Interested in joining the Silph Research group? Learn More »

Silph Study: #057


Souvenir Survey: Lone Earrings

Have you ever lost an earring? According to the world of our Pokémon buddies, it happens to a lot of us! Since the buddy feature was introduced in December 2019, Silph Pokémon Buddies have found the Lone Earring souvenir almost 2,000 times. While these souvenirs serve no in-game purpose beyond collectibility, we remain intrigued by the way they connect the real world to gameplay.

In previous articles on location-dependent souvenirs, we have discovered connections between Skipping Stones and fresh water features, Beach Glass and beaches, and Tropical Shells and coastlines. Lone Earrings seems to be available only in certain locations as well, and our researchers’ hard work has given us some clues. Time to dive down this Raboot hole and see what our team has uncovered!

Key Findings

  • Buddy souvenirs collected in areas with OSM key/value pairs Landuse=Residential, Highway=Residential, and Amenity=Parking were more likely to contain Lone Earrings than areas without those tags.
  • Lone Earrings are not found at equal base rates across where they are available.

Background and Methodology

Buddy souvenirs can be classified as either universal souvenirs, which are available everywhere, or location-dependent souvenirs, which are only available in a subset of locations. In our researchers’ quest to collect Lone Earrings, we received reports that these misplaced jewels were often found in urban and suburban areas and not at all in more rural areas. This tendency makes sense in real-world terms: The more crowded a place is, the more likely items are to be mislaid. How might the game define the location dependence of Lone Earrings, though?

For answers, we turned again to OpenStreetMap (OSM) for connections. To recap, OSM is a mapping source used in Pokémon GO to build the visible overworld map. OSM data has been shown to influence many hidden features, such as biomes, nests, and — in our previous studies — location-dependent souvenirs.

We define a location area as the level 14 S2 cell containing the player’s avatar. (S2 Cells are a spatial indexing tool used by Pokémon GO, with Level 14 cells used to determine PokéGym eligibility.) An OSM tag is a key/value pair used to distinguish map features. If any OSM element within an S2 cell location has a specific OSM tag, then the entire location is classified as having the OSM tag.¹ For example, the map below shows how we used OSM map data and S2 cell boundaries to classify a town in Norway.ᵃ

The black lines represent the Level 14 S2 cell boundaries. Cells shaded in red are classified as containing the Natural=Coastline tag, which is related to the Tropical Shell Souvenir. You can try this out using the PogoMap tool to visualize the OSM data and S2 cell geometry of your home area. Under Settings/Visual, set Map Type to OpenStreetMaps and S2 Cells as Level 14.

Experiment Design

Given our previous observations of other location-dependent souvenirs, we hypothesized that one or more OSM tags could act as “triggers,” providing a condition allowing Lone Earrings to be found. From a subset of locations where Lone Earrings were found, we narrowed down a list of OSM tags to test, namely: Landuse=Residential, Landuse=Commercial, Landuse=Industrial, Landuse=Retail, Amenity=Parking, Amenity=University, Amenity=School, and Highway=Residential. We’ll refer to these eight tags as “Urban” tags for the remainder of this article.

We looked at 151 different locations where more than 35 souvenirs had been reported. Each location was classified based on the presence of the Urban tags and whether Lone Earrings had been found. Using a Fisher exact test of independence, we tested whether the presence of one or more Urban tags was related to finding a Lone Earring at a given location.


Urban Tag Locations with
Lone Earrings
Locations without
Lone Earrings
Bonferroni Corrected p-value
At least one Urban Tag Yes 111 30 6.09 x 10⁻⁶
No 0 10
Landuse=Residential Yes 69 5 3.13 x 10⁻⁷
No 42 35
Landuse=Commercial Yes 10 3 1.00
No 101 37
Landuse=Industrial Yes 14 2 1.00
No 97 38
Landuse=Retail Yes 20 3 1.00
No 91 37
Amenity=Parking Yes 66 12 0.0150
No 45 28
Amenity=University Yes 12 0 0.33
No 99 40
Amenity=School Yes 25 7 1.00
No 86 33
Highway=Residential Yes 104 27 0.0010
No 7 13

A significant relationship was found between locations which contained at least one of the Urban tags and Lone Earrings. In addition, the single tags Landuse=Residential, Amenity=Parking, and Highway=Residential showed a significant relationship with Lone Earring availability. For the remaining five Urban tags, we couldn’t reject the null hypothesis (that there is no relationship between the OSM tag and Lone Earrings).

Limitations of Results

In our previous study on Beach Glass and Tropical Shells, we found that the related OSM tag was necessary to find the souvenir. Here, while we found significant associations of 3 tags with Lone Earrings, we are less confident that these tags tell the whole story.

