SafeAscent

Climbing safety predictions powered by historical accident analysis and real-time weather data.

Last Updated: February 2026

---

Project Overview

SafeAscent is a web application that predicts climbing route safety by analyzing ~5,000 historical accidents weighted by:

• Spatial proximity - Gaussian decay by distance (route-type-specific bandwidths)
• Weather similarity - 7-day pattern matching with cubic weighting
• Temporal relevance - Year-scale decay with seasonal boosting
• Route type - Asymmetric cross-type influence matrix

The algorithm provides daily risk scores (0-100) with confidence metrics, helping climbers make informed decisions about route conditions.

For detailed algorithm documentation, see ALGORITHM_DESIGN.md.

---

Current Status

Component	Status	Notes
Backend API	✅ Production-ready	FastAPI + PostgreSQL
Safety Algorithm	✅ Complete	7 modular services, 50/50 tests passing
Frontend Map	✅ Complete	Two-view system with stratified heatmaps
Data Pipeline	✅ Complete	~168K routes, ~6.9K accidents, ~25K weather records

---

Tech Stack

Backend

• FastAPI - Async Python API framework
• PostgreSQL - Primary database with PostGIS
• Redis - Caching (2-day TTL for nightly safety scores, 6-hour for weather forecasts)
• Open-Meteo - Weather API (free, no key required)

Frontend

• React 18 + Vite - UI framework and build tool
• Material-UI - Component library (Material Design 3)
• Mapbox GL JS - Interactive 3D terrain maps
• React-Map-GL - React wrapper for Mapbox

Data Pipeline

• Pandas - Data processing
• Google Maps API - Geocoding (primary)
• Gemini 2.0 Flash - AAC extraction + geocoding fallback

---

Frontend Features

Two-View Map System

View	Purpose	Implementation
Cluster View	Navigation	Mapbox native clustering, color-coded by avg risk
Risk Coverage	Safety Analysis	5 stratified heatmap layers

Key Features

• Dark mode UI with Material Design 3
• Search by route/mountain name
• Date picker for 7-day forecast
• Season filter: All / Summer (rock) / Winter (ice/mixed)
• Season-specific map styles: Warm outdoors / Cool winter theme
• Hover tooltips with route details
• Tight grid clustering for overlapping routes
• 8-tab route analytics dashboard
• Custom climbing-themed favicon
• Boulder routes excluded (different risk profile)

---

Backend Architecture

Algorithm Services (backend/app/services/)

• algorithm_config.py - All tunable parameters
• spatial_weighting.py - Gaussian decay by distance
• temporal_weighting.py - Exponential decay + seasonal boost
• weather_similarity.py - 7-day pattern matching
• route_type_weighting.py - Asymmetric type matrix
• severity_weighting.py - Fatality/injury weighting
• confidence_scoring.py - Confidence calculation
• safety_algorithm.py - Main orchestrator

API Endpoints

Core

• POST /api/v1/predict - Get safety prediction
• GET /api/v1/routes - List routes with filters
• GET /api/v1/routes/map - Routes optimized for map display
• GET /api/v1/mountains - List mountains

Route Analytics

• POST /api/v1/mp-routes/{id}/safety - Safety score for date (bypass_cache=true available for forced live recompute)
• GET /api/v1/mp-routes/{id}/forecast - 7-day forecast
• GET /api/v1/mp-routes/{id}/accidents - Accident history
• GET /api/v1/mp-routes/{id}/risk-breakdown - Factor analysis
• GET /api/v1/mp-routes/{id}/time-of-day - Hourly analysis
• GET /api/v1/mp-routes/{id}/historical-trends - Historical risk trend series
• GET /api/v1/mp-routes/{id}/ascent-analytics - Monthly ascent/accident rates

Scoring Strategy

• Map rendering uses bulk precomputed cached scores (/mp-routes/map-with-safety) for scale.
• Route analytics/details fetches can request live recomputed score (/mp-routes/{id}/safety?bypass_cache=true) for higher-fidelity route-level display.

---

Database Schema

For detailed database documentation, see data/DATABASE_STRUCTURE.md.

Core Tables

Table	Records	Description
accidents	~6,900	Combined from AAC, Avalanche.org, NPS
weather_patterns	~25,000	7-day windows for each accident
mp_routes	~168,000	Mountain Project climbing routes
mp_locations	~45,000	Location hierarchy (areas → crags)
historical_predictions	Growing	Daily safety score history

---

Data Sources

Source	Records	Coverage
AAC (American Alpine Club)	2,770	1990-2019
Avalanche.org	1,372	1997-2026
NPS (National Park Service)	848	Various
Mountain Project	~168,000 routes	Complete

---

Quick Start

Backend

cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload

Frontend

cd frontend
npm install
# Add VITE_MAPBOX_TOKEN to .env
npm run dev

---

Documentation

• ALGORITHM_DESIGN.md - Detailed algorithm decisions
• DEPLOYMENT.md - Infrastructure architecture
• data/DATABASE_STRUCTURE.md - Database schema
• data/README.md - Data sources and collection

---

License

MIT