# GeoLifeCLEF 2022 - LifeCLEF 2022 x FGVC9 — Data

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

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

<p>
  This year's challenge, GeoLifeCLEF 2022, uses a cleaned-up version of the data used for last year's challenge, GeoLifeCLEF 2021.<br>
  A changelog is available at the end of this page.<br>
</p>

<p>
  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 <a href=https://www.inaturalist.org/>iNaturalist</a> and <a href=https://plantnet.org/>Pl@ntNet</a> citizen science platforms.<br>

  There are 17K species in the dataset, 9K plant species and 8K animal species. <br>

  Each observation consists of a species name with the GPS coordinates where it was observed.<br>

  In addition, observations are paired with a set of covariates characterizing the landscape and environment around them.<br>

  Covariates include high-resolution remote sensing imagery, land cover data, and altitude, as well as traditional low-resolution climate and soil variables.<br>
</p>

<p>
  A <a href=https://www.kaggle.com/tlorieul/geolifeclef2022-data-loading-and-visualization>walkthrough notebook</a> describing the data structure and how to load it and visualize it is available in the <code>Code</code> tab.<br>
  A complete description of how the original GeoLifeCLEF dataset was built can be found in the <a href=https://arxiv.org/abs/2004.04192>associated paper</a>.<br>
  Data loading and visualization utilities are provided on the <a href=https://github.com/maximiliense/GLC>GitHub of the competition</a>.
</p>

<img src="https://raw.githubusercontent.com/maximiliense/GLC/master/images/observations_distribution_FR.png" alt="Observation distribution in France" width="350">
&nbsp;&nbsp;&nbsp;&nbsp;
<img src="https://raw.githubusercontent.com/maximiliense/GLC/master/images/observations_distribution_US.png" alt="Observation distribution in the US" width="450">
<br>

Here is an example of the patches corresponding to observation <code>10171444</code> on the Pic Saint-Loup mountain, France:
<img src="https://raw.githubusercontent.com/maximiliense/GLC/master/images/patches_sample_FR.jpg" alt="Patch data for observation 10561900" width="1000">

<p>
  In more detail, each observation is paired with the following covariates:
  <ul>
	<li>Remote sensing imagery: 256mx256m RGB-IR patches centered at each observation</li>
	<ul>
	  <li>Format: 256x256 JPEG images, a color JPEG file for RGB data and a grayscale one for near-infrared</li>
	  <li>Resolution: 1 meter per pixel</li>
	  <li>Source: NAIP for US and IGN for France</li>
	</ul>
	<li>Land cover data: 256mx256m patches centered at each observation</li>
	<ul>
	  <li>Format: 256x256 TIFF files with Deflate compression</li>
	  <li>Resolution: 1 meter per pixel</li>
	  <li>Source: NLCD for US and Cesbio for France</li>
	</ul>
	<li>Altitude data: 256mx256m patches centered at each observation</li>
	<ul>
	  <li>Format: 256x256 TIFF files with Deflate compression</li>
	  <li>Resolution: 1 meter per pixel</li>
	  <li>Source: SRTMGL1 for US and France</li>
	</ul>
	<li>Bioclimatic data: 19 low-resolution rasters</li>
	<ul>
	  <li>Format: GeoTIFF files with LZW compression</li>
	  <li>Resolution: 30 arcsec per pixel (~ 1 kilometer per pixel)</li>
	  <li>Source: WorldClim for US and France</li>
	</ul>
	<li>Pedologic data: 8 low-resolution rasters</li>
	<ul>
	  <li>Format: GeoTIFF files with LZW compression</li>
	  <li>Resolution: 250 meters per pixel</li>
	  <li>Source: SoilGrids for US and France</li>
	</ul>
  </ul>
</p>

 <hr>

<b>Changelog from GeoLifeCLEF 2021 dataset:</b>
<ul>
  <li>completed species metadata with genus, family and kingdom information from GBIF</li>
  <li>removed observations of species from kingdom different from Plantae and Animalia (29240 observations, 2072 species)</li>
  <li>kept only one observation when it is duplicated - same latitude, longitude and species (110556 observations removed)</li>
  <li>removed all observations at exactly the same location - same latitude and longitude, different species (103887 observations removed)</li>
  <li>removed observations from species only present in test set (208 observations, 188 species)</li>
  <li>removed species with strictly less than 3 observations in the train set (13336 observations, 9913 species)</li>
  <li>species_id updated (not aligned with GeoLifeCLEF21), in the end, 17037 species are retained</li>
  <li>updated all altitude patches</li>
  <ul>
    <li>re-extraction uses bi-cubic interpolation instead of bi-linear interpolation</li>
    <li>an issue which resulted in artifacts on a few altitude patches from past years was also fixed</li>
  <ul>
</ul>