我们需要一份可溯源、可商用、可开源的全球罕见度数据。 评估了几个候选数据源之后,GBIF(Global Biodiversity Information Facility)是唯一同时满足规模、许可、可引用性三项的: We needed citable, license-clean, fully open rarity data with global coverage. Among the candidates we evaluated, the Global Biodiversity Information Facility (GBIF) was the only one that checked all three boxes — scale, license, citability:
为了让「0-100 分罕见度」易读,我们映射到 5 个层级,每级配一个圆形充填图标和颜色。 To make a raw 0-100 score legible, we bin it into 5 tiers, each with a filled-circle glyph and a colour.
| Glyph | 中文 | English | Score | 大致语义 Roughly means |
|---|---|---|---|---|
| ○ | 常见 | Common | 0 – 7 | 后院常客、城市鸟、广分布 Backyard regulars, urban birds, wide distribution |
| ◔ | 能见 | Occasional | 8 – 24 | 季节性或局部常见,野外能见到 Seasonal or locally common; findable in the field |
| ◑ | 少见 | Uncommon | 25 – 49 | 需要特意寻找,不是天天能见 Requires deliberate searching |
| ◕ | 罕见 | Rare | 50 – 74 | 多数鸟人一年看不到几次 Most birders see these only a few times a year |
| ● | 传奇 | Legendary | 75 – 100 | 一辈子能拍到一次就值得记入鸟生 A once-in-a-lifetime capture |
阈值刻意设成非均匀(8/25/50/75),原因是真实分布是右偏长尾—— 大多数常见鸟挤在 0-15 分,把 0-100 平均切五段会让 90% 的鸟变成「常见」。现在的阈值能让每一档都有足够多的鸟落入。 The thresholds are intentionally non-uniform (8/25/50/75) because the real-world distribution is right-skewed long-tail — most common birds cluster in 0-15. A uniform split would collapse 90% of birds into "Common". The current cuts give each tier a meaningful population.
s3://gbif-open-data-us-east-1/occurrence/<YYYY-MM-DD>/occurrence.parquet/*
Source: s3://gbif-open-data-us-east-1/occurrence/<YYYY-MM-DD>/occurrence.parquet/*
license ∈ {CC0_1_0,
CC_BY_4_0}
License filter: only CC0_1_0 and
CC_BY_4_0 records are counted
reverse_geocoder
离线解析为 ISO 国家代码,再查国别罕见度
Country resolution: photo GPS is reverse-geocoded offline
to an ISO country code, used for country-scoped lookups
直接用 GBIF 的 count 当排名会让「麻雀级」鸟把 99% 的分数挤掉。
我们用 log-normal 归一化把分布拉平到 0-100:
Raw GBIF count is heavily right-skewed (city sparrows dwarf rare
endemics). We apply log-normal normalization to spread the score across 0-100:
score = 100 × (1 - (log10(count + 1) - log10(min_count + 1)) /
(log10(max_count + 1) - log10(min_count + 1)))即:观察记录越少,分数越高。 Fewer observations → higher rarity score.
GBIF 观察次数低不一定意味着真的「濒危」——也可能是冷门科研对象。 我们用 IUCN 红色名录给「应该罕见」的鸟设下限: A low GBIF count doesn't always mean a species is actually endangered — sometimes it's just understudied. We use the IUCN Red List status as a floor to ensure threatened species are scored as rare:
| IUCN | 下限分数 Floor score |
|---|---|
| CR (Critically Endangered) | 90 |
| EN (Endangered) | 75 |
| VU (Vulnerable) | 60 |
| NT (Near Threatened) | 45 |
实际 GBIF 分数若高于 IUCN 下限则保留;若低于则提到下限。 The IUCN floor only applies when the GBIF score falls below it.
某些常见鸟在 GBIF 上 count = 0(地区性数据采集差异、学名同义词未合并等原因),
原始算法会给 100 分——明显错误。我们对这 47 个异常物种,用同科属邻居 GBIF 中位数作 proxy
重打分,写进离线预计算表里。
Some common birds have count = 0 on GBIF due to regional sampling
gaps or unmerged synonyms; the naive algorithm scores them 100 — clearly wrong. For these 47
species we substitute the median GBIF score of same-family neighbours as a proxy
and bake the corrected value into the offline precomputed table.
4 个物种因为分布特殊(岛屿特有 / 单点遗存 / 中国大陆稀客等),算法分数明显违反鸟圈直觉,单独修正: 4 species where algorithmic scores clash with field-birder intuition have explicit hardcoded overrides:
| 学名 Scientific name | 分数 Score | 备注 Note |
|---|---|---|
| Quoyornis georgianus (White-breasted Robin) | 40 | 澳洲西南端特有 Endemic to south-west Australia |
| Butorides sundevalli (Galapagos Heron) | 12 | Galapagos 特有但常见 Galapagos endemic, but common there |
| Pyrocephalus obscurus (Vermilion Flycatcher — Galapagos clade) | 18 | 亚种分级争议 Contested subspecies split |
| Pteroglossus erythropygius (Toucanet sp.) | 20 | 局部常见 Locally common |
照片选中后,右侧详情面板会显示三行连续的鸟类信息: The right-side detail panel shows three adjacent rows when a photo is selected:
鸟种 / Species: Black-capped Chickadee ◔ (点击复制学名)
全球罕见度: Occasional (12/100)
IUCN: LC (Least Concern)点击鸟种行可一键复制学名到剪贴板。 Click the species row to copy the scientific name to the clipboard.
