Monday, December 4, 2017

Open Source Analyze Week 2 - Food Deserts Duval Co., FL

This week continues with the analyze portion of the overall open source project for module 4.  Here data for a selected city is used in the processes outlined last week, particularly using MapBox and Leaflet.  I chose Jacksonville, Florida as my city which lies within Duval County.  A link to my web map is here: (although I have not been successful in getting it to work yet.){z}/{x}/{y}?access_token=pk.eyJ1IjoibW0yNDYiLCJhIjoiY2pha2N2M3IzMmhiazMzcXU4dXpycm9kdSJ9.89dpeOKCBq4iDjk0VnXDPg

Since the link is not working, I am also including a screenshot of the MapBox version of my map of Duval County/Jacksonville food deserts.

The tiling process was the easiest part of this lab and made the most sense.  It is easy to add layers and edit them however is desired.  In the realm of mapping this is quite useful in that changes can be very specific and are easily reversed or further edited.  It is also helpful in that working with different extents becomes easier as the programs adjust for the user and editing is very streamline.  The following are food for thought responses presented during this lab assignment.  My data is for Duval County Florida which is home to my chosen city of Jacksonville.  I used the USGS and census bureau websites to obtain data for Duval County including the parcels and outline.  I used the US Dept. of Agriculture website to obtain the food desert shapefile.  This data was created by using a selection by attribute tool and specifying areas that classify as food deserts.  To be classified as a food desert, the USDA states “at least 500 people and/or at least 33% of the census tract’s population must reside more than one mile from a supermarket or large grocery store within an urban area and more than 10 miles for rural settings.”  Using these parameters and the selection by attribute tool, areas of “food desert” were selected and sent to a shapefile that could be used across different applications.  Given my source is the USDA I believe the quality/credibility of my data is good.  The data shows that there are several food deserts within Duval county which did surprise me given that Jacksonville, FL lies at the heart of the county and is a sprawling urban area.  I would have expected better access to grocery stores in this entire area being urban and progressive city.  The trend is, not surprisingly, that the food deserts occur in the areas surrounding Jacksonville, particularly to the Northwest.  Here, I was surprised by how close the food desert is to downtown Jacksonville.  The data shows the location of grocery stores and the food desert designation shows the time investment required to obtain fresh, healthy food.

I had far too many technical difficulties to say I enjoyed the lab this week but I did benefit from the tiling section and further use of online mapping applications.  

Wednesday, November 29, 2017

Open Source Analysis - Food Deserts

The first analyze week for Project 4 introduces Mapbox and Leaflet as tools used in open source analysis.  Mapbox was relatively easy to use but somewhat difficult to get used to with no "save" option.  The files imported into Mapbox had to be compressed into a zip folder in order to be used.  Leaflet is an open source JavaScript library for use in web and mobile platforms.  From Leaflet the source code was accessed and copied into Notepad for editing.  The code (html) was intuitive and, in my opinion, easier to understand than Python; however, I have experienced several problems getting the code correct.  Despite proofing and double checking I have not yet located where the problem lies.  My web map is not opening so I will continue to go over the code for any incorrect spacing, syntax, etc.  The code was edited to point to the "I" drive and to incorporate my web map coordinates, titles, and settings such as color.  These additions and edits of the code result in a marker being placed in Pensacola and a popup showing a food desert and non-food desert.  In order to add a legend, a large amount of code needed to be copied and pasted into the notepad text document (which is also saved continually as an .html file).  The indicated sections were changed to reflect my map and the legend is formed in layers much the way layers are added and edited in Mapbox.  Finally, to add the Geocoder to the finished product required the addition of some placement text and required a search for a plugin that worked with the application.  Despite some system and coding issues, the lab was informative and helpful.  For what it is worth, I will continue to work on this code in hopes that I can make it work. 

