Before discussing the mechanics of change detection, it is very important to examine and consider the spatial and temporal characteristics of any detected change. It must be stressed that whatever change detection technique is employed, quality control is essential to meaningful analyses and interpretations. The process of change detection can generate a number of errors and questionable results, with misregistration and inconsistent classification being the most common problems. When a change detection result suggests that 10% of an area has changed from grassland to mature forest in just a few years (an impossible scenario), artifacts from poor quality control during processing should immediately be suspected and addressed. Benefit from change detection techniques will only be realized with careful use of change detection methods and cautious, thorough analysis and interpretation of results.

2.1 Spatial Characteristics

The spatial characteristics of a case of change refer to the patch size over which the alteration occurs, as well as to the spatial continuity of the changes. For example, in Figure 7.01 on the southern fringe of the Amazon delta, both large, rapidly cleared commercial patches occur next to smaller plots cleared more gradually. Though it is easier to show large changes than small ones, efforts toward assessing the small patchiness could be worthwhile in two respects: first, in total the effect of little changes may be much greater than those of big changes, and secondly they present a far greater challenge in terms of field detection and redress against perpetrators (if any), so that the independent detection afforded by remote sensing often provides sole irrefutable evidence of activities and are thus quite valuable.

2.1.1 Resolution and scale -- The patch size of changes in vegetation cover dictate the type of data that can be used effectively for detecting and monitoring dynamics. For instance, if the size of the areas undergoing change were one hectare (ha = 100 m x 100 m), then imagery with a spatial resolution appropriate for monitoring areas of that size should be selected; a rule of thumb is that 4 pixels are needed to confirm identity, and ten pixels are needed to measure acreage. The ultimate choice of appropriate spatial resolution is influenced by the size and shape of the changing areas, as well as the spectral properties of the cover type before and after alteration. For example, in comparing areas that were forested and then converted to bare soil, the contrast between pre- and postchange images would be great, so a slightly larger pixel size (coarser resolution) could be used. If on the other hand, the change of interest was the growth of a secondary forest back to a primary forest, the contrast between a time series of images would be less, so a smaller pixel size (finer resolution) would enhance the prospects for detecting change.

When there is mention of the spatial scale of a change event (as different from the spatial scale of a map), it usually refers to the surface area of that event relative to the entire area of interest. It therefore addresses the issue of how much of the area is being changed. This should not be confused with the patch size of the area being altered. For example, in the rainforest region of central Africa, the area of interest is very large (i.e., the Congo Basin), whereas individual areas of deforestation are often not very large. We can refer to African deforestation as local scale, whereas South American deforestation, and deforestation in the United States Pacific Northwest are broad scale.

2.1.2 Fragmentation -- Another spatial characteristic of a change in vegetation cover is fragmentation. Fragmentation refers to the continuity of areas that are being changed relative to unchanged areas. A highly fragmented area is one where the landscape has been converted to a patchwork of different cover types with no large contiguous parcels of the original vegetation cover remaining (see Figure 7.02). The degree of fragmentation can greatly affect the biodiversity of an area by severing dispersal corridors; disturbing the continuity of nest sites, hunting and mating; increasing predation and gradients of physical/environmental conditions by increasing the relative amount of perimeter around a habitat; and/or by reducing the habitat areas below the minimum required to maintain viable populations.

Many studies have documented the negative effect of habitat fragmentation on biological richness (species abundance) and diversity. One of the more dramatic examples is the decline in neotropical migratory birds as a result of habitat fragmentation in their tropical winter nesting and foraging sites. Other documented cases of biodiversity decline associated with habitat fragmentation include grizzly bears in Montana, mountain lions in New Mexico and California, and spotted owls in Oregon and Washington.

2.2 Temporal Characteristics

Two questions dealing with the timing of change detection analysis must be addressed: first the optimum seasonal time to observe an area, and secondly the frequency at which dynamics should be monitored.

2.2.1 Resolution and scale -- Temporal resolution in remote sensing refers to the observation frequency, that is, the number of observations per unit time. The optimum temporal resolution is usually determined by the expected rate of vegetation change. For instance, if an area is uninhabited, the frequency for data acquisition could be very low since changes in land use most likely would occur at a slow rate. However, if a road has recently reached a previously remote area and settlers are moving in along the fringes, a more frequent acquisition rate would be needed to keep pace with the rate of change. To monitor the effects of an insect attack on a particular crop, image acquisitions would probably require a high (frequent) temporal resolution (perhaps weekly) since changes in vegetation condition may occur quickly.

It is important to select imagery that meets or exceeds the temporal resolution needed by the occurrence being monitored. First, the platform carrying the sensor must be capable of observing the area of interest at the desired frequency and for a sufficient period of record. This can pose a problem with some satellite platforms due to their minimal repeat interval of two weeks or more. The weather of an area also plays an important role if passive sensors are being utilized, because a locale experiencing frequent cloud cover may be difficult to image successfully at the required intervals. This is especially problematic for flood and storm studies, and other events that occur during particularly cloudy conditions.

2.2.2 Optimum time for observation -- For overall contrasts between cover types in the humid tropics, the optimum time is usually the dry season because of decreased cloud cover and the increased contrast between seasonal and nonseasonal vegetation. In temperate forests the optimum time might be after the deciduous trees have lost their leaves so it is easier to differentiate between evergreen and deciduous trees. In some instances, it may be desirable to acquire imagery at several times throughout the year to distinguish vegetation types that emerge early, senesce late, or have special flowering or leaf discoloration periods. To avoid confusion between similar land cover types that appear different in some seasons, it is usually best to compare multiyear imagery acquired during the same season.