排序下拉框里的 「罕见度↓」 是默认排序,照片按 GBIF 罕见度从高到低排列, 让最难得的鸟一眼可见。下拉框另有文件名、锐度↓、美学↓、鸟种颜值↓ 四个选项。 The "Rarity↓" option in the sort dropdown is the default — photos are ordered by GBIF rarity score, putting your rarest finds at the top. The dropdown also offers filename, sharpness↓, aesthetics↓ and species-beauty↓.
每只识别到的鸟,控制台日志会跟一个 tier 图标: Each identified bird appears with a tier glyph in the log:
[BirdID] Common Crane (Grus grus) — confidence 84% ◑
[BirdID] House Sparrow (Passer domesticus) — confidence 91% ○跑批结束时还会打印整批次的 tier 分布摘要: At the end of a batch, a tier breakdown summary is printed:
[Rarity Summary] ○ Common: 124 | ◔ Occasional: 38 | ◑ Uncommon: 12 | ◕ Rare: 3 | ● Legendary: 0「按物种优先 / 按评级优先」二选一开关在设置 → 输出: A toggle under Settings → Output picks between two organization strategies:
Black-capped Chickadee/3 Star/…
Species-first (default): Black-capped Chickadee/3 Star/…
3 Star/Black-capped Chickadee/…
Rating-first (legacy layout): 3 Star/Black-capped Chickadee/…
不管选哪种,1 星 / 0 星 / -1 星 的照片都会被归到 Other Birds/
分支,避免主目录被低质量副本污染。
Regardless of which is selected, 1★ / 0★ / -1★ photos always go
to the Other Birds/ branch to keep the main tree free of low-quality copies.
罕见度与相关分数写入这三个冷门 XMP 字段(Lightroom / Bridge 主面板都不显示它们, 所以不会和你自己的元数据打架)。写入的都是纯数值,没有前缀文字: Rarity and related scores go into three rarely-used XMP fields (invisible in the Lightroom / Bridge main panels, so they won't collide with your own metadata). Values are written as bare numbers, with no prefix text:
| 字段 Field | 内容 Content | 示例 Example |
|---|---|---|
XMP-iptcExt:Event |
GBIF 罕见度 0-100 GBIF rarity 0-100 | 18.00 |
XMP-iptcCore:IntellectualGenre |
IUCN 红色名录等级 IUCN Red List category | LC |
XMP-iptcExt:AdditionalModelInformation |
iRateBird 鸟种颜值 0-100 iRateBird species beauty 0-100 | 63.50 |
Lightroom / Bridge / digiKam / ExifTool 都能直接读到这些字段。 These fields round-trip through Lightroom / Bridge / digiKam / ExifTool.
GBIF 罕见度反映的是全球观察密度,不直接等于「物种保护级别」: The GBIF score reflects global observation density — it is not a substitute for conservation status:
| 场景 Case | GBIF 高分 ≠ IUCN 濒危 High GBIF score ≠ IUCN threatened |
|---|---|
| 极偏远岛屿特有种 Remote-island endemics | 全球观察少 → GBIF 高分;本地可能并不少 Few global records → high score; may be locally abundant |
| 难以拍照的夜行性 / 林冠层鸟 Nocturnal or canopy-dwelling birds | 观察少 → GBIF 高分;种群可能稳定 Under-observed → high score; population may be stable |
| 学名最近被拆分 / 合并 Recently split or lumped taxa | 数据库未同步 → 分数可能偏差 Database lag → score may drift |
最佳实践:把 GBIF tier 当作「拍到这张照片的稀有度」—— 它帮你快速识别值得打星的瞬间。要评估保护意义,再去查 IUCN 等级(已经显示在详情面板里)。 Best practice: treat the GBIF tier as "how rare it is to capture this shot" — it's optimized to surface the photos worth starring. For conservation significance, cross-reference the IUCN row in the same panel.
如果你在科研、媒体或博客中使用 慧眼选鸟 的罕见度结果,建议同时引用 GBIF 的数据快照 DOI
(每次构建时记录在 birdid/data/bird_reference.sqlite 的 gbif_rarity_index
表 metadata 行)以及 IUCN Red List。
If you use SuperPicky's rarity output in research, media, or blog posts, please
cite the GBIF snapshot DOI (recorded in the gbif_rarity_index table metadata row of
birdid/data/bird_reference.sqlite at build time) together with the IUCN Red
List.
如果你发现某只鸟的 tier 明显违反直觉(比如你家后院常客被标成「传奇」), 欢迎提 Issue 附上学名 + 你的地区 + GBIF 链接,我们会评估加进后续版本的手动 override 列表。 If you spot a tier that strongly disagrees with field intuition (e.g. a backyard regular flagged as "Legendary"), open an Issue with the scientific name, your region, and the GBIF taxon link — we'll review it for the manual override list.