Project avatar for Iguanas from Above

Iguanas from Above

We are back! Count marine iguanas from Fernandina – expect to find lots!

Results

Preliminary results 2nd phase (Española and Floreana islands) - April 2023

We want to thank to the 9,525 volunteers who so far have helped us to identify and count marine iguanas, other species and objects in the Galápagos. We also want to kindly thank all our loyal volunteers for working together with us over these years!

On the second phase, 3,039 volunteers helped us make 196,067 classifications belonging to 9,097 subjects. We experienced a great response of volunteers to the beautiful christmas iguanas.

Special thanks to our 2nd phase top 10 classifiers. We really appreciate all your effort!
User name and number of images classified:

Initial approach

On our second round, we selected once again a gold standard (GS) set of images, from which we identified presence or absence and counted marine iguanas, to compare volunteer results against expert results. This phase included 12 colonies from Española and Floreana islands. We randomly selected the same number of images per site to come up with a total of 456 (5%) images of the 9097 that our volunteers classified. We used a majority vote criteria for the comparisons.

What did we find?

First, we found that from the 456 images, only 176 (39%) had marine iguanas. This time, iguanas were more common on the images than first phase, where only 5% of the GS set had iguanas.

Second, we analyzed volunteer accuracy regarding marine iguana presence or absence in the images. From images WITHOUT marine iguanas, volunteers classified with 100% of accuracy. This means that tendency of volunteers to not overcount individuals in our images remained over the new phase. This helps to reduce the amount of images to review.

From images WITH marine iguanas, volunteers classified with 82% of accuracy. There was a 30% increase in the correct identification of marine iguanas by volunteers in this phase compared to the previous one.

For 2nd phase we improved image quality and marine iguana abundance in the images, in response of the results and feedback from 1st phase. We can infer that this change influenced our new results, especially from the frequent volunteers who, by classifying a greater number of iguanas, we believe, increased their searching experience.

However, ideally we need to have results with a similarity of at least 95%. Therefore, we did not move forward to compare counts but we looked for a different approach to analyse identification results.

Loyal-volunteer approach

Instead of making our comparisons based on a majority result, we identified the minimum number of volunteers needed to confirm that an iguana is present in an image. Within these volunteers, it is very likely to find classifications from our most loyal volunteers (since they are classifying hundreds and thousands of images) and since they are the ones with the most experience, we expect more accurate results.

So, what did we find this time?

We found that we need at least five volunteers classifying yes for iguana presence to accept that as correct. After applying this new criteria, we obtained a 95% accuracy of the volunteers. With the previous approach we were accepting as correct with at least 11 volunteers (majority vote), and that reduces the accuracy as you can see in the graph below.

As we achieved the ideal similarity percentage, following this criteria, we proceeded to compare the counts of the volunteers with those of the experts.

Let's remember that each image is classified by 20 volunteers - we used the arithmetic median to aggregate these results into a single value per image. Next, we compared both counts for the 456 images (GS image set).

How accurate are volunteers counting marine iguanas?

From our GS image set, experts counted 601 marine iguanas while volunteers counted 480. The latter represents an 80% of accuracy. We went back to our data from 1st phase and after applying the same criteria, we found similar results. Therefore, based on both phases, we know that we need to add a 20% to the final counts we will obtain from all our images. We are waiting to finish our third phase to confirm the pattern.

What’s next?

We are looking forward to analyze results of 3rd phase, confirm our pattern and validate with a publication citizen science approach as an accurate method to count marine iguanas from aerial images and help to estimate the population size of this species.

Moreover, we are planning to launch a 4th phase this year, which will include images from Isabela and Fernandina islands. The Fernandina colonies are the most abundant in terms of marine iguanas and we need to analyze, when lots of iguanas are present in an image, how accurately volunteers can count. But also, we would love to finally give our volunteers images with numerous iguanas so that they can have more fun while classifying them.

Remember April is Citizen Science Month, help us classifying all the images you can!

