SunShield AI

Data-driven solar shade placement for campuses. Combines heat sensors, solar potential, and AI to recommend optimal sites in seconds.

Comment

SunShield AI — Your Campus Solar Planning Co-Pilot

🌟 Inspiration

Campus heat is dangerous—students experience heat exhaustion, reduced productivity, and unsafe conditions during heat waves. Urban planners currently spend 2-3 weeks manually analyzing where to install solar shade structures, hiring consultants for $5,000-$10,000 feasibility studies.

No tools integrate heat risk data, solar generation potential, and budget optimization in one platform. We wanted to give planners a data-driven decision co-pilot that transforms weeks of analysis into minutes of clarity while generating export-ready files for their existing GIS tools.

The dual-benefit approach—combining heat protection with renewable energy generation—creates a compelling value proposition for both student health and campus sustainability goals.

Global Impact Vision

While our proof-of-concept focuses on UCSD campus, SunShield AI's methodology is universally applicable to any geographic region with heat exposure challenges. The system is designed for immediate deployment in high-heat urban environments worldwide:

  • Pilgrimage Routes (India, Saudi Arabia): Multi-kilometer walking paths where heat exposure is unavoidable—such as Kumbh Mela routes, Hajj pathways, or temple circuits. Our optimization identifies critical shade intervention points along pedestrian corridors where solar infrastructure provides both cooling relief and renewable energy for lighting and water stations.

  • Urban Heat Islands (Global South Cities): Rapidly urbanizing regions experiencing extreme heat without adequate cooling infrastructure. Our data-driven approach works with any heat sensor data source and adapts to local solar irradiance patterns.

  • Transit Corridors (Bus stops, train platforms): High-dwell-time locations where vulnerable populations (elderly, children, outdoor workers) face extended heat exposure during daily commutes.

The platform's modular data pipeline accepts heat measurements from any source (satellite thermal imaging, IoT sensors, mobile weather stations), integrates with regional solar databases (NREL for US, similar APIs globally), and optimizes for local cost structures and infrastructure constraints. One methodology, infinite applications.

💡 What it does

SunShield AI analyzes real Scripps Institution heat sensor data and NREL solar radiation to recommend optimal solar shade installation sites on campus.

Dual-Benefit Scoring Algorithm

Our core innovation evaluates each location using a weighted multi-criteria score:

$$\text{Dual-Benefit Score} = (0.40 \times \text{Heat Risk}) + (0.30 \times \text{Solar Potential}) + (0.20 \times \text{Public Impact}) + (0.10 \times \text{Feasibility})$$

Where each component is scored 0-100:

Component Formulas:

Heat Risk (40% weight): $$\text{Heat Risk} = \frac{\text{temp_delta}}{5.0} \times 100$$

Solar Potential (30% weight): $$\text{Solar Score} = \frac{\text{sun_hours_daily}}{9.0} \times 100$$

Public Impact (20% weight): $$\text{Impact Score} = \frac{\text{pedestrians_per_day}}{500} \times 100$$

Feasibility (10% weight): $$\text{Feasibility} = \frac{45000 - \text{cost}}{20000} \times 80 + \text{grid_bonus}$$

Key Features

For any budget, we show which sites to build first and why. The system:

  • Processes 1,247 GPS sensor readings → clusters into 25 candidate sites using DBSCAN
  • Enriches with government solar data via NREL API
  • Runs greedy optimization to maximize impact per dollar
  • Exports as CSV or GeoJSON for ArcGIS, QGIS, Google Earth integration
  • Generates AI-powered grant justification text via Google Gemini
  • Validated against 6,192 real Sullivan Solar installations (1,794 in San Diego)

Result: Planners get optimized recommendations in 30 seconds vs. 2-3 weeks of manual analysis.

🛠️ How we built it

Data Pipeline

Step 1: Spatial Clustering

from sklearn.cluster import DBSCAN

clustering = DBSCAN(eps=0.0005, min_samples=3)
clusters = clustering.fit_predict(gps_coords)
# Result: 1,247 readings → 25 distinct hot spot sites

Step 2: Multi-Source Data Integration

  • Scripps Heat Data: Real campus temperature measurements (1,247 GPS readings)
  • NREL Solar API: Government solar radiation database (US Dept of Energy)
  • ZenPower Validation: 6,192 Sullivan Solar installation records (2012-2023)
  • Campus Infrastructure: Site classification, cost estimates, grid access

Step 3: Budget Optimization

Greedy algorithm maximizes cumulative Dual-Benefit Score:

$$\text{Maximize: } \sum_{j \in S} \text{Score}_j$$

Subject to constraint: $$\sum_{j \in S} \text{cost}_j \leq \text{Budget}, \quad S \subseteq {1, 2, ..., 25}$$

