In an urban environment, the quality of landscaping is a crucial component of the living infrastructure. It plays an important role not only from an aesthetic perspective but also for health reasons. Spaces such as parks, as opposed to just street trees, make significant contributions to green spaces, which can influence surrounding property values. Therefore, landscaping, both as a valuable asset and as a determinant in the allocation of available space, inevitably impacts rent prices. In this series, we will analyze the relationship between the greenery index—examined both from a macro scale using satellite images and from a street-level perspective—and rent prices through data analysis.
In this article, we utilize the traditional method of estimating the green view index using GIS. Specifically, I applied the traditional approach to approximate the green view index by using the Normalized Difference Vegetation Index (NDVI) derived from Sentinel-2 satellite images. The satellite's vision sensor can differentiate captured images into various color bands through image processing. For example, black represents areas with no relevant objects (such as water bodies or regions without vegetation), while darker shades of green indicate areas with more abundant vegetation.
Figure 1. NDVI around NYC
When zooming in on Central Park and Times Square, it looks like this: The residential areas around Central Park West and Central Park East appear to have well-planned urban landscaping with neatly arranged street trees.
Figure 2. Zoomed in version of the NDVI around central park.
And then, I joined the data by Census tract units.
Figure 3. aggregated NDVI per census tracts boundary
However, when examining the distribution of vegetation by region, it becomes evident that in urban settings, vegetation is often limited to the remaining spaces after the installation of infrastructure like buildings and roads. As a global trend, areas with less vegetation tend to have higher rent prices. In other words, less vegetation is often associated with urban areas. For example, comparing dense areas in Manhattan with rural areas in Queens, Manhattan has higher rent prices but less vegetation. This suggests a hierarchy based on region.
The relationship between rent price per square foot and NDVI is shown to be negatively correlated, as illustrated in Figure X. In the figure, the y-axis represents rent price per square foot, and the x-axis represents mean NDVI. The two variables have a slope of -1.48 and a p-value of 0.00*, indicating a statistically significant relationship
Figure 4. Global trend of relationship between NDVI vs. rent price per sqft in NYC.
To reflect regional differences in a data-driven manner, I classified the areas as follows. Using the statistics of building height and building footprint area, I applied PCA followed by K-means clustering for classification. This feature is also included in the MarketStadium app. Areas marked in purple indicate higher average building heights and are identified as dense urban areas.
Figure 5. PCA-K-means clustering result: cluster 1 represent more dense urban area.
However, once again, the results showed a negative relationship between NDVI and rent price per square foot, consistent with the global trend.
Figure 6. cluster 1 is denser urban area. in both Cluster 1 and 2, NDVI and rent price per sqft are negatively correlated.
Therefore, instead of a data-driven approach, I conducted the same analysis by dividing the data based on boroughs: Manhattan, Brooklyn, Queens, and the Bronx. The results showed a significant negative correlation in Manhattan and Queens, while the correlation in Brooklyn and the Bronx was not statistically significant.
Figure 7. Results of NDVI vs. rent price per sqft
Discussion of Results. The analysis did not yield sufficient insights. The initial hypothesis was that more greenery would positively affect rent prices, but the results showed the opposite. However, this outcome does not necessarily reflect the true relationship between greenery and rent prices. The limitations of this analysis likely stem from the fact that many buildings are included within a single census tract boundary, and the rent per square foot data is either dissolved or aggregated, leading to significant missing information. Additionally, the exclusion of other relevant variables may have contributed to these unexpected results.
As a result, more granular analysis at the street or point level is needed. In the next analysis, I will explore the relationship between green view index and rent price on a micro-scale using MarketStadium's point-level rent price data combined with surrounding Google Street View images. I will also investigate whether there is a discrepancy between NDVI derived from satellite imagery and greenery observed in street view images. Although this approach has been discussed in various urban planning journal papers, the goal here is to explore its application using MarketStadium data and to demonstrate how it can be effectively utilized.
Figure 8. multiple buildings and vegetation data in a single census tract.
Reference
Google Earth Engine. (n.d.). Sentinel datasets catalog. Retrieved September 2, 2024, from https://developers.google.com/earth-engine/datasets/catalog/sentinel
Comments