BRENNEN

Industries / Government, education and community / Government / Parks and recreation

Nature in parks

Surveying and protecting the plants, animals and night sky inside parks, and managing how visitors affect them.

5

Milestones achieved, from 1995 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1995

    Wolves return to Yellowstone

    At least 136 wolves were killed in Yellowstone between 1914 and 1926, the year its last pack was killed. After a 1987 federal recovery plan and an environmental impact statement that drew more than 160,000 public comments, 31 gray wolves from western Canada were moved to the park in 1995 and 1996. From 1995 to 2003 wolves killed fewer livestock than expected: 256 sheep and 41 cattle. Park scientists write that most studies agree willow and aspen began to recover from browsing after wolves returned, but disagree on why, because "this was not a properly run experiment" and many factors were uncontrolled.

    LessonA reintroduction is not a controlled experiment, so claims about its knock-on effects need care.

    nps.govhome.nps.gov

    Credit

    US Fish and Wildlife Service; National Park Service

  2. 1996

    The first bioblitz

    The word "bioblitz" was coined in 1996 by a National Park Service employee organizing the first such event, at Kenilworth Park and Aquatic Gardens in Washington, D.C.: a 24-hour push by scientists and the public to find and record every plant and animal in one place. More than half of the 90 scientists who took part, from federal agencies, the Smithsonian Institution and local universities, had never been to Kenilworth before. The format has since been used around the world in parks and natural areas.

    nps.gov

    Credit

    National Park Service, with federal, Smithsonian and university scientists

  3. 2007

    A park for dark skies

    In spring 2007 the International Dark-Sky Association (IDA) named Natural Bridges National Monument, in Utah, the world's first International Dark Sky Park, citing its dark sky and the Park Service's treatment of night skies as a natural resource. The association's 2018 park guidelines require a lighting management plan, at least two-thirds of a park's outdoor fixtures to conform when it applies, 90% within five years, and a written commitment to 100% within ten.

    nps.govdarksky.org

    Credit

    International Dark-Sky Association; National Park Service

  4. 2016

    Cities race to count species

    Lila Higgins of the Natural History Museum of Los Angeles County and Alison Young and Rebecca Johnson of the California Academy of Sciences started the City Nature Challenge in 2016 as an eight-day contest between Los Angeles and San Francisco, in which more than 1,000 people made over 20,000 observations. In April 2026 the four-day event drew 106,354 observers in 754 cities in 61 countries, who recorded 3,001,825 observations of at least 76,422 species on iNaturalist, a shared app for nature records, by the UK National Biodiversity Network's count.

    citynaturechallenge.orgnbn.org.uk

    Credit

    Lila Higgins, Alison Young and Rebecca Johnson

  5. 2026

    Bird song sorted by AI

    Researchers led by Mary Clapp of the Institute for Bird Populations used recorders that the Park Service places each year at 185 sites in Olympic National Park for spotted owl monitoring, and ran the audio through BirdNET, a bird-sound model from Cornell University scientists, to track each species' singing through the season. They reported that BirdNET's "results needed significant verification from reviewers with expertise in bird sounds", and that the model had more trouble late in the season, when birds shift from songs to calls.

    LessonThe recorders let researchers "be in 180 places at once", which no field crew could do, but their output still needs experts to check it.

    nps.gov

    Credit

    Mary Clapp and colleagues, Institute for Bird Populations, with the National Park Service; BirdNET, Cornell University

Watch for

  1. One model for many animals

    On 7 August 2025 Google DeepMind released an updated Perch model, an AI model for sorting wildlife recordings, as an open model on Kaggle. It says the new model gives better bird species predictions than its predecessor without retraining, adapts better to new settings such as coral reefs, and was trained on nearly twice as much data, including mammals, amphibians and human-made noise. It says the first Perch, from 2023, was downloaded over 250,000 times and that its vector search library is now part of Cornell's BirdNET Analyzer.

    Achieved whenan independent study compares Perch and BirdNET on the same expert-labeled park recordings and reports which needs less expert checking.

    deepmind.google

    Claimed by

    Google DeepMind, August 2025

Open

  1. 01

    Light fixtures logged by camera

    A phone app, with a drone workflow for large sites, that logs every outdoor light fixture in a park with its location, a photo, whether it is fully shielded so that no light shines upward, and its color temperature, and produces the lighting inventory that dark-sky certification asks for.

    The brief
    Credit · unclaimed
  2. 02

    Late-season bird calls tested

    A public set of park recordings from late summer, when many birds have switched from songs to calls, labeled by experts, and scores for BirdNET and Perch on it, reported as precision and recall by species and as expert minutes needed per 100 confirmed detections.

    The brief
    Credit · unclaimed

Open 01 · the brief

Light fixtures logged by camera

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A phone app, with a drone workflow for large sites, that logs every outdoor light fixture in a park with its location, a photo, whether it is fully shielded so that no light shines upward, and its color temperature, and produces the lighting inventory that dark-sky certification asks for.

Why it is still open

The IDA's guidelines define a lighting inventory as "a formal audit of all artificial outdoor lighting in a park." Grand Teton National Park's, carried out on 15–19 June 2022 and reported by Bryan Boulanger, counted 2,834 fixtures, of which 34% (959) met the guidelines by being fully shielded with a color temperature of 3000 K or less. I found no tool that judges shielding and color temperature automatically.

darksky.orgnps.gov

Built with what exists

a phone camera with GPS; a handheld spectrometer, such as the Sekonic C-800, to measure color temperature; an image model fine-tuned to tell shielded from unshielded fixtures (see the Note "One GPU to fine-tune"); QGIS for the map.

  1. First step

    Photograph and log 50 outdoor fixtures in a city park by hand, measuring each with a spectrometer, as the first labeled set.

  2. Done looks like

    In one park or campus with at least 300 fixtures, the app's inventory compared fixture by fixture with a hand inventory, reporting the share of fixtures found, the share of shielding calls that agree, and the color temperature error in kelvin.

Open 02 · the brief

Late-season bird calls tested

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of park recordings from late summer, when many birds have switched from songs to calls, labeled by experts, and scores for BirdNET and Perch on it, reported as precision and recall by species and as expert minutes needed per 100 confirmed detections.

Why it is still open

In Olympic National Park the late-season shift from songs to calls "posed a challenge for the algorithm", and the study's lead author hopes to reach a point that needs less human verification. The largest public benchmark, WABAD, covers over 90,000 vocalizations of more than 1,100 species worldwide, and researchers using it still call expert validation "essential". I found no labeled set aimed at the late-season shift.

nps.govbou.org.uk

Built with what exists

AudioMoth recorders, small open-hardware sound loggers; BirdNET Analyzer; Perch, open on Kaggle; Raven Lite or a web labeling tool; Python.

  1. First step

    With a park's permission, record a week of late-July audio with two AudioMoths, run BirdNET, and time an expert checking 200 of its detections.

  2. Done looks like

    2,000 expert-labeled late-season clips from at least three parks, released with their labels, and both models scored on them, with a count of expert minutes per 100 confirmed detections.