GIS Portfolio - Michael Malone

  This weeks module focused on classifying images using multispectral signatures. Above you can see the completed classified land cover of Germantown, Maryland. To create this image above signatures were collected that correlated to each required feature. Then bands were chosen (R:4 G:6 B:5) that contained the largest separation amongst features to minimize spectral confusion. In the above image roads and urban areas were often confused leading to a much larger area of roads than actually exist. The inset map contains a classification distance map which displays the distance each cell is (spectrally) from the sample points with brighter pixels being further than darker pixels. This indicates that brighter areas have a higher chance of error.

  This weeks lab focused on manipulating multispectral images by adjusted which layers are displayed in order to reveal information that is otherwise omitted. The above image shows snowcaps that were identified by searching for a spike in pixel brightness in layers 1-5, a RGB combination of 3 5 4 was used to make the snowcaps contrast heavily with the surrounding area.

  The above map shows the land cover in a section of forest in Washington taken by remote sensors with the area of each land cover category given in hectares. The focus of this module was spatial, spectral, and radiometric resolutions paired with an intro to ERDAS Imagine. The program was used to analyze and isolate the image chosen for the map above. 

  The above map shows digitized land use/land cover (LULC) classifications for Pascagoula, MS. A set of 30 random sampling points were created post creation of the LULC layer and they were verified in google street view to determine accuracy. Based on these points the LULC model was found to be 86.6% accurate.