Let’s start with this observation: we found Lone Earrings only in locations which had at least one Urban tag. Does this then mean that Lone Earrings can only be found if there is an Urban tag? Not quite. Unlike Natural=Beach and Beach Glass, some of the Urban tags were incredibly common. In this study, our data included only 10 locations that didn’t have at least one Urban tag. The lack of contradiction is less conclusive in this case and could instead be attributed to a majority of our researchers residing in urban areas.

In addition, we noted many locations where an Urban tag was present, but Lone Earrings were not found. Although Lone Earrings are uncommon, we estimated a 95% likelihood that at least one would be found if 36 or more souvenirs were collected in a location. However, our data showed that Lone Earrings were found in only 79% of all Urban-tagged locations. This discrepancy could indicate an additional factor or tag that determines Lone Earring availability (more on that later).

Furthermore, let’s look at the individual tags in isolation. Buddies found Lone Earrings in locations where one of Highway=Residential, Landuse=Residential, and Amenity=Parking was the only Urban tag present. However, Researchers also found Lone Earrings in locations where Landuse=Industrial was the only Urban tag, suggesting it too could be a source of Lone Earrings. This suggests that, in the case of the less common tags, we may simply lack the statistical power to identify a definite relationship.

Given all of the complications discussed so far, we are not quite ready to claim Lone Earrings are always found near one of these Urban Tags. What other factors should be considered?

Lone Earring Rarity

To help unravel this mystery, let us take a closer look at the distribution of Lone Earrings per location. Thus far, we have classified souvenirs into two types. Universal souvenirs are available at every location and are found in consistent proportions at each location. Location-dependent souvenirs are unavailable at certain locations. When available, location-dependent souvenirs are found in consistent proportions to universal souvenirs. The graphs below show the distributions for Cactus Fruit and Small Bouquets as representative of universal and location-dependent souvenirs, respectively.

Excluding a few outliers, both Cactus Fruits and Small Bouquets closely follow their mean base rate at each location, as do the other universal and location-dependent Souvenirs.² Lone Earrings, however, break this trend — they are not found at equal base rates.³

This variability in the Lone Earring base rate actually undermines some of our earlier test assumptions. During the Urban Tag test design, we estimated that Lone Earrings would account for 8.1% of souvenirs in the worst case scenario, where every possible souvenir was available. At a sample size of 36 and above, we estimated that we could determine with 95% confidence that Lone Earrings were or were not available at a location; however, that initial assumption appears to have been incorrect.

Suppose that Lone Earrings are found at a base rate of 30% at some locations and base rate of 3% at others. The 36 minimum sample size would be insufficient to determine if Lone Earrings were available in rarer locations. This would increase our number of false negatives and make it harder to reject the null hypothesis for certain tags. It could also help explain why we didn’t find Lone Earrings in areas which did have an Urban tag. Alternatively, it could also be the case that Lone Earrings are actually universally available but appear at an extremely low base rate in some locations. If that is the case, might there be other metrics that could determine Lone Earrings’ base rate?

Speculation on Alternative Models

Now, let’s propose a few models that could help explain the distribution of Lone Earrings. These models discuss what could be used, not what necessarily is being used, in Pokémon GO.

A simple solution could be the existence of an additional OSM tag which is a strong predictor of Lone Earring availability. This would help explain the numerous locations which had an Urban tag but did not produce Lone Earrings. Even then, an additional tag relation would explain availability but not the variable base rate.

A more nuanced approach would be an interactive effect between multiple OSM tags. In this model, we would expect a location would exhibit a higher Lone Earring base rate if it contains more tags or if it has larger tagged regions than other locations. There is some evidence of this relationship in our dataset. At locations where Lone Earrings were found, we found a positive rank correlation between the number of Urban tags and the observed base rate (rs(107) = 0.37, p < 0.001). Due to limits in the data we collected, we are unable to test tagged region size or multiple instances of the same tag.⁴ While this interactive effect from multiple Urban tags helps explain some of the base rate variance, it could also be the case that Urban OSM tags are indicative of another factor altogether.

Another possibility is that satellite imagery data is used to classify areas according to their land use. It has long been speculated that another spatial data set is used in Pokémon GO to inform Pokémon habitats. A variety of supervised, semi-supervised, and unsupervised algorithms can be used to generate clusters of similar land cover. The pictures below show an OSM map, satellite map, and the Copernicus imperviousness land cover map for the same region in Nordland, Norway.

Impervious areas are the result of human-made buildings and surfaces which prevent water from being absorbed into the ground. The Copernicus database assigns a scaled measure of imperviousness (0-100%) marked in varying shades of red, with more impervious areas in darker shades. The Urban OSM tags that we found correlated with Lone Earrings (roads, housing areas, and parking lots) are the same features that the Copernicus land cover classification algorithm defines as impervious areas.⁵

One could imagine that the different Lone Earring rarities we observed are driven by the level of urban development in the area, with highly impervious areas yielding Lone Earrings more often than less impervious areas. We can see this in the map above, where Lone Earrings were found at higher base rates in highly impervious (dark red) areas.