Wednesday, November 22, 2017

GIS Day Event

Since I was navigating U.S. air travel system on GIS day I decided that I would contemplate the benefits of GIS to travel and specifically air travel.  Traveling with family, I decided to have this discussion with my father and sister.  My GIS "event" took place in Charlotte International Airport and Orlando International as well as points in between up in the air.  This was an informal discussion on the implications of GIS on travel.  This is also of particular interest to me as a private pilot.  There's many features of aeronautical charts which rely on GIS, such as the distances surrounding airports and defining airspace which is critical to maintaining proper separation between aircraft. 

The following are the major points which were discussed.  First, the core issue that the entire process of air travel wouldn't happen without coordinated GIS results that allow for many planes to be in the air simultaneously in a relatively small airspace.  The monitoring and directing of air traffic couldn't happen without directional awareness and precision maps including approach charts and terminal maps.  GIS allows for the most efficient and safest routes to be undertaken.  We discussed that this is particularly important for transatlantic and transpacific flights which follow the great circles to reach destinations.  We also discussed the use of GIS on the roads and recounted several instances when electronic GPS units have failed and knowledge of maps and directions was essential to reaching a destination.  In this discussion it was noted that the building of new roads relies heavily on GPS in engineering roads around and through the mountains we are surrounded by.  

Wednesday, November 15, 2017

Food Desert Analysis - Open Source

Project 4 in the Special Topics course will analyze Food Deserts using open source software and tools such as QGIS which was introduced this week.   Food Deserts are areas, particularly rural areas, that do not have regular, easy access to fresh produce.  This is the basis of an array of concerns but the project will focus on the spatial relationships and implications.  My completed map of Escambia County, FL with food deserts and non-deserts shown. 

This map was completed in QGIS which I found to be much easier, generally, to navigate and use than ArcMap.  I do believe there are some deficiencies in my map product, namely that the desert and non-desert appear to be reversed.  The food deserts should be the areas furthest from the grocery stores and the food oasis closest.  I believe the problem lies in the selection by expression function performed.  I will revisit this to revise results.  The image here is slightly different than the one viewed in the print composer in that the legend is overlapping the study area and the two images of Escambia Co. are not lined up as they were in QGIS.  The program is taking some getting used to but I did enjoy the use of it.  Most of the tool used, such as statistics and other vector tools were easier to use, the design was also simpler to figure out after trial and error with locking layers to keep items from changing.  After joining an excel table and "cleaning up" the attribute table by removing unnecessary columns the selections were made using expressions: Dist <=1 representing locations within a mile of the grocery store and Dist >1 representing those locations further than a mile from a grocery store. 

I enjoyed the lab, despite some trouble with parts.  I look forward to using open source and QGIS for the remainder of this project. 

Tuesday, November 14, 2017

Supervised Classification

Module 10 has been my favorite lab thus far in the Remote Sensing and Photo Interpretation Course.  The lab goes through the various steps and methods associated with supervised image classification in ERDAS Imagine.  My completed map is seen here:

This map is of Germantown, MD and has been re-coded into eight feature classes using tools in ERDAS Imagine and then opening the image in ArcMap for further symbology and color classification.  First, spectral signatures were created by examining pixels and using coordinates in conjunction with the inquire cursor.  From there the "create signature" option in the signature editor was used to create classes for each of the features in the image.  These features were then re-coded and merged into coherent classifications that make sense to the end user. The next step was to check for spectral confusion which indicates that pixels are overlapping and/or mis-classified.  Histograms and mean plots were examined from the signature editor.  The bell shaped curves in the histograms need to be separated in order to avoid spectral confusion.  Similarly the mean plot lines should be separated and not on top of one another.  The image was then opened in ArcMap for further editing, particularly the colors used in the main image.  ArcMap also allowed me to insert the inset for the Output Distance File.  The image uses symbology that makes sense to the user and all the features are distinct.

This lab was quite helpful especially in examining the pixels and creating the spectral signatures.  It drives home the precision that must accompany classifying images.