Performance achieved: 98.2% average budget utilization

Technology Stack

  • Backend: Python 3.11, Pandas, Scikit-learn, NumPy
  • Analytics: Custom multi-criteria scoring + greedy optimization
  • Frontend: Streamlit 1.32 (interactive dashboard), Plotly 5.19 (maps)
  • AI Integration: Google Gemini 2.0 Flash API (grant text generation)
  • APIs: NREL PVWatts (solar data), Scripps sensors (heat data)
  • Deployment: GitHub + Streamlit Community Cloud
  • Data Export: CSV, GeoJSON formats for GIS compatibility

Scalability Architecture

The system is location-agnostic by design:

  1. Flexible Data Inputs: Accepts any coordinate-based heat measurements (satellite, IoT, mobile sensors)
  2. Modular Scoring: Component weights adjustable for regional priorities (e.g., increase Public Impact weight for pilgrimage routes)
  3. API-Agnostic: Works with NREL (US), Solargis (Europe), PVGIS (Global), or custom solar databases
  4. Cost Adaptation: Installation cost formulas parameterized by regional labor/material costs
  5. Export Standards: GeoJSON follows international GIS standards—compatible with global planning tools

Deployment Example for India Pilgrimage Routes:

  • Input: Mobile heat sensor data collected along Kumbh Mela walking paths
  • Solar Data: PVGIS API for Indian solar irradiance patterns
  • Public Impact: Pedestrian traffic from festival attendance data
  • Output: Ranked shade station locations optimized for maximum heat relief per rupee spent

🚧 Challenges we ran into

API Rate Limiting: NREL API limited to 1 request/second. Solution: Implemented time.sleep(1) throttling between 25 site calls (~25 seconds total).

Unrealistic Solar Scores: Initial scores showed flat distribution (all ~67). Root cause: No site-specific factors. Solution: Applied microclimate modifiers:

  • Parking structures: +10-20% (elevated, minimal shade)
  • Tree-lined walkways: -5 to +5% (variable canopy)
  • Building-adjacent areas: -10 to 0% (shadow effects)
  • Result: 26% variance increase → realistic 62-78 score range

Missing Pedestrian Data: No actual traffic counters available. Solution: Evidence-based estimation from site classification (parking: 400-500/day, plazas: 250-350/day, validated against similar campus studies).

Streamlit UI Issues: Sidebar controls existed in HTML but rendered invisible. After 30 minutes debugging CSS, we pivoted to main-page controls—saving time and improving mobile UX.

Scope Management: Requested features totaled 9.5 hours; only 4 hours available. Solution: Ruthlessly cut ML models, real-time sensors, complex what-if scenarios to ship working MVP. Lesson: Shipping beats perfection.

Git Submodule Conflicts: ZenPower dataset integration created nested repository issues. Solution: Careful path management and .gitignore configuration.

🏆 Accomplishments that we're proud of

🌍 Universal Methodology: Built location-agnostic system applicable from UCSD campuses to Indian pilgrimage routes to Middle Eastern transit corridors—one algorithm, global impact potential.

📊 Multi-Source Integration: Combined 3 real-world datasets (Scripps research + NREL government + ZenPower industry benchmarks spanning 6,192 installations).

⚡ Dramatic Efficiency: Reduced planning time from 2-3 weeks → 30 seconds (99.7% reduction).

🗺️ Professional Tools: Built GIS exports (CSV, GeoJSON) compatible with ArcGIS, QGIS, Google Earth—enabling seamless integration with existing urban planning workflows globally.

🤖 AI-Powered Justification: Google Gemini generates grant proposal text, saving planners hours of writing while maintaining technical accuracy.

✅ Industry Validation: Validated cost model against 6,192 real Sullivan Solar installations (1,794 in San Diego specifically)—proving methodology reflects actual market economics.

💰 Optimization Performance: Achieved 98.2% average budget utilization through greedy algorithm—maximizing impact per dollar spent.

🔄 Complete ETL Pipeline: Raw GPS readings → DBSCAN clustering → API enrichment → Multi-criteria scoring → Constrained optimization → Professional exports.

📈 Realistic Variance: Fixed flat scoring issue—achieved 26% solar score variance (62-78 range) reflecting real microclimate differences.

📚 What we learned

Data is Messy: Cleaning and preprocessing consumed 60% of development time. Real-world data requires extensive validation, outlier handling, format standardization, and missing value imputation.

Domain Knowledge > Technical Sophistication: Understanding urban planners' workflows (GIS tools, budget presentations, stakeholder justifications) shaped our UX more than algorithmic complexity. Lesson: Talk to users early.

Explainability > Accuracy: Planners need to justify decisions to budget committees. A transparent 4-component score they can explain beats a black-box ML model with 2% higher accuracy. Lesson: Interpretability enables adoption.

Integration is Adoption: GeoJSON export for ArcGIS/QGIS was critical. Building a standalone tool that ignores existing workflows = zero real-world usage. Lesson: Fit into existing ecosystems, don't replace them.