The Iguanas from Above team

Preliminary results 1st phase (San Cristóbal and Santa Fe islands) - October 2021

A huge thank you to the 5,060 volunteers who helped us to identify and count marine iguanas and other species from the Galápagos! You helped us make 495,472 classifications belonging to 24,373 subjects over four interesting months – every click counts, what a great effort! We are really grateful for all your input.

Special thanks to our top 10 classifiers.
User name and classifications done:

In this first round of the project, we are focusing on understanding how to get a similar quality of results from our volunteers as we get from our expert team. What we learn here, we will use in the next phase, and our end goal is to calculate population sizes of marine iguanas directly from the counts of our Zooniverse volunteers.

As a first step, we have randomly selected a small set of images, which we call the “gold standard” set, to analyze the results; we took 2733 (11%) of our total images from three localities on San Cristobal and Santa Fe islands: Loberia, Punta Pitt and El Miedo. In this dataset we compared your counts from Zooniverse to the counts from our team of experts in order to see how similar the outcomes were.

How accurate were the volunteer counts?

The first thing to note is that marine iguanas were relatively rare (only around 5% of images contained them). In images where no iguanas were shown, volunteers classified with 99.8% accuracy when compared with experts (Fig. 1A). This is great because it shows you are not likely to accidentally overcount the iguanas, well done!

However, in the 5% of photos where iguanas were found, only 52.3% (Fig. 1B) of the images were correctly classified, so volunteers do seem to be missing quite a few of the animals. We think this is related to image quality, as well as the amount of experience each volunteer has. Based on discussions we saw on our message board, we are confident that many of our “super volunteers” were just as good as our experts at finding the iguanas, but perhaps some volunteers who only tried a few images might be missing them.

Next, we separated the images into those with good quality (visible and clear) and bad quality (hard to spot and/or blurry), according to an expert view. When the image quality was good you obtained 75% (Fig. 1C) of accuracy. At Loberia—the site with the best quality images—you obtained an 80% accuracy (Fig. 1D), this is close to the 95% agreement we are looking for!


Figure 1. First results regarding accuracy from volunteers’ data analysis. (A) from images with no iguanas; (B) from images with iguanas; (C) from images with iguanas and good quality condition, and (D) from images with iguanas and good quality at the Loberia site.

These findings confirm insights given to us from various volunteers regarding poor image quality and scarcity of iguanas being the major difficulties. We are also using volunteer feedback to improve our tutorials and workflow. Although the overall expert counts are not yet similar enough to the Zooniverse results, we are still working on this data and may be able to filter based on volunteer experience to see if this improves the accuracy of the counts. We will keep you posted on our findings.

What else did we find?

We also had some interesting inputs regarding other species and the presence of marine plastics in our data. We will continue to collect this data, and we are starting to collaborate with other scientists at the Galapagos Conservation Trust who are interested in the topic of oceanic plastics on the islands. Also, since you were able to identify reproductive males in the photos, we will be able to use the data to better understand the reproductive behavior of the iguanas.

Where are we going from here?

We were delighted to see so much interest in the project and although the first results are not quite good enough for publishing, we really think with some improvements we will get the results we need. This is just the first test stage of the project, and we had a lot to learn about how to take and process the images. The good news is that we have identified some important issues and we are modifying our fieldwork and image processing accordingly. We are confident that the results of our next phase will be much better, and so you will be able to contribute directly to our objective of estimating the population size of this endangered species in the shortest possible time.

We have already done fieldwork for the second phase, on Española and Floreana islands early this year, and here we have obtained images with better quality. In these sites, the abundance of the iguanas was very high, so we think this will help to train the eye and avoid missing any animals. Furthermore, the iguanas this next time will be very colorful (red/pinkish with turquoise, known as christmas iguanas) and looking cute — this should keep you all engaged!


Christmas iguana (Amblyrhynchus venustissimus) from Española Island, Galápagos, Ecuador. A photo of an adult male in the left and a drone image of another adult male from above in the right.

We plan to post the next round in early 2022. This is a little later than originally planned for various reasons, but we really look forward to seeing our loyal volunteers once more. Thanks again for your help and interest, and please stay tuned for the next phase!

The Iguanas from Above team