Exploratory: generated automatically from the UmaTerra platforms' data; not independently validated or peer reviewed.
A national park in the United States, one in Canada and one in Brazil, of similar size, read with the same tools: the official outline, tree-cover loss, land use and what GBIF holds on birds, plants and mammals. It is neither a ranking nor a proof: the story shows where the tools work the same in all three countries and where the parks differ, in the land around them, the kind of observation and the quantity of records, and why a count of records measures observation effort, not richness.
Loading the numbers…
measured by satellite, the official outline or a GBIF count. estimated a reading that depends on a choice. missing a part that could not be obtained.
Great Smoky Mountains (United States), Gros Morne (Canada) and Iguaçu National Park (Brazil) are national parks, all in IUCN category II. The three outlines come from the same world database, the WDPA, and the area was recomputed from the polygon. The three appear side by side, at the same scale, and the “Highlight” buttons pick out one of them in every step.
Four things run with no change in all three countries: the outline (WDPA), the satellite image (Sentinel-2), tree-cover loss (Hansen) and the count of records (GBIF). One changes from country to country by necessity: the land-use map, which each country produces with its own method and legend. The table says which.
GBIF adds up what museums, herbaria, observers and cameras have published. Each row is what exists inside each park's outline for one group: records, species and records per km². A record is someone's encounter with a species at a place and a date; many records say, first of all, where people were looking.
Each square is a 0.05° cell (about 5 km), coloured by the density of records of birds, plants and mammals inside the park. The literature shows that records cluster near roads and towns, and far from them there are rarely visited areas, even when the park is the same.
GBIF separates collection specimens, human observations and machine observations (cameras and recorders). All three count but say different things: a specimen has a name and a place that can be checked in a collection; a human observation depends on who saw it and who identified it.
Behind every number there is a dataset, with a publisher and its own citation on GBIF. The largest ones in each park show what kind of science produces them: bird-observation platforms, naturalists, or museum collections and institutional inventories.
Records do not grow evenly through time: citizen-observation platforms pushed the number of records up in recent decades, and not every park followed the same pace. The chart shows records per km² per year, for the three groups together.
Hansen counts all tree-cover loss since 2001, from any cause: fire, storm, pests, felling. Applied to the inside of each park and to a 10 km ring around it, it shows where the land changed, without saying why. The map colours mark the period in which each pixel lost its cover.
Each country has its own annual land-cover map, with its own method and legend, and the three are not added together. In the United States it is the Annual NLCD, 1985 and 2024; in Canada, the NTEMS VLCE2, 1985 and 2022, chosen for Gros Morne because it has forest, shrub, wetland and water classes, the ones that count in a park of forest, bog and coast, with the limits that it ends in 2022 and has no built-up or cropland class; in Brazil, MapBiomas Collection 11, 1985 and 2024, with classes of its own, such as mosaic of uses and forestry, that the other two maps lack. So the three parks’ tables should not be read side by side. Drag the bar: it moves the three panels together.
Without the official lists, we can read what GBIF records inside each outline: how many species of each group and how many appear in only one park. It is a list of what was recorded, not of what exists, built from GBIF’s own species count and the names from its species API.
The most recorded species in each park say what is easy to see and identify; those found in only one park show what is local to the outline. The names and common names come from GBIF’s species API.
GBIF attaches to each species its IUCN Red List category, when the species has been assessed: it is the one threat yardstick the three countries share. Here species (not records) are counted in the CR, EN, VU and NT categories inside each outline. It is not an official park list: it is GBIF’s name match with the IUCN.
Some categories look odd on a list of a park’s living species: extinct, extinct in the wild, conservation dependent and data deficient. Instead of dropping them, we looked at which species they are and whether the names are accepted in GBIF’s catalogue, to say what is known and what is not.
Totals published by each agency, as context and not as a comparison: each source has a different scope, method and date. Great Smoky comes from NPS pages and an inventory of all taxa; Iguaçu from an ICMBio bird guide and a scientific survey of mammals; for Gros Morne, the Parks Canada and IUCN pages carry no totals for birds or plants. Ecology expects more species in tropical forests than in temperate and boreal ones (the latitudinal diversity gradient); this table does not test that, because the numbers have different methods and there are empty cells. An expected difference between biomes, not a result of this story.
The idea was to compare GBIF with each park's official species list, and with the national threatened-species lists: how many listed species have at least one record, and how recent and precise. In this round those lists could not be obtained, species by species, in machine-readable form (what could be read were the totals of the previous step); the table says what was tried and what stopped it. The per-species lists in the previous steps come from GBIF and are not official. Nothing was made up to fill the gap.
The Iguaçu polygon falls entirely in IBGE's Atlantic Forest biome. The story of the Mello Leitão museum, in Santa Teresa (Espírito Santo), reads the same biome with the same tools from a very different point of view: four small places instead of one park, and what GBIF holds for each, with the location and precision of the records.
Written with Claude Sonnet 5.5, by Anthropic, from the cited papers and the numbers Earth Engine and GBIF returned: no fact or number comes from the model. Versão em português.