This project is an analysis of previous sightings of two rare birds, the caucasian grouse and the caucasian snowcock, to assist with planning of an outing to see the birds in the mountains of Georgia. The analysis has been completed in Python with use of ecological and citizen science data. The conclusions of the analysis can be found at the bottom of this document.
- Background Information
- Data
- Caucasian Grouse and Snowcock Sightings
- Time of Sightings
- Modelling Occurence
- Partial Dependence Plots: Optimising Factors
- Similarly Sighted Species
- Conclusion
The caucasian grouse (Lyrurus mlokosiewiczi) and the caucasian snowcock (tetraogallus caucasicus) are two birds endemic to the Caucasus Mountains, meaning that they can only be found in the wild in this region.
As of July 2023 there are only 572 sightings of the caucasian grouse on eBird, and 487 sightings of the caucasian snowcock. In comparison, the black grouse has over 9,000 sightings while the red grouse has almost 30,000 sightings. The two caucasian species are less-studied then many of their Eurasian relatives.
The caucasian grouse is typically found in the altitunidal range "between upper limits of mountain forests (Picea orientalis, Betula) and subalpine meadows with rhododendron (Rhododendron caucasicum) thickets and stunted birch". After the chicks hatch in the summer, the females often move to more open meadows with the young while the males move to more protected ravines.
The caucasian snowcock inhabits areas of the mountains between the treeline and the permanent snow areas. During breeding season this is typically between 2400m and 3500m above sea level on north-facing slopes and cliffs. After breeding season they have been known to climb to higher altitudes before descending in autumn and winter.
The Caucasus Mountains themselves span a length of 1200km from the Black Sea to the Caspian Sea, and reach to a height of 5,642m at the peak of Mount Elbrus (the tallest mountain in Europe). This is a huge potential range for the birds.
Data on bird sightings is from the eBird Basic Dataset as provided by Cornell Lab. Under the terms of use of the data, the original files cannot be stored on GitHub and so have been included in the .gitignore. If you wish to re-run this analysis you will need to download the from eBird and store it in the correct file path.
GeoJSON data for creating maps of Georgia are from Humanitarian Data Exchange. There provide maps of the country at three different resolutions. The relevant files have been stored in the data folder.
Open Topo Data is used for elevation data. Although the data can be downloaded, we will use their API.
The first step in sighting these two birds is to look at where they have been sighted in the past. This can be achieved relatively simply by filtering the eBird data for Georgian sightings of the two birds and plotting them on a map.
To ensure that locations are accurate (within a few kilometers) we have filtered out any sightings where the birder travelled over 5km. We have also removed sightings that took place over the timeline of more than 5 hours to keep consistent with future analysis.
The code for the filtering and map plotting can be found in bird_sightings.py.
As we can see, sightings for both birds are primarily concentrated in the Mtskheta-Mtianeti region in the north-east of the country. Therefore we will now focus our analysis on this area. The below figure, produced by bird_sightings_kazbeg.py shows the sightings in just the Mtskheta-Mtianeti region.
The time of bird sightings is also important. Different species are most active at different times of the day and your chance of seeing the birds will change along with this.
In time_of_day.py we create two simple plots of when the species were sighted based on the hour that the observations started.
These graphs are not perfect for two reasons. Firstly, an observation on eBird can span multiple hours. If the start time is 10:00 and the duration is 2.5 hours then the bird could have been spotted at any combination of 10am, 11am, or 12pm.
Secondly, observers may be out looking for the birds more commonly at certain times. While most observations of the birds are in the morning, this does not show whether observers are out more in the morning or whether the birds are more likely to be seen in the morning.
An alternative is to model the occurence of the birds and then analyse the models to see what times the birds are most active.
We will fit a simple XGBoost model (a gradient boosted random tree model) to model the occurence of the two caucasian birds.
The features we will use in our model are: latitude, longitude, month, hour, duration, distance travelled, number of birders, speed of travel, and elevation.
All features can be obtained from the eBird data except for elevation. To obtain elevation we will use the Open Topo Data API, which requires an API call for every hundred data. Since we do not have hundreds of thousands of rows of data this will not be an issue, but we would need an alternative method if we had more checklists.
The modelling is done in model_occurence.py. We split our checklists into a training and validation set using a 90/10 split. The caucasian grouse model has an F1 score of 0.8387. The caucasian snowcock model has an F1 score of 0.6857.
We can create a map of occurence chance by feeding data back into the models for predictions. These maps are centred on the town of Stepantsminda, an easy-to-get-to location where many of the bird sightings have been.
The map above is from the caucasian grouse map. We can conclude that we have the best chance of seeing a caucasian grouse in the mountains to the west of the town, but there is also the possibility to see them on the hiking trails to the north east as well.
The modelling for the caucasian snowcock suggests a more restricted area for spotting the bird. If we want to see the caucasian snowcock we should focus our time on the mountains to the west.
Once we have our model we can create partial dependence plots to see how we can optimise each factor.
Returning to our previous hypothesis regarding what time to look for the birds, there is a strong preference for birdwatching in the morning. However, for the caucasian grouse we can also search in the evening.
As you would expect, a longer birdwatching journey has a greater chance of seeing these rare birds. In the case of the caucasian grouse we can increase of chances by birdwatching for at least an hour. For the caucasian snowcock we want to birdwatch for as long as possible.
However, for both birds we don't need to cover as much ground as possible. Distance covered does not have a major impact on the model, and there is a preference for travelling at a slower pace (<2kmph).
Birds rarely inhabit an environment by themselves. If we are looking for the caucasian grouse or caucasian snowcock, what other birds might we spot that could indicate we are in the right environment?
There are several ways of tackling this question with data. The most simple is to look at each checklist and say, whenever a caucasian grouse is spotted, how often do we see each other species? This analysis has been done in indicator_species.py but has a major flaw. Whenever we see a caucasian grouse we might have a 90% chance of seeing a blackbird. Great, so a blackbird indicates we are in the right environment? Not necessarily. We might also have a 90% chance of seeing a blackbird even when we don't see a caucasian grouse.
In cosine_species.py we also calculate a cosine similarity between the target species and every other species, where each species has been vectorised based on what checklists it is in. The output for the caucasian grouse and caucasian snowcock can be seen below.
From this chart we know that if we see one of the caucasian birds then we are likely to see the other one. Ring ouzel, water pipit, northern wheatear, and fire-fronted serin are other birds that are likely to indicate the presence of the two caucasian birds. The blackbird, a bird that our rudimentary analysis suggested was a key indicator, still does appear but it's importance is not as strong.
From our analysis, we can conclude that our best chance of seeing the caucasian grouse and caucasian snowcock in Georgia is by heading to the Mtskheta-Mtianeti region. Specifically we want to head to the mountains to the east of Stepantsminda. For the caucasian grouse we also have a chance of seeing them by walking on the eastern hiking trails.
In terms of our activities, we should head out early in the morning, as the chances of sighting the birds drops off after 10am.
Despite this, we do not need to cover as much ground as possible. It is more important to cover the ground slowly, and to stay out birdwatching for at least an hour. The longer we birdwatch the better, particularly for the caucasian snowcock.
Other birds that we are likely to see are the ring ouzel, water pipit, northern wheatear, and fire-fronted serin. These birds can be used as indicators that we are in the right location to spot the two caucasian birds.