion is more accentuated in the earliest images as this is when considerable land cover changes occurred in the study area. The high percentages of overall accuracy of the images revealed that the supervised classification, which was strongly supported by ground-based information, provided a suitable identification of land cover types in each of the satellite scenes processed.

Although the classification conducted for the oldest scene (MSS 1975) suffered from the disadvantage of limited ground validation, a cautious interpretation considering field control points that had not changed over time enabled high values of accuracy to be obtained, and clearly identifiable categories of land cover in the classification. In order to manage these errors propagated through the analyses of change and patterns, some measures were adapted from Brown et al. (2000).

In the present study, the error was minimised by applying improved topographic correction algorithms in the image processing and by aggregating some of the sub-categories of land cover types. Additionally, different ground-based surveys conducted from previous works and additional field visits were used to support the image classifications. Also, the error was minimised by filtering the classified image to remove small patches of less than four 30 m-pixels (equivalent to 0.36 ha).

source: Rapid deforestation and fragmentation of Chilean Temperate Forests
by Cristian Echeverria a,b, * ,1 , David Coomes a , Javier Salas c , Jose´ Marı´a Rey-Benayas d , Antonio Lara b , Adrian Newton