Forest degradation, a methodological approach usingremote sensing techniques: literature review

3. Mapping forest degradation

3.1. Remote sensing in forest degradation

In the idea of the study of Kauppie et al. (2006) the way to quantify change in the forest is to select four forests attributes area, volume, density of growing stock, biomass, and sequester carbon) that provide a useful starting point for global forest monitoring. According to Cogalton (2007), these dates are particularly essential when attempting to estimate forest volume, biomass and carbon using remote sensing technology.

11

Lambin (1999) explained the role of remote sensing in forest degradation, in this study, the author is trying to explain how spectral, spatial and temporal information could be used to access the concept of forest degradation.

In the study of Herold et al. (2009) they explained that mapping forest degradation with remote sensing data is more challenging than mapping deforestation because the degraded forest is a complex mix of different land cover types (vegetation, dead trees, soil, shade) and the spectral signature of the degradation changes quickly (i.e., < 2 years) (Souza et al. 2009 in herold et al). High spatial resolution sensors such as Landsat, ASTER and SPOT have been mostly used so far to address forest degradation. However, very high resolution satellite imagery, such as Ikonos or Quickbird, and aerial digital imagery acquired with videography has been used as well. Methods for mapping forest degradation range from simple image interpretation to highly sophisticated automated algorithms (GOFC-GOLD,2008).

Higher spatial resolution imagery is more suitable for detection of specific forest degradation impacts. For example, Ikonos imagery can easily detect forest canopy structural damage (Read et al., 2003; Souza, Jr et al., 2005), but, given the cost for image acquisition and computational challenges to extract information from these very high spatial resolution images, their use in operational applications such as monitoring logging is limited.

3.2. Forest change detection analysis

Various classifications of change in forest ecosystems have been proposed. Aldrich (1985) approached the variability in forest cover from a thematic angle, enumerating nine general forest disturbance classes: no disturbance, harvesting (areas subjected to timber removal operations), silvicultural treatments (e.g., thinning), land clearing (vegetation removal and site preparation), insect and disease damage (epidemic conditions), fire (prescribed burning and wildfire), flooding (man-caused and natural), regeneration (artificial or natural), Other (not fitting any of the above categories).

Coppin (2001) tried to explain different types of change detection in forest ecosystems with remote sensing using digital imagery, he used many techniques change detection, image acquisition, data reprocessing for change detection methods, multidimensional temporal feature space analysis, image differencing, etc.

Since early days of earth observation systems, various techniques of change detection have been developed for forest monitoring using high resolution optical remote sensing. These approaches are focused on the identification of forest cover change, described by Geist (2006),

12

Highly Detectable		Detection limited & increasing data/effort		Detection very limited
·	Deforestation	·	Selective logging	·	Harvesting of most non-
·	Forest fragmentation	·	Forest surface fires		timber plants products
·	Recent slash-and-burn	·	A range of edge-effects	·	Old-mechanized
	agriculture	·	Old-slash-and-burn		selective logging
·	Major canopy fires		agriculture	·	Narrow sub-canopy
·	Major roads	·	Small scale mining		roads (<6m wide)
·	Conversion to tree monocultures	·	Unpaved secondary roads (6-20m wide)	·	Understorey thinning and clear cutting
·	Hydroelectric dams and other forms of flood disturbances	·	Selective thinning of canopy trees	·	Invasion of exotic species
·	Large-scale mining

Table 3: Forest degradation activities and their degree of detection using Landsat-type data, Source:

Peres et al.(2006).

The literature indicates that forest canopy changes can be detected by a variety of analysis methods. Although most methods provide generally positive results, few studies have compared and evaluated alternative approaches. Since Singh's 1999 paper, two recent studies have attempted to determine what change detection method is most appropriate.

Using SPOT multispectral, multitemporal data, Muchoney and Haack (2004) compared four methods, merged PCA, image differencing, spectral temporal change classification, and post-classification change differencing, for identifying changes in hardwood defoliation by gypsy moth. Defoliation was most accurately detected by the image differencing and PCA approaches.

Collins and Woodcock (1996) have compared three linear change detection techniques, multitemporal Kauth-Thomas, PCA, and Gramm-Schmidt orthogonalization. Better and similar results were obtained with the multitemporal Kauth-Thomas and PCA methods than for the Gramm-Schmidt technique; however, the authors recommended the Kauth-Thomas approach because it identifies change in a more consistent and interpretable manner. These authors also examined to what extent the digital images should be preprocessed.