Module 10 - Supervised Classification of Germantown, MD

The first portion of this lab allowed us to experiment with a supervised classification.  We used tools such as the signature editor, imported or created and AOI layer, used UTM coordinates to hone in on classification types and used the polygon method versus the grow/grow properties option to "train" the classification tool.  After our first attempt we used the histogram plots and Mean plots to evaluate the results of our spectral signature.  We set the signature colors to approximate true colors using a band combination that would have the least spectral confusion.We then applied and saved our signature file.  We then used the Classify-Supervised tool to classify our image.  We then merged multiple classes of a similar nature to narrow down our actual classes.  Once complete we generated a distance image and a recoded image.  We then used the attribute table for the recode image to add our class names and calculate the area.  It took me a few tries to get the entire process good results.  I then moved on to the actual assignment which was to perform a supervised classification of Germantown MD.  Here is my end result:

Lab 9 Unsupervised classification using ERDAS - UWF high resolution .sid image using only visible bands of light (RGB)

This lab provided an exercise in both ArcGIS and ERDAS Imagine to perform an unsupervised classification.  The deliverable component came from the ERDAS Exercise 2 portion of the lab.  We utilized the Unsupervised tool within the Raster tab Classification group and set the nuber of classes to 50, accepted the approximate True Color for the color scheme and assigned Red as 3, Green as 2 and Blue as 1; we also used a maximum iteration of 25 and a convergence threshold of 0.950 and set the skip factors to 2 for X and Y.  Our next task was to reclassify the 50 classifications we just created via the attribute table.  We selected known items in the image which were the highlighted in the table; we then set the Class_Name field to one of our 5 categories: trees, buildings/road, grass, shadow and mixed; we also changed the color of our new class to something appropriate.  We repeated this process until we were left with items which were difficult to discern and placed them in the mixed category.  We then turned on the original .sid image and used a combination of the Swipe, Flicker, Blend and Highlight tools by first selecting an unclassified item in the attribute table, changing it's color to something distinct and bright and using the tools to aid in identifying the item at which point we could properly reclass it.  This process was repeated until all 50 classes had been reclassified and assigned the required color.  At this point we used the Raster Tab, Thematic button Recode to merge our 50 classes into 5 for our final product.  This new recoded  image was then saved.  The image was imported into my ArcMap geodatabase and a final map created.  We also added a new column for Area to the attribute table.  These were then summed and used to develop permeable and impermeable acreages and subsequently percentages.  This information was included in the final map.

Lab 8 - Thermal and Multispectral Analysis using ERDAS and ArcMap

      After reviewing the use of histograms and learning to combine multiple images in this case obtained using LANDSAT ETM+, we practiced converting the multiple images into a composite using the Composite Bands tool in ArcMap of the Layer Stack tool in ERDAS.  We also reviewed items from past lectures and applied them during our analysis phase.  This included the use, in ERDAS, of multiple views, histogram editing, and using the Discrete DRA tool.  We used symbology in ArcMap and altering color band layer associations in ERDAS during our Multispectral analysis.

      I noticed an area in the Midwestern portion of the image which caught my eye due to both its shape and coloration.  By using the ETM Composite image I created from the ETM layers 1-8, I had the basic image ready to adjust.  Again, I prefer to work in a geodatabase so I imported each layer into the ETM.gdb and stored the ETMComposite there as well.  While running various comparisons made different areas of the image stand out, using the stretch symbology and reviewing each layer, visually I prefer the multispectral imagery.  I set Red to layer 4, Green to layer 2 and Blue layer 6 (the thermal layer).    As I panned around the image I noted a feature south of Guayaquil which was within an urban area, oval in shape with a bright green outline and dark red center.  This became my AOI.

I noted that by using the Stretched symbology the object was most visible in layers 1-3.  Band 6 did not illustrate central red blurs fading to yellows and then surround blues.  I knew the red blurs were potential hot spots, assumed the fading to yellow was a blend of greens and blues and that the outer more distant blues were urban areas.  By switching to the RGB composite symbology as described in the first paragraph, the shape and heat signature were much more clear.  I suspected that this could be some sort of civic area or sports arena.  Knowing the incredible following of Fútbol (soccer) I began to lean toward this area being a stadium.  The coordinate value obtained from the information icon in ArcMap  (Coordinate System WGS 1984 UTM Zone 17N, Projection Transverse Mercator, Linear Unit Meter, Angular Unit Degree, Datum WGS 1984)  is 79°55'29.684” W, 2°9.956"S.  I noted this on the map.  Wanting to confirm my theory, I entered those coordinates into Google earth and discovered that the location was in fact a sports arena known as Estadio Monumental Isidro Romero Carbo AKA Estadio Banco Pinchincha which is in the parish of Tarqui in Northern Guayaquil, Ecuador.