Tuesday, November 7, 2017

Unsupervised Classification

Module 9 involved taking an image in ArcMap and using the Spatial Analyst tools of Iso Cluster and Maximum Likelihood Classification.  This transformed the image and from here each pixel could be categorized so that the number of classes was drastically reduced.  Similarly in ERDAS Imagine, though more involved, another image was reclassified and later regrouped.  My final regrouped map is seen below:

The process was far easier in ArcMap but, as is common, the ability to specify greater detail was found in ERDAS Imagine.  Here there were 50 classes and by zooming in on individual pixels (taken at a scale of 1:500 or less so that features were slightly identifiable), each was classified into one of 5 classes: trees, grass, buildings, shadows, and mixed.  The mixed field aided in classification of areas that were more indistinguishable.  From here the percentage of areas falling on permeable and impermeable areas was determined to be 69% and 29% respectively.  The classes were given more natural colors which are intuitive of what the features are in "real life." 

I enjoyed this lab and it gives a good idea of how a map can be drastically different simply by the way it is visualized.  This draws attention to the fact that this type of work must be precise and accurate. 

Friday, November 3, 2017

Methamphetamine Lab Busts - Analysis

In order to represent the Methamphetamine Lab distribution in the West Virginia region, the various fields from the spatial join performed previously must be examined in order to display the most effective which will demonstrate the most accurate situation.  I chose to analyze the fields of age groups 18-21 and 21 and up, several ethnic backgrounds and household dynamics of whether or not the home is occupied by a single parent.  These appear to be the most logical demographics to analyze when it comes to issues relating to public health and socioeconomic status.  The Ordinary Least Squares Regression was run twenty-one times and various fields were removed individually in order to examine the effects on the summary table, particularly the probability, coefficient and VIF elements.  I eliminated many fields that had coefficients greater than 0.05 and probability which was large.  VIF values that were obviously higher and “out of place” were also removed and the effects reviewed.  The coefficients and probability are quite closely related and generally as one field was removed those values behaved similarly.  When exceptionally large VIF values were removed, the remaining tended to become more uniform and close in nature.  The end result has all the values similar and related to one another.  The Six-Checks were a useful tool that helped to determine which fields would stay and which would be removed.  By answering various questions as to the behavior of the data it was easier to determine whether to remove the field or let it remain.  Another factor was to consider the “common sense” scenario.  For example, the “age under 5” category is intuitively not producing Methamphetamine and therefore should be removed from the regression analysis.  The resulting summary table and OLS standard residual map are seen below:

A small probability indicates that the residuals are not normally distributed, as in a bell shaped curve, and therefore the model displays a bias.  A lesser probability than 0.05 indicates statistical significance and is related to the coefficient in that when the probability is small the coefficient, or independent variable, is near zero.  A coefficient near zero indicates a small probability. The coefficient is akin to the slope of a line, indicating the relationship between the independent/explanatory variable to the dependent variable and therefore when it is near zero, there is a low relationship.  

 I predicted that the least significant variables will be “households with a single male parent” and “households with a single female parent.”  My reasoning behind this hypothesis is that I believe it is not the fact that the single parent is male or female but rather the fact that they are a single parent under excessive stress and socioeconomic depression.  I believe the more significant factors are income levels, lever of education and employment status.  The results of this analysis show a concentration of the Methamphetamine laboratory busts concentrated in the central Charleston, West Virginia area and to the west and east of the center of the city.  There is a trend, as seen in the Ordinary Least Squares Regression, of a younger population of diverse backgrounds and lower socioeconomic status that sees the highest occurrence of meth lab busts.  The fact that these are of a high concentration in the city is also in keeping with the demographics involved with this crisis.  The circular problem creates the circumstances and in turn the circumstances and consequences keep these groups in their current situation.  The coefficients, probability and VIF factors of the age groups of 18-21 and 21-29 are most reasonable and indicate the highest likelihood of a laboratory bust.