A Study of Avian Population Recovery after Habitat Restoration using Remote Sensing and Citizen Science Data


Link to Slides:


My project goal was to use satellite remote sensing data along with citizen science eBird observation data to study of how habitat restoration has affected various bird populations at Fernhill Wetlands, an important bird habitat near Portland Oregon.

I used Google Earth Engine (GEE) to generate and analyze multi-band LANDSAT and SENTINEL satellite imagery using GEE’s programming interface. I extracted quantitative information from the images, including indices for vegetation (NDVI and EVI), and water surface area (NDWI), among others.

Fernhill Wetlands has over 100 bird species. I expected that land-bird populations would be positively correlated with vegetation, while deep-water species would benefit from higher NDWI. Shorebird behavior would be more complex.

The NDVI, EVI, NDWI, temperature and rainfall data from 2013-2018, as well as eBird bird counts through time were combined in a multivariate analysis using R.

The analysis matched expectations for land and deep-water birds, with interesting results for shorebirds. For example, marsh songbirds showed a positive correlation with the water index, probably because the pond perimeter that they inhabit increases with water area. This data will be extremely useful for Fernhill authorities to manage the refuge.

My work shows that satellite remote sensing, in combination with eBird data, can be a powerful tool for analyzing and studying habitat impact on wildlife. I plan to use this technique to study other important wetland habitats in Southern California. This is especially important given how climate change will continue to affect the ecology of California.

Question / Proposal

Fernhill Wetlands is an important bird habitat near Portland, Oregon. In 2014-2015, it underwent a massive habitat restoration, transforming it from large unused sewage ponds into a complex native wetland habitat, complete with dry land area, open water and marsh.

The key question that needed to be answered was:

How did the habitat restoration affect the populations of various avian species at Fernhill Wetlands? 

The Portland Audubon Society used bird survey data to study the avian population recovery and published their results in early 2017.

The goal of my project was to go further and expand on the Audubon study by using remote sensing and citizen science eBird data to do a detailed, quantitative study of avian fauna at Fernhill.

Specifically, I wanted to:

1.       Generate satellite remote sensing multi-band data maps of Fernhill Wetlands

2.       Extract quantitative metrics for vegetation cover, water surface area, etc. from the satellite data

3.       Perform a statistical analysis to establish correlation between vegetation area, water area, temperature, rainfall, etc. with the populations of various bird species obtained from eBird.

My hypothesis was that increasing the vegetation index would benefit land species like raptors and songbirds, while decreasing open water would adversely affect diving ducks and other deep-water species.

Additionally, the more interesting thing I wanted to find out was how shorebirds would be affected. Since they depend on both water and vegetation, I expected them to behave in more unpredictable ways depending on the relative impact of the vegetation and water indices.


The Portland Audubon Society performed an earlier study of the Fernhill Wetlands project. They used human observers to collect bird data and used it to perform an analysis through time to estimate the effect of habitat restoration on avian populations. The study was performed in 2016 and did not have some of the latest data on the further evolution of the habitat and how it continues to evolve and affect avian populations.

There has been a long history of the use of satellite imagery to study how habitats change and evolve. What is new is the advent of Google Earth Engine. This is an extremely powerful technique that makes it much easier for the end user to generate, analyze and extract remote sensing data from a variety of spacecraft.

eBird is a birder database where bird enthusiasts and citizen scientists can enter their observations for public use. Since the people doing the entries are largely knowledgable amateurs, the quality of the data is generally very high. 

My goal was the use the power of Google Earth Engine to quantify the vegetation and water areas at Fernhill and use R to identify correlations with bird populations as documented in eBird. In this manner, I was building on the work done by Portland Audubon Society in conjunction with Clean Water Services. I consulted both of these organizations and the researchers who had done the earlier work, and they were very generous with their time and data and were excited to see what additional insight could be obtained through the use of remote sensing data.

My goal with this project is to establish the methodology for combining satellite remote sensing data with eBird data to perform powerful analyses of avian populations and how they are affected by habitat change. I believe this is the first time this kind of analysis is being performed. 

With global warming becoming an increasingly serious threat, I believe that this kind of analysis is going to be critical to study how habitats need to be adjusted and protected to minimize pressure on avian populations.

Method / Testing and Redesign

Figure from Summary Powerpoint:


1. First, I explored and identified all needed datasets for the investigation. Satellite images would be obtained from LANDSAT-8 and SENTINEL-2, climate data from PRISM, and bird population data would be obtained from eBird.

