Methods

Analysis began by converting the previously processed LANDSAT images of Chicago’s albedo and temperature change that Mackey et al. had used in their study into an ArcGIS-compatible GeoTIFF format. Both the GeoTIFF and the building shape file were then opened in the GIS and the two data sets were geo referenced in relation to one another. Since the vector shape file was already too large for most types of analysis and it was difficult to get a meaningful higher resolution out of the 30m LANDSAT imagery, it was determined that a raster-based analysis at 30m would be optimal.

Accordingly, the building polygon was converted to 30m raster format using ArcGIS’s “Polygon to Raster” tool. For this operation, the field used to assign values to raster cells was a “mask” field that had been generated such that all polygons had a value equal to 1. Running this tool with the “Maximum Area” setting and using the LANDSAT image as a snap raster ensured that the resulting mask included all LANDSAT pixels for which there was some building footprint within them. While this strategy admittedly results in a lot of mixed pixels which sometimes barely contain any area of building foot print, it was felt that it would be better to include these mixed pixels as buildings since a new reflective roof on even the smallest portion of a building in a pixel would produce a noticeable increase in the albedo of that pixel.

The resulting raster of buildings was then used to mask out the non-building pixels of the albedo and temperature change images using the “Extract by Mask” tool. Attribute tables were generated for both the original city-wide albedo and temperature change images as well as the new building-masked versions using the “raster to point” tool. While this tool may not be the most computationally efficient means of generating attribute tables for a raster images, it was deemed favorable for its ability to preserve the decimal values of albedo and temperature in the raster cells.

Next, the tables of albedo and temperature change were joined based on the generated “pointid” field, which was the same for each pixel in the raster pair so long as the initial raster images had identical numbers and locations of active pixels. This resulting table was then exported to a text format and brought into Excel. Here, the albedo change and temperature change pixels were plotted against one another in a scatter plot and values of correlation were derived by taking the square root of the R2 value generated from the best fit line. For better graphical representation, the data was also binned into intervals and plotted with error bars of 1 standard deviation. From these graphs, a slope of regression curve was taken.

This Excel file was also used to obtain the number of albedo increasing pixels in each bin that lied within building footprints. Ultimately, this allowed the production of a graph showing the percentage of albedo increases attributable to reflective roofs based on intensity of albedo increase (figure 5 in the next section).

After studying the albedo increases in building foot prints in relation to those of the whole city, the building albedo increases were also evaluated in relation to the building criteria of height and number of dwelling units. To obtain raster images of these two criteria, the same “polygon to raster” process was used except, this time, the value of the given criterion was chosen as the cell value instead of a uniform “mask” field. From here, the same process of joining tables and exporting them into Excel was deployed to get correlations, slopes of regressions and graphs. It should be noted that the number of building units was evaluated only for residential buildings with a non-zero number of units. This was achieved by sorting the data in excel and deleting all values with zero dwelling units.

Continue to the Next Section: Results Part I: Verification that Albedo Increases Resulted from New Reflective Roofs

Contents:

Methods