Alcatraz Seabirds from Space

Alcatraz Island is home to one of the largest urban seabird breeding colonies in the world. The National Park Service and partners have been monitoring the colony for 32 years and have gained valuable insight into the breeding population trends on the island. Here we investigate the use of satellite imagery to assist with population estimates of the Brandt's cormorant colony on Alcatraz, with potential applications to remote colonies where on-the-ground monitoring is not practical.

Monitoring a Growing Population

Three pairs of Brandt’s cormorants began nesting on the island in 1991, and in 2023 over 4,800 breeding pairs were counted. With this growth, the colony is expanding into areas not readily visible from island viewpoints or the water without causing disturbance to nesting birds. Additionally, in 2020, unexpected island closures due to the Covid pandemic prevented researchers from accessing the colony to collect data. These instances of decreased detectability and potential future limited accessibility have led us to investigate new methods to estimate the size of the breeding colony on Alcatraz. 

In 2020, island-based photographs were used to obtain best estimates of the breeding population size. However, aligning individual photographs to ensure area coverage without overlap and hand-counting all birds was time-consuming. This method also relied on some access to the island, and is limited by island viewpoints. Aerial surveys have been employed for censusing nearby seabird colonies in a similar way. However low-flying aircraft and drones have the potential to stress breeding seabirds and, in extreme cases, cause nest abandonment. Additionally, cost, time and accessibility may limit the scope of detecting and quantifying seabird colonies via aircraft. 

Satellite Imagery and Wildlife Research

High-resolution satellite images are becoming increasingly available and are being used worldwide for various wildlife monitoring projects. In one such study, Adelie penguins were censused by identifying the spectral signature of guano stains on the snow. Larger animals, such as elephant seals, have also been individually counted at their rookeries using satellite data. 

Cormorant colonies are well suited to potential aerial and satellite surveys due to the bird's physical characteristics and nesting tendencies. They are large birds, standing almost three feet tall, and their dark coloring contrasts with the typically light guano-covered nesting grounds. They nest above-ground rather than in burrows, usually along cliff edges or atop rocky islands. Within a colony group, nests are systematically evenly spaced, with a ~1-m radius around every nest. On Alcatraz, the Brandt’s cormorant nesting season spans approximately from March through August, with peak nesting activities (sitting on nests, prior to chicks leaving nest) occurring during May and June. 

Acquiring Satellite Images

To conduct proper analysis, we searched for sources of high-resolution aerial images. Ideally, the images would be less than 1-m resolution, to account for the cormorant's size and nest spacing, and be available at least once per week, to capture different phases throughout the breeding season. The available Google Earth imagery of Alcatraz easily resolved individual nests, but is only updated every one to three years.

Maxar and Planet each maintain publicly available repositories of historical high resolution satellite data. These images exist almost daily around around large urban areas, making Alcatraz Island, located within the San Francisco Bay, ideal for this study.

We searched the Maxar image catalog for clear, unobstructed images containing Alcatraz from high resolution (<1 m) satellites. We prioritized finding quality images coinciding with all months during the cormorant breeding season, although frequently in the late spring and summer the island was fully obscured by clouds. We identified 156 suitable images from WorldView-2 and -3 satellites between January 2015 and January 2023, and were able to obtain 57 through NASA’s Commercial Smallsat Data Acquisition (CSDA) program. All images were ordered with a 30-cm resolution panspectral (black and white) component, and 1.24-m multispectral (8-band, 400–1040-nm) components.

Defining and Extracting Data

For initial visual inspection, we created pansharpened images of the colony during the nesting season using ArcGIS. Pansharpening integrates the multispectral and panspectral components into a higher resolution color image. While the pansharpened images showed clear delineations between known nesting areas and areas without cormorants, we were unable to resolve individual nest sites. 

As we were unable to identify individual nests, we sought to extract the spectral signature of nesting cormorant sites. We selected three control sites and two sites with consistent cormorant colonies. The control sites were: 1) Cistern (CIS), a flat cement block at the top of the island with no obstructions and no vegetation, 2) Parade Grounds (PG), a large cement slab on the  southwest end of the island, and 3) Rec Yard (RY), a cement yard at the top of the island surrounded by four tall walls. The colony sites were: North Colony (NC), the largest colony on the island, flat with no obstructions, and New Industries North (NI), the second largest historical nesting site.

The data associated with each site were extracted using ArcGIS’s Extract by Mask tool. The output was an individual file for each site and date containing digital numbers corresponding to every pixel within the site, each with eight spectral bands. Basic atmospheric correction was done using the dark object subtraction model, which subtracts the darkest pixel values of an image from every individual pixel. We used the darkest values with at least 100 representative pixels within the original satellite image to account for possible anomalies and edge effects. These corrected digital numbers were converted to Top of Atmosphere (TOA) Reflectance following the protocols provided by Maxar ( Absolute Radiometric Calibration white paper ). 

Visualizing Data

To analyze seasonal trends in the cormorant colony spectral signature, we created various plots from the corrected reflectance values. In the plots, each spectral band is represented by a different line corresponding approximately to the band's wavelength. Averaging all yearly values for each month, we first looked at seasonal trends of individual sites.

We then normalized the colonies against the controls by dividing the monthly average reflectance of the colony by the control. The resulting value would be higher if the colony was lighter relative to the controls, and lower if the colony was darker relative to the controls. We expect lower values when the dark cormorants are present in their nesting areas.

Results

There appears to be a darkening of the spectral signature of the seabird colonies from March to July, precisely when Brandt's Cormorants arrive to nest on Alcatraz Island. This is observed in the raw data from the colonies (Monthly Reflectance by Site; NC and NI) and in the colony data that have been normalized using the control sites (Normalized Monthly Reflectance; All). This effect is observed across all visible bands and generally at both sites when compared with all control sites. The spectral signature then lightens from July to November as nest sites are abandoned. The nests themselves naturally break down after being abandoned and periodically nest sites are also cleared by National Park Service staff after the breeding season is finished. The decrease in reflectance observed from November to January is something that we can not fully explain. This could be due to smaller sample sizes in the winter or perhaps the impact of precipitation on the spectral signature, as this area generally receives rain only from November to April.

Conclusion

We believe that this project provides promising evidence that seabird colonies impact the spectral characteristics of nesting sites and can be observed from space. The primary limitation to pursuing this sort of work on a larger scale is data availability. Alcatraz Island is in the middle of an urban setting and is frequently imaged, but data are still fairly limited. Sites that are more remote, where the majority of seabird colonies exist, are surely even more data limited. This could be solved by moving away from commercial satellites and using publicly available data. Landsat data are more widely available and could be used over larger spatial scales, but the resolution is much more coarse (30 m for Landsat 8 data). Alternatively, commercial high-resolution satellites could be tasked by researchers to collect sufficient data at specific time intervals if financial resources are available. Certainly as satellite technology continues to improve and become cheaper, satellite monitoring of small seabird colonies such as Alcatraz Island will become even more feasible.