Module 7 - Performing Multispectral analysis and using the NDVI (index) to enhance specific image features

Both the exercises and the final lab were pretty interesting as they gave us insights in how to utilize the multispectral tab, NDVI creation and the use of the panchromatic tab as well as using histograms and adjusting the breakpoints and LUT (visible brightness on screen) Histogram; also worth mentioning is the Discrete Data button which will automatically adjust breakpoints to create a balanced histogram which produces good results for most but not all situations.  We also experimented using a few the tools found in ArcMap, but after a bit of fiddling with ERDAS, I find that I preferred it over ArcMap.  For our lab we were given three distinct features to locate by using the inquiry button and observing pixel values.  Once each feature was located we applied what we learned in the exercises and then created a subset & chip of the area after selecting the optimum band combination for each map.  Although some maps could have used the same band combination we were asked to use three distinct band combinations, one for each feature.  Below are the maps I produce in Arcmap based upon the multispectral analysis performed in ERDAS.

Module 6 - Spatial Enhancement

I found this lab to be quite difficult regarding the final portion.  The exercises were easy to follow and made sense.  Using them in both ERDAS and ArcMap for cleaning up stripting in a Landsat image proved much more difficult.  Despite the lecture and readings, I still felt far out of my league.  I did additional research online, but failed to grasp the high-level concepts.  I did my best and here is my result:

Module 5a Intro to ERDAS Imagine and Digital Data - Producing a class map within ArcMap

This week we learned the basics of the clearly powerful ERDAS Imagine software.  Our ultimate goal was the exportation of a portion of a .tif file of an area within Washington State to an image file for use in ArcMap.  I chose to use a portion of Jefferson County in Washington State that showed several different classes within a small area.  This information was brought into ArcMap, symbology was used to add a description for the area in hectares so that it could both a permanent part of the map and be depicted in the legend.

Module 4 Ground Truthing and Accuracy Assessment

Our goal for this lab was to ground truth our LULC classification map from module 3 utilizing Google Maps earth view and its 3d pan/tilt/zoom features as well as street view feature which had much higher resolution than our original imagery.  I've included my overall approach to the project and my final map:

NOTES:

1.      Because of my previous work as a land surveyor and the need to use aerial imagery, Autocad, GIS and google earth/google maps I saved some google map searching time.

2.      Because I had made sure that my map projection and truthing shape file were in the NAD_1983_HARN_StatePlane_Mississippi_East_FIPS_2301_Feet projected coordinate system, the point I created were also in this system.

3.      By creating all the points and then using the identify feature set to display in degrees minutes and seconds, I was able to paste this information into the google map search field.

4.      This took me directly to where my point was located and I could use the earth view in conjunction with the 3d tilt/pan and zoom options to quickly identify the classification of my point.

5.      If it matched, I entered yes in the True_YN column during my edit session

6.      If it did not match, I entered no in the True_YN column and entered a more accurate classification in the New_Code column.

7.      After completing my ground trothing, I saved my edits, updated my legend and added an accuracy note with explanation to the final map.

8.      The majority of my initial mistakes from module 3 were in determining the wetland classification.  I knew that they should be comprised of either some sort of marsh grass or trees/mangroves, but it was very difficult to discern in the original imagery.  The google earth view was more helpful, but it was still difficult to discern as much of the area looked like mud flats. 

9.      I did a bit of additional research for that region to determine what the local forestry association described as the flora for these tidal wetlands.  The majority seemed to be varying types of marsh grass.  As that was not one of our specific provided codes I used non-forested wetland.  I think in this case, using a level 3 or 4 code would be more beneficial if actual detailed wetland information were needed.  I feel there may be some areas which are mud flats and oyster beds visible during low tide, but since I did not know at during what tide cycle the imagery was obtained I could not speculate as to the validity of this conclusion.

Module 3 Land Use / Land Cover Classification

Our goal this week was to recognize elements of land use and land cover (LULC) classifications, learn to identify these various types of features on aerial photography and create a LULC map.

Below is my LULC map and a description of the features within the Codes:

Code	Code_Descr	Features
11	Residential	Area containing single and multi family dwellings
12	Commercial and Services	Any retail or fast food
13	Industrial	Areas which appeared to be industrial sites
14	Utilities	Water Tower
15	Industrial and commercial complexes	Areas of both industrial and commercial uses
16	Mixed Urban or Built-up Land	Land which was developed cut could not determine a specific use
24	Other Agricultural Land	Land which appeared to have been planted but currently fallow
43	Mixed Forest Land	Areas with Trees
51	Streams and Canals	Water feature, mainly river and tributaries
52	Lake	Water body
61	Forested Wetland	Wetlands
62	Non-forested Wetland	Wetlands
73	Sandy Areas other than Beaches	Area which appeared to be sand but in an area not likely to be a beach
121	Cemetery	Headstones, road network, and manicured grass
122	Education	Schools
43/33	Mixed Forest Land/Mixed Rangeland	Areas with Trees and grassland
61/62	Forested/Non-forested Wetland	Wetlands
62/61	Non-forested/Forested Wetland	Wetlands

Module 2 - Basics of Aerial Photography and Visual Interpretation of Aerial Photographs

This Modle was designed to provide basic skills in recognizing tone (brightness/darkness) and texture (smooth/rough) in areas and features; identifying features by their shape/size, shadow, pattern and aassociation; finally we compared a True Color Image to a False Color Image.  We used similar techniques for identifying areas (via polygons) and locations (via symbols) and converting them to .shp files so that they could be edited and labels added based upon information input into the attribute table.  Using similar marking techniques we learned to recognize how feature color changes depending on whether you are using a True Color or False Color image.

The first two exercises resulted in the following maps: