Remote Sensing Final Project: Change in NDVI Over Time

Change in NDVI by Neighborhood in Richmond, VA from 2018-2021

Introduction

For my final project, I was interested in analyzing vegetation change in urban areas in Richmond, VA. I was inspired by the article “Urban environmental stewardship and changes in vegetative cover and building footprint in New York City neighborhoods (2000–2010),” in which Locke et al. (2014) assess whether there is a correlation between the number of environmental stewardship groups in New York City neighborhoods and whether that neighborhood gained or lost vegetation between 2000 and 2010. This study used Spectral Mixture Analysis on Landsat 5 imagery from 2000 and 2010 to determine the percent vegetation change for each neighborhood. For my project, I decided to measure vegetation change for each neighborhood in Richmond using the simpler method of NDVI.

Remote Sensing Imagery

I downloaded imagery from the National Agriculture Imagery Program (NAIP), provided by the USDA-FSA Aerial Photography Field Office. This program had imagery for Virginia approximately every three years, the most recent being 2021. I looked at older imagery from several different years, but encountered various issues, such as a lack of imagery with the infrared band.. I wanted to use the 2011 imagery in order to compare from a 10 year range, but it was captured several months earlier in the year than the 2021 imagery. It also looked very different from the 2021 imagery, with the colors (especially the blues and greens when displayed in false-color infrared–that is, the green and red bands) being much more muted. Additionally, the NDVI values I calculated from the 2011 imagery did not extend all the way from -1 to 1. Although there may have been user error involved, this data seemed sufficiently different from the 2021 imagery that it did not seem worthwhile to compare them. Instead I chose to compare imagery from 2021 and 2018.

Both the 2018 and 2021 imagery are orthorectified aerial images captured with a Leica ADS100 digital pushbroom sensor with a spatial resolution of 60 cm. Both captured 4 bands, but the data I downloaded provided 3 bands in false-color infrared. The 2018 imagery for the Richmond area was captured between 9:36 am and 9:50 am on 8/25/2018, between 9:58 am and 12:29 pm on 8/26/2018, and between 11:07 am and 11:15 am on 8/28/2018. The 2021 imagery was captured between 10:58 am and 12:12 pm on 9/10/2021.

Image Processing and Analysis

I downloaded shapefiles of Richmond’s city boundary and neighborhood boundaries from the Richmond GeoHub website and clipped the 2018 and 2021 rasters to the city boundary. I used the NDVI tool to calculate the NDVI for each raster with Band 2 as Visible Red and Band 1 as Near-Infrared. I then calculated the median NDVI for each neighborhood in each year with Zonal Statistics. I chose to summarize by the median instead of the mean to reduce the influence of possible outlier values. I subtracted the 2018 NDVI raster from the 2021 NDVI raster to get the difference in NDVI for each pixel. I couldn’t get Zonal Statistics to work with the raw NDVI difference values, so I multiplied the raster by 100 and converted the values to integers, then ran Zonal Statistics to get the median difference in NDVI for each neighborhood.

Results

The median change in NDVI for the whole city is 0.0035, measured on the typical scale of -1 to 1. Although this is a small change, I was pleasantly surprised that this increased NDVI suggests that Richmond had either more or healthier vegetation in 2021 than in 2018. The neighborhood with the greatest decrease in vegetation was Bryan Park with a difference value of -5. (Since I had to multiply by 100 and convert to integers, this corresponds to about a 0.05 difference in NDVI.) This neighborhood, which has a very high NDVI in both years, is mostly comprised of a large park. Upon inspection of the false color infrared imagery, I noticed that the park’s fields in the 2021 imagery appear to be reflecting less infrared energy than in 2018. This could be explained by the grass being drier and less verdant due to the later date or a lack of water in 2021. The neighborhoods with the next greatest decrease in vegetation are Forest Hill Park, Broad Rock Sports Complex, Cullenwood, and Cherry Gardens, all with a difference value of -4. Forest Hill Park is a forested park with no visible decrease in tree cover, so the difference could be due to decreased vegetation health. Broad Rock Sports Complex includes forest, sports fields, and houses, so vegetation health likely influenced its decreased NDVI. There are also a few new buildings in this neighborhood in 2021. Cullenwood and Cherry Gardens both include residential areas and forest, but there are many other neighborhoods that appear to be similar in land cover makeup, so it is unclear what caused the larger decrease in NDVI in these particular areas. 

