Wildlife Intelligence Dashboard

Historical wildlife observation patterns across U.S. national parks.

WildWhere AI combines real public GBIF, NPS, NOAA, and visitation datasets into a cleaned SQLite warehouse, analytics layer, machine-learning model, and Streamlit-ready dashboard. This deployed page surfaces the project analytics and prediction workflow with an explicit park-species eligibility gate.

Prediction safety gate enabled

Analytics Overview

1,555Clean reported observation records

5National parks covered

10Target wildlife species

0.876ML test accuracy after eligibility filtering

Reported Sightings By Park

Park	Records
Yellowstone	570
Grand Teton	367
Glacier	271
Olympic	187
Mount Rainier	160

Top Reported Species

Species	Records
Elk	252
Black Bear	203
Mountain Goat	192
Grizzly Bear	157
Bald Eagle	150

Seasonal Pattern

Season	Records
Summer	700
Spring	396
Winter	278
Fall	181

Estimated Reported-Observation Likelihood

The project uses a two-stage system. First, selected park/species combinations are checked against official park-species reference data. Only documented combinations are evaluated by the model for historical reported-observation likelihood.

Valid Prediction Grid Preview

The deployed preview shows precomputed rows from data/processed/valid_prediction_grid.csv. Unsupported combinations are excluded before model scoring.

Park	Species	Season	Time Window	Output	Records
Yellowstone National Park	Bison	summer	unknown	high	47
Grand Teton National Park	Moose	summer	morning	high	28
Mount Rainier National Park	Mountain Goat	summer	morning	high	52
Grand Teton National Park	Mountain Goat	summer	unknown	Insufficient data	3
Olympic National Park	Bald Eagle	summer	unknown	high	54
Glacier National Park	Mountain Goat	winter	unknown	low	37

Example blocked combination: Bison in Mount Rainier National Park returns Not supported with no model score because the species is not documented for the selected park in the project reference data.

Step 3.5 SerpApi Enrichment

WildWhere AI now includes an optional SerpApi enrichment layer for public search context: named viewing areas, seasonal advice, time-of-day mentions, and source/domain signals. This is supplemental context, not scientific ground truth.

350Search queries planned across park/species/template combinations

0Live SerpApi results collected in this deployment because no API key was provided

1,555Rows in wildlife_features_enriched.csv with zero/unknown context defaults

60Tests passing after prediction-safety patch

Search-context enrichment is optional and depends on configured API credentials. Search-derived context should support interpretation, not replace GBIF, NPSpecies, NOAA, or NPS visitation data.

SQL + Warehouse Layer

The SQLite warehouse includes cleaned tables for parks, species, sightings, weather, visitation, and joined wildlife features.

6SQLite tables

10Portfolio SQL queries

9Processed analytics CSVs

SELECT park_name, common_name, COUNT(*) AS sighting_count
FROM wildlife_features
GROUP BY park_name, common_name
ORDER BY sighting_count DESC;

Data Transparency

17.2%Weather join success

47.5%Visitation join success

1,050Bounding-box-only records

505Coordinate-uncertain records

GBIF records are reported observations collected near parks with bounding boxes. They are not confirmed wildlife presence, real-time tracking, complete wildlife distributions, or true population density.

Project Entry

The deployed Vercel page is the public portfolio preview. The repository also includes a full Streamlit application with entry point app/main.py.

https://wildwhere-ai.vercel.app