Multi-Scale Evaluation Of The Region Of Eastern Amazonia Using Ikonos, Landsat, SRTM And MODIS Data
Thomas
A
Stone, The Woods Hole Research Center, tstone@whrc.org
(Presenting)
Ima
Celia
Vieira, Museu Emilio Goeldi, ima@museu-goeldi.br
Arlete
Silva de
Almeida, Museu Emilio Goeldi, arlete@museu-goeldi.br
Eric
A
Davidson, The Woods Hole Research Center, edavidson@whrc.org
Satellite data from four sensors and topographic data of a portion of Para were compared to explore the information gained and lost for identifying land covers at various scales in areas east and south east of Belem. Because much of this area has been occupied for up to 200 years, this is an older area of clearing with few remaining large forest patches.
We used 1 and 4m Ikonos, 30m Landsat ETM+ data, 250 and 500m MODIS data and 90m Shuttle Radar Topography Mission (SRTM) data for this analysis. Also, we developed intermediate resolution scale data by aggregating higher resolution data.
We examined locales where these data cover common areas by evaluating how scale changes (changes in grain size) affect mean pixel values, variance and entropy, a measure of the image texture complexity. Using a 3 X 3 pixel moving window or filter, we have examined all scales and found that with coarser resolutions, mean pixel values decline, and variance increases. However, entropy increases and then declines with the point of inflection for entropy varying from sensor to sensor. In general, entropy is usually lowest at highest resolution, increases rapidly at an intermediate resolutions and then undergoes a slow decline as resolution increases.
Understanding how changes in scale and sensors affect our ability to extrapolate from fine scales to broad scales or basinwide helps us to estimate if information is lost or gained as we change scales.