# GeoLifeCLEF 2022 - LifeCLEF 2022 x FGVC9 — Data

来源：https://www.kaggle.com/competitions/geolifeclef-2022-lifeclef-2022-fgvc9/data

采集时间：2026-10-24T01:21:12Z

This year's challenge, GeoLifeCLEF 2022, uses a cleaned-up version of the data used for last year's challenge, GeoLifeCLEF 2021.
A changelog is available at the end of this page.
The challenge relies on a collection of 1.6 million of geo-localized observations of plants and animals in the US and France, coming from the
[iNaturalist](https://www.inaturalist.org/)
and
[Pl@ntNet](https://plantnet.org/)
citizen science platforms.
There are 17K species in the dataset, 9K plant species and 8K animal species.
Each observation consists of a species name with the GPS coordinates where it was observed.
In addition, observations are paired with a set of covariates characterizing the landscape and environment around them.
Covariates include high-resolution remote sensing imagery, land cover data, and altitude, as well as traditional low-resolution climate and soil variables.
A
[walkthrough notebook](https://www.kaggle.com/tlorieul/geolifeclef2022-data-loading-and-visualization)
describing the data structure and how to load it and visualize it is available in the
Code
tab.
A complete description of how the original GeoLifeCLEF dataset was built can be found in the
[associated paper](https://arxiv.org/abs/2004.04192)
.
Data loading and visualization utilities are provided on the
[GitHub of the competition](https://github.com/maximiliense/GLC)
.
![Observation distribution in France](https://raw.githubusercontent.com/maximiliense/GLC/master/images/observations_distribution_FR.png)
![Observation distribution in the US](https://raw.githubusercontent.com/maximiliense/GLC/master/images/observations_distribution_US.png)
Here is an example of the patches corresponding to observation
10171444
on the Pic Saint-Loup mountain, France:
![Patch data for observation 10561900](https://raw.githubusercontent.com/maximiliense/GLC/master/images/patches_sample_FR.jpg)
In more detail, each observation is paired with the following covariates:
Remote sensing imagery: 256mx256m RGB-IR patches centered at each observation
Format: 256x256 JPEG images, a color JPEG file for RGB data and a grayscale one for near-infrared
Resolution: 1 meter per pixel
Source: NAIP for US and IGN for France
Land cover data: 256mx256m patches centered at each observation
Format: 256x256 TIFF files with Deflate compression
Resolution: 1 meter per pixel
Source: NLCD for US and Cesbio for France
Altitude data: 256mx256m patches centered at each observation
Format: 256x256 TIFF files with Deflate compression
Resolution: 1 meter per pixel
Source: SRTMGL1 for US and France
Bioclimatic data: 19 low-resolution rasters
Format: GeoTIFF files with LZW compression
Resolution: 30 arcsec per pixel (~ 1 kilometer per pixel)
Source: WorldClim for US and France
Pedologic data: 8 low-resolution rasters
Format: GeoTIFF files with LZW compression
Resolution: 250 meters per pixel
Source: SoilGrids for US and France
Changelog from GeoLifeCLEF 2021 dataset:
completed species metadata with genus, family and kingdom information from GBIF
removed observations of species from kingdom different from Plantae and Animalia (29240 observations, 2072 species)
kept only one observation when it is duplicated - same latitude, longitude and species (110556 observations removed)
removed all observations at exactly the same location - same latitude and longitude, different species (103887 observations removed)
removed observations from species only present in test set (208 observations, 188 species)
removed species with strictly less than 3 observations in the train set (13336 observations, 9913 species)
species_id updated (not aligned with GeoLifeCLEF21), in the end, 17037 species are retained
updated all altitude patches
re-extraction uses bi-cubic interpolation instead of bi-linear interpolation
an issue which resulted in artifacts on a few altitude patches from past years was also fixed