2. I verified that all the data would have the temporal and spatial resolution required to be able to do an effective analysis at Fernhill Wetlands.

3. I acquired Landsat 8 and Sentinel 2 imagery and PRISM climate, and eBird datasets. These involved the use of Google Earth Engine and special access permission to download eBird datasets.

4. I preprocessed Landsat and Sentinel imagery using Google Earth Engine by picking least cloudy scene and applying masking techniques to discard problematic pixels in the scene.

5. I sanitized eBird data to exclude reports with no bird counts,  observations that did not include bird species, and observations of hybrids and domestic species.

6. I developed Javascript code in Google Earth Engine to obtain NDVI, EVI, NDWI indices and summary temperature and precipitation data.

7. I performed time series raster data analysis in Google Earth Engine for year to year changes in vegetation index, water index, mean temperature and precipitation.

8. I classifed eBird raw data into bird categories and obtained mean counts of species per category.

9 .After organizing the eBird data in this manner, I inputted the merged eBird and satellite data into R for correlation analysis.

10. I performed correlation analysis in R between eBird, satellite, and climate data. The data was obtained in correlation plots as well as tables.

11. After completing the correlation analysis, I observed the data to look for patterns in how different types of bird species reacted to the input variables. I thus evaluated the effectiveness of using this alternate mechanism to assess biodiversity changes after restoration activity.


My independent variables were time, as well as the amount of vegetation, the amount of water surface area, the temperature and precipitation. The dependent variables were the observation count for various bird species at Fernhill Wetlands.

I ensured that the experiment was fair by averaging the data collection over multiple months, multiple observers and by clustering groups of bird species into one group. For example, I clustered all the sparrows and other small passerines into one group called songbirds because their behavior and their reaction to vegetation and water indices are going to be similar, and there is more data to enable robustness against noise.

The data collection was done by myself and other eBird volunteers at Fernhill Wetlands in Forest Grove, Oregon. I performed all my data analysis at home. The only equipment I used were binoculars, spotting scope and camera, as well as a notebook to take counts of bird observations. I also used a laptop to connect to Google Earth Engine and to do the statistical analysis using R.


Figiure 1: Correlation Matrix Plot of Bird Categories with Vegetation/Water Indices and other Bird Categories:



  • As expected, there is a strong correlation between NDWI and diving ducks, cormorants and swans, as all these birds prefer deep water. These are also negatively correlated with NDVI as more vegetation implies less open water. The data showed that diving duck population decreased from 2013 to 2018, because the area of open water was drastically reduced after the habitat restoration was complete.
  • LIkewise, dabbling ducks prefer vegetation, both on shore and in the water but at the shoreline, and therefore, are positively correlated with EVI, as well as NDVI. 
  • Raptors require large areas of vegetation for their territories as well as for cover and nesting. Also, they need to find prey and therefore are strongly positively correlated to NDVI
  • All of the above were expected results. However, the study, as expected for shorebirds, also highlighted some unusual results.
  • Interestingly, secretive marsh birds, which are shore birds that live in marshy vegetation, are positively correlated, not with EVI or NDVI but with NDWI.This may seem paradoxical, but the main reason is that more water area implies a longer perimeter or shoreline that provide habitat to these birds. I knew that shorebirds could have unusual results, but I did not anticipate this when I started the project. It is an example of how shorebirds sometimes have unpredictable responses to the habitat
  • Similar bird groups show positive correlation to each other (for example, raptors and dabbling ducks). However, this is co-incidental and not causal.
  • The climate data (temperature and precipitation) did not have a strong effect on bird populations when the same times are compared from year to year. 

Error Analysis: 

  • Being in a very rainy and cloudy area of the country, Fernhill Wetlands had many cloudy days and Landsat images were not available for a few weeks. Other images close by temporally had to be substituted for the analysis.
  • The Landsat image resolution was 30m, which is coarser than what I would have liked. The Sentinel image resolution in 10m which is much better, but the data only was available from 2015 onwards
  • The lack of resolution prevented me from being able to distinguish between different plant species, something that would have been very useful
  • eBird data is based on human observation and hence subject to error or biases. I attempted to remove bad data and increase sampling to reduce errors and biases in eBird data


