A Mapping project for the rutgers university bloustein school course command line gis

sean andrew chen

Table of Contents



Urban areas suffer from what is called the urban heat island effect. Because of the materiality of the built environment (i.e., the heat retention properties and albedo of concrete and asphalt) as well as the lack of vegetation in urbanized areas, cities experiences hotter temperatures than rural areas. I wanted to know if urban morphology also had anything to do with urban land surface temperatures. That is, if the shape and configuration of the city, its streets, blocks, and buildings contributed to higher temperatures or not. In this final project for the course Command Line GIS at Rutgers University’s Bloustein School of Planning and Policy under Dr. Will Payne, I attempt to answer that question. Using Brooklyn, NY as a case study, I calculated land surface temperature, a vegetation index, albedo as well as a host of built environment morphological characteristics. On this webpage, I provide static as well as interactive maps of my data. I also employed a simple regression analysis to see what if any relationship there was between those morphological characteristics and land surface temperature.


Raster data for land surface temperature (LST), normalized difference vegetation index (NDVI), and albedo were calculated using the Google Earth Engine. All three were calculated using Landsat 8 data from the USGS from the summer of 2019. That is, the data was averaged across the months of June, July, and August for the year 2019 in the New York City region. Building footprints were retrieved from the New York City Open Data portal. I had originally experimented with the Microsoft National Building Footprint database, but these did not have height data while the NYC Open Data portal data did. The Brooklyn street network was retrieved from Open Street Maps via the OSMNx Python package. Finally, family poverty rate and percent of population that is non-White was retrieved via API from the US Census’ 2019 5 Year American Community Survey data at the Census tract level. Tract geometries were retrieved from the Census’ TIGER files online. Since the TIGER files have tracts by state, I clipped the tracts using a the borough of Brooklyn. For the most part, data quality was high with the exception of a few missing pixels in the land surface temperature data.


Using Google Earth Engine, raster values for land surface temperature, NDVI, and albedo were calculated. Then, using RasterIO in Python, the raster values were sampled for building footprints and Census tracts with the median value being taken. That is, each building footprint and each tract was now given the median value of the raster pixels within that object polygon. This way, we could connect land surface temperature, albedo, and NDVI to building morphological characteristics. We could also compare tract level demographic variables to those raster values. The next step was calculating those morphological characteristics. Using the PySAL subpackage of Momepy on building footprint and street network data, we created tessellations and blocks. Now using building footprints, tessellations of those building footprints, the street network, and blocks, I calculated morphological metrics. To map these metrics, I spatially joined the buildings and their metrics to Census tracts and grouped the measurements by Census tract taking the mean value. Connecting those morphological metrics to the raster values, I was able to run a simple OLS regression with LST as the dependent variable and the morphological metrics as well as NDVI and albedo as the independent variables. The OLS was run with a spatial weights matrix to perform spatial diagnostics. However, a spatial regression, even though advised by the diagnostics, was not undertaken at this time.


Initial findings suggest that there is a relationship between a few morphological metrics and land surface temperature. The controls of NDVI and albedo did indeed show very strong relationships with land surface temperature. But for morphological characteristics, building adjacency, which is defined as how much buildings tend to join together into larger
structures calculated as a ratio of joined built-up structures and buildings within a certain distance, had a a relatively strong effect. The shared walls ratio, which is length of shared walls divided by the perimeter of the building, had a slightly weaker effect. But it was building volume facade ratio, which is the volume of a building divided by the product of the perimeter and height, which had the largest effect of all the building morphological characteristics. Street openness also had a relatively strong effect. However, the strongest morphological characteristic was network closeness, or how close together the nodes of the street network are. 


These choropleths show two variables at the Census tract level: percent of families living below the poverty line and percent of population that identifies as non-White. Data was taken from the 2019 5 Year American Community Survey. Tracts with a coefficient of variation greater than 40 were crosshatched in the choropleth. This is to show what tracts have highly unreliable data.


Data for land surface temperature (LST) in Celsius, albedo, and normalized difference vegetation index (NDVI) were all calculated from Landsat 8 imagery via Google Earth Engine. The rasters were then processed in Python using RasterIO and sampled for each Census tract polygon. The median value of all the pixels in each polygon object was taken. These maps show the distribution of those values aggregated at the Census tract level.



The above small multiples aim to help viewers see if there is a relationship between land surface temperature and family poverty rate or percent of the population that identifies as non-White. There does seem to be a bit when inspected visually. Land surface temperatures are higher in the north and eastern parts of Brooklyn, roughly the same as where non-White and poor populations reside. However, I also ran a simple Pearson’s correlation test on the values. There is not really a relationship between percent non-White and land surface temperature. There is, however, a positive relationship between the poverty rate and the land surface temperature. As predicted, there is also very strong relationships between NDVI and LST. There is also a relationship between albedo and percent non-White. And of course, there is a strong relationship between poverty rate and percent non-White.


This interactive web map helps you visualize the relationship between demographic variables and variables like LST, NDVI, and albedo.


In order to go ahead with morphological analysis of the built environment in Brooklyn, NY, you need some foundational “building block” data. The most fundamental is building footprints and the street network. From these, you are able to tessellate shapes around each building footprint bounded by the street network. From these tessellations and the street network, you are able to create blocks. These “building blocks” of building footprints, street network, tessellations, and blocks are what we calculate morphological characteristics off of.

Building footprints and Street Network in Williamsburg, Brooklyn

Tessellations around Building Footprints and Street Network in Williamsburg, Brooklyn

Blocks created by Tessellations and Street Network in Williamsburg, Brooklyn


Morphological metrics were calculated off of blocks, streets, tessellations, and buildings. There were up to 75 metrics, and spatially lagged metrics were also calculated. However, for simplicity’s sake, we show only 10 building metric variables in this interactive web map. The data itself was created by spatially joining the building polygons to Census tracts and then aggregating the metrics by the mean for each Census tract. Thus, this map shows the average value of each metric by Census tract. These metrics show larger patterns afoot in Brooklyn. There seems to be a split between the morphology of north Brooklyn and south Brooklyn.


First let us look at the correlations between land surface temperature and other variables:

Here are the correlation coefficients. All but Tessellation Circular Compactness have a p-value of less than 0.05.



Variable Pearson Correlation Coefficient
NDVI -0.241728
Building Adjacency 0.164737
Albedo -0.163848
Building Perimeter Wall -0.146870
Street Network Closeness 0.144760
Building Volume Facade Ratio 0.144430
Street Openness 0.143223
Building Squareness -0.116923
Tessellation Circular Compactness 0.114319
Building Shared Walls Ratio -0.103934



Variable Regression Coefficient
NDVI -5.5656404
Building Adjacency 0.4319604
Albedo -9.1601112
Building Perimeter Wall -0.0003460
Street Network Closeness 51832.6334535
Building Volume Facade Ratio 0.8523624
Street Openness 0.3353771
Building Squareness -0.0068810
Tessellation Circular Compactness 0.0272492
Building Shared Walls Ratio 0.2961113

The adjusted R2 of the regression is 0.1895.