The setting

The goal of the mission was to automate the process of counting pelican nests on Queen Bess Island, Louisiana, USA.

The Trinity F90+ drone was used to map the island off of coastal Louisiana that is critically important for brown pelicans, and other seabird, nesting. Traditional surveys at the islands are done on foot and with airplanes, but the Nelson Ecosystem Lab at the University of Louisiana at Lafayette wanted to apply new count methods and automate the counting process. The Trinity platform allows the researchers to access these remote islands and count the pelican nests while avoiding disturbing these delicate habitats.

The individual mission challenges were:

• Flying at an altitude that avoids any disturbance to the birds and achieve a pixel size of less than 2.5 cm.
• Mapping the entire island, about 17 ha, creating a mosaic in Pix4D
• Counting nests using object-based image analysis in eCognition and pairing it with on the ground survival and breeding information from other bird researchers

The story of pelicans and Queen Bess Island

The island was recently restored by the State of Louisiana because it was rapidly eroding. Louisiana loses lots of marshes each year, but this island is very important for pelicans (the official state bird) and supports 15-20% of all nests in the entire state. The restoration project was a big deal and cost about 19 million USD funded by BP oil spill money.

6,600 pelican nests were estimated in 2018 when only 2 out of 15 hectares were habitable for nesting, prior to restoration. Most of the island was becoming open water and limiting the nesting area. Sand was pumped in, rock barriers installed, and new vegetation planted to keep the island in place. Restoration is an important research topic in Louisiana because it occurs on such a large, coast wide scale, and it is difficult to predict how animals might respond to construction.

The mission was a success because of proper planning, execution, and minimal disturbance to wildlife. The UAV’s mapping capabilities allowed the researchers to study the site quickly and more effectively. The real value of the data is how it can be paired with on the ground survival and breeding information from other bird researchers to test small-scale patterns over the entire island using drone surveying.

The object-based image analysis approach helped make the counts more efficient. Nests were outlined very well using eCognition’s segmentation algorithms. After figuring out the appropriate scale and parameters, John M. Harris was able to effectively delineate nests from other objects and identify them based on spectral and geometric features. Since some nesting areas were more densely populated than others, he had to use multiple rounds of segmentation and classification. The object-based approach increased accuracy because different sections of the island could be classified using slightly different methods.