The Copernicus land-use classification algorithm is limited to Europe, so it is likely that Niantic would instead choose to source data from a global source or perhaps even use an in-house system. Without more data, it is hard to say which other land classification source could be used to determine Lone Earring rarity, or if any are being used at all. We are left with quite the mystery!

Parting Words

Special thanks to the many researchers who dedicated countless hours collecting data in blazing sun and freezing winds. To the analysis team, who spent well over a year wrestling data and persisting through query timeouts. To the buddies, who on occasion took their sweet, sweet time picking out a souvenir. And to you for reading this far. In the end, we are left with more questions than answers. After all, that is the magic behind the world of Pokémon…the reason we continue to GO! Until next time,
-Project Tamagotchi


Authors: Scientists CaroKann and Titleist and Lead Researcher Sinkalingsveis
Analysts: Lead Researcher Sinkalingsveis, Scientist CaroKann, Scientist Titleist, Lead Researcher RDC-DCIfan, and Senior Researcher Lauracb
Project Leaders: Scientist Pancake and Lead Researcher Belle
Graphic Artist: Scientist CaroKann
Editors: Scientist Cham1nade and Lead Researcher JinianD
Additional OSM support: Scientists Gustavobc and WoodWoseWulf

Over 350 researchers and 2,200 Pokémon buddies lovingly contributed nearly 22,000 souvenirs to this study. The following researchers ransacked their home ranges in search of misplaced jewelry:

  • archer
  • ~Cables~
  • CaptSinise
  • DeeDillyDawn
  • JMcQueen81
  • marinaj
  • Mudkip
  • Reykius
  • shenannygans
  • silvietta23
  • turtleduck
  • Zenigel


¹ A location’s OSM tags were collected using Overpass Turbo. At each location, we established a bounding box which contained the level 14 S2 cell area. The bounding box was then passed to a custom Overpass Turbo query which searched the area for the Urban tags. Using the output, we classified the location-area as either having or not having the specified OSM tag.

Each query only considered tags created on or before February 25, 2019, the most recent map source behind game mechanics. The map update from July 2021 visually changed the map but did not affect underlying features such as nests. We have no evidence of location-dependent souvenirs being affected by this update either.

Because the bounding boxes used in Overpass do not perfectly match the S2 cell geometry, we selected a slightly larger bounding box area for our queries. Manual verification was done for Urban OSMs in which the tag lay close to the level 14 S2 border.

Due to query runtime, storage, and privacy limitations, we only collected data on the 8 urban tags defined in this study. Below are the custom queries used to test each location:

Urban Tag Query (no Highway=Residential)
Highway=Residential Tag Query

² We calculated a souvenir’s “base rate” as the ratio between a souvenir’s frequency and the frequency of all other universal souvenirs present. This is expressed below for universal souvenirs and location-dependent souvenirs, respectively:

The base rate represents the “best case” scenario where no other location-dependent souvenirs are present at the location. Across all locations where Lone Earrings were found, the mean base rate in our study was 15.44%. Due to the observed variation in base rate, however, we cannot confidently apply this base rate to each individual location.

³ Using a chi-squared goodness of fit test, we compared a souvenir’s frequency at each location to the number we would expect given the souvenir’s base rate and the total number of universal souvenirs found at the location. We considered locations with 60+ total souvenir observations. For non-universal souvenirs, we excluded locations where the souvenir was not found and would thus not follow the base rate.

We rejected the null hypothesis that Lone Earrings are found at an equal base across the 50 locations where they were available (χ²(49, N = 626) = 278.5, p-value < 0.001). We failed to reject the null hypothesis for both Cactus Fruits (χ²(64, N = 1,036) = 36.4, p-value = 1.00) and Small Bouquets (χ²(29, N = 751) = 22.2, p-value = 0.88). The latter result held true for the other universal and location-dependent souvenirs for which we had sufficient statistical power to test.

⁴ Our study collected only limited OSM data. We defined whether a tag was present or not; we did not store each tag instance. Even if we had collected more detailed tag data, there is no clear way to define tag density. Do we count each node instance, each polygon instance, the polygon area, or only the polygon area within the S2 cell? Unclosed polygons, such as Highway=Residential, do not have areas and further complicate this question.

⁵ For more detailed information on how the Copernicus land-cover algorithm works, see the Copernicus user manual.

Mapping sources

ᵃ Copyright OpenStreetMap contributors. For more information, please visit

ᵇ Copyright Esri, all rights reserved. For more information about Esri® software, please visit

ᶜ Copyright Copernicus Land Monitoring Service 2018, European Environment Agency (EEA). For more information, please visit

Additional modifications were made to the map sources including S2 cell and Souvenir overlays.