  • The use of satellite remote sensing data, in combination with eBird bird observation data is a powerful combination that enabled the analysis of the effect of habitat change on bird populations in a very straightforward and accurate manner
  • The analysis uncovered several unusual effects of habitat on certain bird species, especially those that prefer the shoreline. Marsh wrens are a perfect example, where they actually had a positive correlation with NDWI
  • The songbird group also had a weak negative correlation to NDVI which is unexpected. It is believed that this is due to the removal of some non-native species that the songbirds have come to rely on. This was a very unexpected result and may need more native plantings to compensate for the loss
  • The final project will be reviewed with Clean Water Services and Portland Audubon to determine what adjustments in the continued habitat restoration would be warranted based on the data
  • The technique of combining satellite and eBird data has been shown to be much more powerful than using eBird data alone as was done in the prior work for this site, because the latter only shows what is happening without being able to pinpoint the cause. This technique will be used in other wetland habitats in the near future

About me

I am a 9th Grade high school student living in Pasadena, California. I was born in Portland, Oregon and lived there until earlier this year, when my family relocated to California.

I have been an active birder since childhood. I love being outdoors and try to go birding every weekend if possible. I am extremely interested in ornithology, ecology, conservation, and evolutionary biology, and my dream is to become a wildlife biologist and researcher after graduate school. I have been an enthusiastic participant in various citizen science projects – including Project Feederwatch for the Cornell Lab of Ornithology, annual Christmas bird counts for the Audubon Society, and as an active contributor to the eBird database.

I am also very interested in geography and computer programming and this is what prompted me to combine them with my love for birding and the environment to study the effect of habitat restoration on avian populations.

My greatest science heroes are, without a doubt, Sir Charles Darwin and the explorer Alexander von Humboldt. Among living people, I admire Sir David Attenborough, the great naturalist, as well as Drs. Peter and Rosemary Grant, the scientists who have spent decades doing amazing research on the finches of Galapagos.

Winning a prize at the Google Science Fair would be a great honor and an amazing experience for me. I am especially excited about the prizes that would allow me to travel with scientists to different parts of the world and learn about wildlife and conservation.

Health & Safety

My project was done at home and independently. The health and safety procedures that I followed mainly had to do with transportation to the birding sites and following good ergonomic practices at home.

My parents assisted me by providing me rides to the birding sites and back in a safe manner and ensured that I did birding in safe spots. 

Since most of my project was done on the computer, I ensured that I followed good ergonomic practices and did not sit at the computer for too long at a time. My parents also provided me with a good ergonomic chair that could be adjusted to the right height so that my eyes were level with the monitor screen. I worked with a mouse and keyboard setup on a desk. I took frequent breaks from the computer and stretched and walked around during the breaks.

My mom's email is rajshreesankaran@yahoo.com

Bibliography, references, and acknowledgements

Avian Response to habitat restoration at Fernhill Wetlands - Joe Liebezeit, Candace Larson, Jay Withgott and Jon Plissner, Portland Audubon Society (https://audubonportland.org/files/citizen-science/fernhill-preliminary-report-2017)

McFeeter, Stuart K. “Using the Normalized Difference Water Index (NDWI) within a Geographic Information System to Detect Swimming Pools for Mosquito Abatement: A Practical Approach.” Remote Sensing 5.7 (2013): 3544-561.

Pettorelli, Nathalie, Jon Olav Vik, Atle Mysterud, Jean-Michel Gaillard, Compton J. Tucker, Nils Chr. Stenseth. “Using the satellite-derived NDVI to assess ecological responses to environmental change.” Trends in Ecology & Evolution 20.9 (2005): 503-10.


I wish to express my sincere gratitude to the following people who encouraged and supported me and were very generous wtih their time, data and patience:

1. Joe Liebezeit - Avian Conservation Manager at the Portland Audubon Society, Portland OR, for giving me data and feedback on the Portland Audubon Fernhill study and encouraged me to conduct this analysis

2. Jared Kinnear - Water Resources Project Manager, Clean Water Services, Forest Grove, OR, for giving me a lot of detailed information on how the restoration was planned and executed at Fernhill Wetlands

3. Erica Higa - Developer, NASA Jet Propulsion Laboratories, Pasadena, CA., for discussions on satellite imagery and extraction of indices.

4. Dr. Benjamin Holt - Member of the Technical Staff, NASA Jet Propulsion Laboratories, Pasadena, CA., for providing guidance and advice on the project.