The neighborhood with the greatest increase in vegetation was Belle and Mayo Islands, with a difference value of 9. This neighborhood comprises part of the James River and some forested islands, so its median NDVI values are very low due to the large amount of water in the image. The NDVI difference could be due to changes in the river instead of a significant increase in vegetation health on the islands. There is no visible change in land cover on the islands. Shockoe Bottom, Shockoe Slip, Central Office, and Biotech and MCV District have the second highest increases in vegetation with a difference value of 5. All of these neighborhoods are in downtown Richmond and have low NDVIs in both 2018 and 2021. In Shockoe Bottom, there is some land cover change, which looks to be new construction of buildings on empty lots. I’m not sure how NDVI differs between parking lots and buildings, so I’m not sure if the construction affected the NDVI. I noticed that building shadows and the light reflected from rooftops often varied somewhat in these downtown neighborhoods between 2018 and 2021. I wondered if the time of day might have affected the NDVI, but all of the imagery was captured in the mid- to late-morning, so it seems like the time of day was controlled for as much as possible.

In the maps below, I chose to display the median NDVI by neighborhood for 2018 and 2021. These maps are very similar, but there are a few neighborhoods with observable differences. I displayed the median change in NDVI by neighborhood to highlight the neighborhoods of greatest change. I also included the raster of NDVI values from 2021, because I thought it would be helpful to readers unfamiliar with Richmond to contextualize where the river, major roads, and most urbanized areas are.

Limitations and Problems

One of my biggest problems was finding good data to compare two different years. I could only find natural color imagery for many years, and some of the false-color infrared imagery I found was taken at different times of year, making it difficult to compare vegetation changes. The imagery from 2018 and 2021 was at most 16 days apart (since some of the imagery from one year was captured on different days), which I think is sufficient to compare the two. I would’ve liked to have a larger time span than three years, since that would’ve allowed me to observe larger and more significant changes. It also would have been interesting to compare more than two years, to see if there is a consistent trend in NDVI changes.

Conclusions

For most of the neighborhoods with the greatest decreases and increases in NDVI, there does not seem to be significant changes in land cover, so there is no evidence that the changes are due to more or less vegetation. I am unsure of what caused the differences between the two years. One theory is that the climate differed between the two years. For example, a drought would decrease NDVI. Land cover change or initiatives to increase urban vegetation are likely more of a factor over a longer timespan. According to an article by Hammond (2022), Richmond set a five-year goal in 2017 of planting 24,000 trees on public land. They exceeded this goal by planting 35,870 trees by December 2022, which was calculated to add 120.3 acres of tree canopy to the city. It is difficult to know whether this initiative may have contributed to the small increase in NDVI from 2018 to 2021, especially given the short time period of my analysis. However, based on these facts, the outlook for urban vegetation in Richmond is positive.

References:

Hammond, K. (2022, December 8). 35,870 trees planted in Richmond blows past 5-year goal, Department of Public Utilities announces. ABC 8News. https://www.wric.com/community/35870-trees-planted-in-richmond-blows-past-5-year-goal-department-of-public-utilities-announces/ 

Locke, D.H., King, K.L., Svendsen, E.S. et al. Urban environmental stewardship and changes in vegetative cover and building footprint in New York City neighborhoods (2000–2010). Journal of Environmental Studies and Sciences 4, 250–262 (2014). https://doi.org/10.1007/s13412-014-0176-x.