Accuracy assessment

chileAccording to the rules suggested by PCI geomatics for ranges of separability values of class signatures, a ‘‘good’’ signature separability was achieved for each image classified, the average Bhattacharrya distance ranging from 1.979 to 1.987. Overall agreement of classification was 82.7% for the 1975 MSS, 83.3% for 1990 TM image, and 84.9% for the 2000 ETM+ image (Appendix 1). The lowest values of accuracy corresponded to shrubland, arboreus shrubland, and native forest categories. These three formations are very similar in their spectral signatures (Bhattacharrya distance less than 1.9) because they correspond to stages in a continuous succession process, which may easily produce misclassifications between the categories assigned in the training site and those classified by the algorithm. Supporting data from other studies and control points were important to distinguish these stages of forest succession.

Major changes in land cover

Changes in land cover were analysed using the area statistics (Table 1) derived from land cover type maps (Fig. 2). The estimated cover of native forests decreased from 119,994 ha in 1975 to 39,002 ha in 2000. In other words, 67% of the native forests existing in 1975 had been replaced by other land cover types by 2000. In 1975 the native forests were distributed throughout the study area. Twenty five years later, these formations were restricted to small patches, sparsely distributed


across the landscape (Fig. 2). Conversely, exotic-species plantations increased from 5% land area in 1975 to 17% in 1990; by 2000 this land cover type was the dominant cover on the map, comprising 36.6% of total land area.

In 1975 exotic-species plantations were located principally in three specific areas within the study area. By 2000, the plantations had rapidly expanded across the landscape, reaching sites at different altitudes and aspects. Shrublands and arboreus shrublands were the dominating vegetation formations in the cover maps of 1975 (33%) and 1990 (58%). Ten years later, these formations decreased to represent 34% of total area, and the exotic-species plantation became the dominant land cover type at that time.

During the first time interval, 36% of the native forest area was converted to shrubland and arboreus shrublands, 29% to exotic-species plantations, and 31% remained as native forest. During the second time interval, a substantial area (50%) of native forest was replaced by exotic-species plantations whereas only 7% was converted by logging to shrublands and arboreus shrublands and 36% remained as native forest. More than half (53%) of the native forest cover existing in 1975 has gradually been converted into exotic-species plantations by 2000; another substantial area (40%) was transformed into shrublands or arboreus shrublands.

From 1975 to 2000, agricultural and pasture lands exhibited a slight increase (Table 1). However, in 1990 the area occupied by this cover type declined to 14% as a consequence of the conversion of land cover types (including pasture lands) to exoticspecies industrial plantations after the promulgation of the law on forestation in 1976. Conversely, shrubland presented an increase from 1976 to 1990 due to the clearance of secondary forests. In 2000, most of this shrubland appeared to be covered by exotic-species plantations.


By overlapping the forest native cover of each year, it was observed that between 36% and 44% of native forest was derived from regeneration of shrubland into secondary forest between time intervals respectively. However, it was noted that between 80% and 90% of this category corresponded to small patches whose area was less than 5 ha. During the whole study period, the annual deforestation rate was 3240 ha year 1 , equivalent to 4.5% year 1 using the compound-interest-rate formula. Most of the forest loss was concentrated in the first 15 years of the study period, at a deforestation rate of 5.06% year 1 , corresponding to 4257 ha year 1 . Between 1990 and 2000, the rate decreased slightly to approximately 3.64% year 1 , equivalent to 1713 ha year .

Variation in forest patch size

One of the basic symptoms of forest fragmentation is the increase in number of smaller patches (Fig. 3). In the Rio Maule-Cobquecura considerable changes were found in the distribution of forest patch size between time intervals (Fig. 3). By 1975, 44% of the forest area was concentrated in a large patch between 20,000 and 100,000 ha; the remaining forest area occurred in isolated patches of less than 10,000 ha, with almost half occurring in very small patches of less than 100 ha. In 1990, 59% of the total area of native forests occurred in patches of less than 100 ha. By 2000, this percentage increased to 69% and only 3% had a size greater than 1000 ha .

Spatial configuration of native forest cover

The highest density of patches was recorded in 1990 with 1.65 fragments of native forest per 100 ha, subsequently decreasing to 1.36 in 2000. During the first period, the native forests were mainly affected by severe fragmentation (increasing largest forest patch (Table 2), ranging from 7% in 1975 to 0.2% of the total area in 2000. Continuous areas of quality habitats decreased following the introduction of disturbed fragments into the matrix. These modifications of the landscape were also characterised by the presence of more patch edges (Table 2).


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