Does Greenery Follow...Green?

There have been several studies that compare the income of a neighborhood and the associated NDVI to other neighborhoods within the city.

My question is: Does the percent change of mean NDVI correlate to the percent change of median household income?

I chose to explore Portland in this project because it is a tree city and has one of the largest urban forests (Forest Park) in the United States. Portland's population has nearly doubled since the '80s and since 2010 it has grown by 13% ( WorldPopulationReview ). With changes in population come changes in income. By comparing changes in income and changes in NDVI, I can answer the question: Does Greenery (NDVI) follow Green (income)?

For my study, I focused on two time periods; 2010 - 2015 and 2015 - 2019. Within those time periods, I am specifically looking for changes in mean NDVI and median household income within each Portland census tract.

Normalized Difference Vegetation Index (NDVI) is a value that measures the reflectance of vegetation or, in other words, the "greenness" of the vegetation. Using satellite imagery and mathematical algorithms, values between -1 and 1 are produced. I'll let the USGS explain a bit more "Certain pigments in plant leaves strongly absorb wavelengths of visible (red) light. The leaves themselves strongly reflect wavelengths of near-infrared light, which is invisible to human eyes."

Thank you USGS.

In areas where there is little reflectance of red light and high reflectance of near-infrared, NDVI values can be close to 1 (0.6 to 0.9), representing healthy, dense green vegetation. Moderate NDVI values, ranging from 0.2 to 0.5, are representative of sparse vegetation or senescing vegetation. Values that are close to 0 (0.1 or less) represent areas that are barren, snow, rock, ice, or, in urban terms, pavement, cement, and asphalt.

The change in 2015 is found by comparing the NDVI from August 2010 to August 2015. Similarly, the change in 2019 was found by comparing the NDVI from August 2015 to August 2019.

The Portland census tract data were easily accessed on  Census.gov , and by applying some filters, the American Community Survey five-year data (ACS5yr) I needed was available. I downloaded data for the years 2010, 2015, and 2019.

Although the ACS provided loads of data, I was in search of two specific columns: FIPS and Median Household Income. The FIPS code is the last 11 digits of the "geoID" column and can be extracted using an excel function (=right("cell",11)). It is associated with a particular census tract location and is used later on in geoprocessing when joining the attribute table with a polygon.

Below is a sliding map where you may compare the changes in Median Household Income (left) and the changes in NDVI (right). Each map has a unique legend and symbology due to the fact that when both maps share the same distribution intervals, one of the maps becomes quite misrepresentative. I have also created unique colors for the symbology to not confuse the values from one map to the other.

In order to analyze the data with more than a visual field, I took a brief and confusing detour into the realm of statistics. I wanted to determine if there was a correlation between the two potentially related data sets. A jog of the memory, phone-a-friend, and a google search later I discovered the statistical tool I was looking for  Pearson's Correlation Coefficient . This statistical analysis determines the strength of correlation between two independent cases, such as in this examination. Similar to NDVI, the values range from 1 to -1. A high degree of correlation is if the coefficient value lies between ± 0.50 and ± 1; Moderate degree of correlation values are between ± 0.30 and ± 0.49; Low degree of correlation values are below ± .29; no correlation is 0.

The moment of truth! Is the percent change of median household income correlated with the percent change in NDVI?!

Hold your breath no longer....

There is no correlation. The Pearson's coefficient came to be -0.0025. Dab smack on no correlation. Bummer, I thought that there would be at least a small relationship, but as far as statistics go, there is no relationship to be had.

Keep your chin up! We're only halfway done, perhaps the changes between 2015 and 2019 will be correlated!

Below is a sliding image for you to compare the changes from 2015 to 2019 (Median Income; left | Mean NDVI; right). Once again, the symbology and values for the symbology are unique to each map to reduce confusion when comparing.

Once again I used the Pearson's Correlation Coefficient to discover if a relationship was present between these two independent values.

How'd it score?

The coefficient value was 0.10. Once again, no relationship, or, possibly, a very weak relationship.

Does greenery (NDVI) follow green (income)?

No, not in this case. There is no relationship between changes in NDVI and median household income during both time periods examined within the Portland census tracts. I thought that there would be some correlation between the two, which is why I chose to undertake this project in the first place, but I am also pleased to come to a conclusion. The methodology I used is a different, maybe unique, approach to measuring the relationship between Income and NDVI. I chose to measure the relationship of the percentage of change in NDVI and median household income whereas other studies, such as Saporito et al., 2015, measured the relationship between NDVI and median household income, not the relationship of the change.

An area that has a change in income should not expect to see a change in greenery.

NDVI is a fantastic way to measure the reflectance of greenery, especially in spring when it is vigorously growing and at the peak of greenness. That being said, Portland is wet in the spring, meaning that there is an abundant amount of clouds and does not promote clear satellite imagery. I chose to use imagery from August instead because it is the month when I was able to find a clear image for each year. NDVI values in this case may be altered by precipitation as it is typically not abundant come August in the Pacific Northwest. In fact, two of the three NDVI images had large fires within them.

Saporito et al. found that there was a positive correlation between median household income and NDVI, where income was high, so were NDVI values. This was done by correlating the values of NDVI and median household income, not the percent change between time periods. This was interesting to me and, because I found no correlation in my project, I chose to replicate the method used by  Saporito et al. 2015 . I had the data I needed to do this side-experiment on hand from my above project. I used the Pearson's Correlation Coefficient once again and used the values from Median household income and mean NDVI.

My findings of the replicated examination for each year all provided a strong positive correlation. The correlation values found were: 0.69, 0.67, 0.61 for the years 2010, 2015, and 2019 respectively. My findings of the correlation between NDVI and median household income in Portland align with those found by Saporito et al. 2015.