Geo Leah

In this week's lab, we performed both unsupervised and supervised classification in ERDAS. For the supervised classification, we created spectral signatures both by digitizing polygons and by region growing from seed (given a set of points representative of different land cover classes). To avoid spectral confusion, we looked at the histograms and mean plots for spectral signatures. The three bands with the least overlap between classes were chosen to display the classified image. Based on our spectral signatures, we ran a supervised classification with the maximum likelihood method and recoded to merge like classes. We also looked at the file of spectral Euclidean distance to determine where there were pixels with the greatest distance from the spectral signatures, indicating likely misclassification. The classified raster (along with the areas of each class in hectares) and the distance file for an area comprising Germantown, Maryland are shown in the map below.

For this week's lab, we performed spatial enhancements using low pass and high pass filters in ERDAS Imagine and ArcGIS Pro, looked at image histograms, and used the Inquire Cursor to explore values in multiple bands at once. For an aerial image of part of Washington state, I identified distinctive features and created the following maps with band combinations that emphasizes those features and contrasts with the surroundings.

In this week's lab we explored ERDAS Imagine and how to view and prepare data for mapmaking in ArcGIS Pro. I took a subset of the image from Landsat Thematic Mapper imagery of Washington State that has been classified according to land cover and created this map in ArcGIS Pro.

For this week's lab, we created a land use/land cover map of an area in Pascagoula, MS from an aerial photo. We classified based on Level II of the USGS Land Use/Land Cover Classification System. I then generated 30 random sample points within this area and used Google Street View to assess to the best of my ability whether my classifications were correct. Based on these sample points, I calculated my overall accuracy to be 73.3%. Quite a few of my mistakes were due to me not knowing the USGS definition of class 54, Bays and Estuaries, and incorrectly classifying rivers or streams in this class. Although this is not reflected much in my sample points, I also had trouble distinguishing between commercial and industrial uses based solely on the aerial photo. Although this lab was pretty time-consuming, I enjoyed the process of classification and would like to learn more about it, especially since it is relevant to my interest in urban planning!