Urban Expansion Analysis

Abstract:

Orlando’s development patterns in the last thirteen years show a departure from past trends that reflect Orlando’s emergence as a business, technology, and AI hub. Past analyses examine Orlando’s development either through a public space lens or fail to cover Orlando’s recent large-scale development trends. While new ways to analyze urban growth have been developed, no existing research has attempted to apply these methods to Orlando. I used IBI and VIIRS data drawn from satellite imagery and Getis-Ord Gi* Hot Spot Analysis to visualize and quantify Greater Orlando’s urban development from 2013 to 2024. By examining the imagery across different year ranges, including from the entire range, I compared the average VIIRS value of hotspot areas to those of non-hotspot areas to determine whether urban development was spatially associated with higher activity areas. My findings suggest that urban growth hotspots were spatially associated with higher pre-existing activity, suggesting that there was a greater focus on intensification of existing urban cores rather than outward expansion. Awareness of these trends can not only help developers and policymakers of Orlando and similar cities anticipate the future needs of newly built-up areas to facilitate further growth. Streamlining the methodology for built-up expansion also allows for analysts to catch built-up encampments in unnaturally occurring areas. 

Data Collection:

Region of Interest (ROI): Lake, Orange, Osceola, Polk, and Seminole counties

• Shapefiles found via Florida Department of Transportation's Open Data Hub

Data used:

• Landsat 8/9 imagery (Path 16, Rows 40-41)

  • From USGS EarthExplorer

  • Taken from the earliest point in the year when cloud cover was filtered for under 10%

  • Imagery used in analysis from

    • March 28, 2013

    • February 16, 2016

    • January 7, 2019

    • January 31, 2022

    • January 21, 2024

• VIIRS Nighttime Lights

  • From the Colorado School of Mines’ Earth Observation Group

  • Monthly composites have cloud-free coverage from 2012-2024

  • Downloaded their 2013, 2016, 2019, 2022, and 2024 composites

Data Cleaning:

ROI

1. Imported county borders into ArcGIS Pro

2. Used the "Project" tool to set the CRS to NAD 1983 UTM Zone 17N, the same projection as the Landsat data

3. Used the "Merge" tool to make the counties into one entity, the ROI

Landsat

1. Imported all downloaded Landsat imagery into ArcGIS Pro

2. Used the "Composite Bands" tool to tie together the RGB, SWIR, and NIR bands into one raster file, repeating for both tiles in a given year

3. Used the "Mosaic" tool to stitch together the two tiles into one that covered the entirety of the ROI

4. Used the "Extract by Mask" tool to clip the mosaicked raster to the county boundaries

5. Repeated steps 2-5 for each year, leaving me with 5 total raster files in the shape of the ROI with all necessary bands included

VIIRS

1. Imported the VIIRS imagery into ArcGIS Pro. I went through the whole process for one year before starting the next to preserve memory and prevent crashes.

2. Used the "Extract by Mask" tool to clip the imagery down to the size of the ROI

        2.1. During this step, I projected the VIIRS imagery to NAD 1983 UTM Zone 17N to match the Landsat data

3. Used the "Resample" tool to apply Nearest Neighbor Interpolation to turn the resolution of the VIIRS imagery from approx. 375 meters to 30 meters

NOTE: Nearest Neighbor Interpolation was used because of its ability to preserve the original values of the raster. Cubic Convolution was considered, but that resampling method changed the original radiometric values of the raster, which provides inconsistent results. For example, Cubic Convolution can make the image seem “brighter” because of higher mean values, which makes the raster near impossible to analyze. Therefore, Nearest Neighbor Interpolation was the obvious choice for resampling. 

The primary limitation of using VIIRS data with Landsat is the resampling, which, by nature, means that some values will be untrue to reality in some cells. Resampling is a necessary evil for using VIIRS alongside Landsat for analysis, but as earlier stated, Nearest Neighbor provides the most preservation of cell values that should not skew results one way or the other.

Methodology:

1. Used the "Band Arithmetic" tool to compute the Indices-Based Built-Up Index (IBI) for each year

2. Used the "Raster Calculator" tool to find the change in IBI in a later year from the earlier year

2.1. This left me with 5 total Change in IBI raster files, one for...

• 2013-2016

• 2016-2019

• 2019-2022

• 2022-2024

• 2013-2024 

3. Used the "Raster to Point" tool to convert all Change in IBI rasters into points, since Getis-Ord Hotspot Analysis requires the use of point layers

4. Used the "Hot Spot Analysis (Getis-Ord Gi*)" tool, using the point layers as the input, the Change in IBI value as the field used for finding hotspots, and Euclidean distance as the Distance Conceptualization method, to carry out my analysis

4.1. The output map gave me a hotspot map where red marked statistically significant high change in IBI (new urban clusters) and blue marked statistically low change in IBI

5. Changed the transparency of each hotspot layer to 50% and overlaid it onto the VIIRS imagery from that year for visualization. Maps 1, 2, 3, 4, and 5 depict the hotspots of urban growth and where activity was prior to that growth

6. Used the "Select" tool to split each hotspot layer into two: one with a Gi_Bin value of greater than or equal to one (singles out hotspots), and one with a Gi_Bin value of 0 (symbolizes non-hotspot areas). This creates ten total layers

7. Used the "Extract Values to Points" tool to assign the overlapping VIIRS value to every point within the ten layers that "Select" created

8. Used the "Summary Statistics" tool to find the mean VIIRS value for each hotspot and non-hotspot area layer. 

9. Compared the mean value for the hotspot areas only and the non-hotspot areas only layers to determine whether urban growth hotspots were spatially associated with higher pre-existing activity. The table with my findings can be found below.

Results:

The biggest conclusion that can be drawn from the maps and Table 1 is that by-and-large, growth hotspots in Greater Orlando were spatially associated with higher pre-existing activity. The only year range where growth hotspots were spatially associated with lower pre-existing activity was from 2016-2019. 

The maps verify the numbers. In Map 2, we see the lone year range where the average VIIRS value of non-hotspot areas were higher than the hotspot areas. This means that new urban development during 2016-2019 occurred disproportionately in lower-activity areas, in the periphery of existing activity. This indicates more sprawl in the traditional sense that mimics Greater Orlando’s sprawl patterns from the 1970s to 2000s, where the focus was on outward expansion rather than the intensification of established urban cores compared to the other studied year ranges. While some development occurred near existing activity centers like Lakeland, southern and northern Orlando, the dominant pattern in urban development was expansion along the periphery. 

As we see in Maps 1, 3, 4, and 5, new urban development is clustering in areas that already have high human activity, which indicates more infilling and redevelopment. Map 5, which compares the cumulative development to activity, shows that overall, Apopka, Winter Garden, Kissimmee, Poinciana, Sanford, Oviedo, and Auburndale experienced the biggest growth since 2013. More importantly though, Map 5 confirms that most of the urban development in Greater Orlando took place in high activity areas from 2013 to 2024. 

Maps:

Implications:

For Orlando:

By observing development trends over the last decade, Orlando planners and developers can use this data to predict where new development will be successful. Since new development has centered around high-activity areas, public transit infrastructure built along high-activity corridors may perform better than if built more generally around Orlando. With the Orange Code and Vision 2050 guiding thinking over how Orlando can be built smarter instead of mindless sprawl, Orlando's future remains bright.

For GEOINT:

This workflow is another means for change detection to be carried out to track adversarial movement and development. Because of the use of VIIRS (and its daily temporal resolution), abnormal activity can be tracked at a speed of the temporal resolution of true-color imagery with SWIR and NIR bands.