Wildfire Research Initiative

California Wildfire Prediction

TAM AIR Club × CAL FIRE × UCSF × UCI

Our research initiative aims to build a pixel-wise wildfire risk prediction system for California at 800m × 800m resolution using machine learning. By analyzing over 147 years of fire perimeter data, we're developing tools to help predict and prevent catastrophic wildfires.

This collaboration brings together high school researchers, university academics, and California's fire management professionals to tackle one of the state's most pressing environmental challenges.

Dataset Overview

  • 22,000+ fire perimeters analyzed
  • 147 years of historical data (1878-2025)
  • 97%+ data completeness for key fields (post-1993)
  • GPS-based high-quality perimeter data

Key Findings

Critical insights from our exploratory data analysis

4.2 Million

Acres burned in 2020 alone - California's worst fire year on record

84%

Of burned acreage occurs during June-September fire season

Top 1%

Of fires cause 58% of all damage - extreme concentration of risk

38 Mega-fires

Fires exceeding 100,000 acres recorded since 1993

Research Visualizations

Key findings from our comprehensive exploratory data analysis

Executive Dashboard - 10-panel overview of California wildfire data

Executive Dashboard

Comprehensive 10-panel overview showing fire counts, burned acreage trends, seasonal patterns, size distributions, causes, and responding agencies over the past 30+ years.

Fire Clock - Seasonal fire patterns visualization

Fire Clock

Polar visualization showing seasonal fire patterns. The outer ring displays acres burned by month, while the inner ring shows fire counts - clearly demonstrating the June-September peak.

Cumulative Fire Risk Map of California

Cumulative Fire Risk Map

Burn frequency heatmap at 1000m resolution showing high-risk zones across California. This visualization helps identify areas that have experienced repeated fire activity.

Trend Analysis showing wildfire acceleration

Trend Analysis

Statistical visualization from 1950-present with fitted regression line providing clear evidence of wildfire acceleration, particularly pronounced after 2000.

Interactive Mega-Fires Map

Explore all 38 mega-fires (>100,000 acres) recorded since 1993

Interactive Folium Map

Click below to explore the full interactive map with all mega-fire perimeters, details, and filtering options.

Open Interactive Map

Note: Large file (~35MB). Best viewed on desktop.

Research Documents

Download our comprehensive analysis report and explore the methodology

Full Analysis Report

Complete exploratory data analysis with 19 high-resolution figures, statistical findings, and methodology documentation.

Download PDF (3.6 MB)

Jupyter Notebooks

View our analysis notebooks with full code, methodology explanations, and reproducible results using Python and GeoPandas.

View on GitHub

Project Roadmap

Our multi-phase approach to building a wildfire prediction model

Phase 1.5: Exploratory Data Analysis

COMPLETED

Comprehensive analysis of 22,000+ fire perimeters. Created 19 publication-quality visualizations. Identified key patterns and data quality thresholds.

Phase 2: Grid Creation

IN PROGRESS

Generate 800m × 800m pixel grid covering California. Rasterize fire perimeters to binary grid cells. Create temporal aggregation (monthly/seasonal).

Phase 3: Data Integration

PLANNED

Integrate PRISM climate data (temperature, precipitation, VPD), DEM for topography features, and vegetation/fuel data. Align all datasets to common grid.

Phase 4: ML Model Development

FUTURE

Develop CNN/LSTM architecture for temporal prediction. Train on high-quality post-1993 data. Address class imbalance challenge (99% of pixels never burn).

Research Partners

This research is made possible through collaboration

TAM AIR Club

Student-led research team from Tamalpais High School

CAL FIRE

California Department of Forestry and Fire Protection

UCSF

University of California, San Francisco - Academic Partner

UCI

University of California, Irvine - Research Collaboration