Greedy is Good (Sometimes): For constrained optimization with 25 sites, greedy selection achieved 98% budget utilization in <1 second. Complex integer programming would add minutes for <2% gain. Lesson: Match algorithm complexity to problem scale.

API Constraints Are Normal: Rate limits, missing data, format inconsistencies are standard in production systems. Building fallbacks and estimation methods is part of professional development—not a failure.

Global Scalability Requires Modularity: Designing for UCSD while thinking "how would this work in Mumbai or Mecca?" forced cleaner abstractions and parameterized assumptions—making the system genuinely reusable.

Ruthless Scope Control: Cutting 5.5 hours of features to ship in 4 hours taught us: Working > Perfect. Hackathon success = shipped product, not feature completeness.

Gemini for Domain Text: Google Gemini 2.0 Flash excels at generating formal, domain-specific language (grant proposals, technical justifications) when given structured data inputs.

🚀 What's next for SunShield AI

Immediate Validation (Month 1)

  • UCSD Pilot: Partner with facilities team to deploy recommendations at 2-3 test sites
  • Pedestrian Counters: Install IoT sensors at top 10 locations for 2-week actual traffic validation
  • Solar Shade Analysis: Commission detailed shade coverage studies to refine solar modifiers
  • Cost Validation: Obtain contractor quotes to verify $27k-$45k installation range

Global Expansion (Months 2-6)

  • India Pilgrimage Routes: Partner with temple management authorities to deploy along Kumbh Mela paths, Tirupati hills, Vaishno Devi trek—optimizing shade shelter placement for millions of annual pilgrims facing extreme heat exposure during religious journeys.
  • Middle East Transit: Adapt for Hajj pedestrian corridors and Gulf city bus stops where heat exceeds 45°C (113°F) regularly.
  • Southeast Asia Cities: Deploy in Jakarta, Manila, Bangkok transit systems experiencing rapid urbanization and rising heat island effects.
  • Sub-Saharan Africa: Partner with NGOs to optimize solar shade infrastructure in markets, water collection points, and school playgrounds.

Technical Enhancements (Months 3-12)

  • Satellite Data Integration: Accept thermal imaging from Landsat/Sentinel for any global location—eliminating need for local sensors.
  • Multi-Language Support: Interface translation for Hindi, Arabic, Swahili, Bahasa—enabling direct use by local planners.
  • Mobile Field App: Lightweight Android/iOS app for site validation and real-time data collection.
  • ML Prediction Models: Train on actual pedestrian counter data for improved traffic estimation.
  • Real-Time Weather: Integrate seasonal optimization based on monsoons, dry seasons, regional climate patterns.
  • Historical Trend Analysis: 10+ year heat data patterns to predict future hot spot migration.

Product Extensions (Year 2)

  • Solar Trash Compactors: Apply same optimization to waste infrastructure
  • EV Charging Stations: Combine solar + charging infrastructure planning
  • Bus Shelter Upgrades: Integrate with transit agency planning tools
  • Emergency Cooling Stations: Hurricane/heatwave response infrastructure optimization
  • Agricultural Worker Shade: Optimize field-edge shade structures for farmworkers

Enterprise & Partnerships

  • ZenPower Sales API: Enable solar companies to generate site recommendations for prospect universities/municipalities—data-driven customer acquisition tool.
  • Multi-Campus Dashboard: Expand to all 10 UC system universities, then CSU system (23 campuses).
  • Government Partnerships: Pilot with Indian Ministry of Tourism (pilgrimage route safety), UAE Ministry of Infrastructure (transit cooling), WHO (heat-health intervention planning).
  • NGO Collaboration: Open-source core algorithm for humanitarian heat relief projects in climate-vulnerable regions.
  • Enterprise SaaS: Role-based access, audit trails, multi-year scenario planning, contractor integration, procurement workflows.

Research & Validation

  • Academic Publication: Document methodology in urban planning / public health journals to establish evidence base.
  • Peer Validation: Collaborate with urban heat island researchers to validate component weights across climates.
  • Impact Measurement: Longitudinal studies measuring actual heat illness reduction, energy ROI, and cost-effectiveness at deployed sites.

Vision: Transform SunShield AI from a campus planning tool into a global heat relief infrastructure platform—protecting vulnerable populations worldwide while accelerating renewable energy deployment. One algorithm solving urban heat from California to Kumbh Mela.

🌐 Global Applicability Summary

Region Use Case Data Source Impact
US Campuses Student heat protection Scripps sensors Validated proof-of-concept
India Pilgrimages Multi-km walking routes Mobile/satellite thermal Millions of pilgrims protected annually
Middle East Transit Bus stops, Hajj paths IoT weather stations Extreme heat (>45°C) mitigation
SE Asia Cities Urban heat islands Satellite thermal imaging Rapidly urbanizing population protection
Sub-Saharan Africa Markets, schools, water points NGO sensor networks Climate adaptation for vulnerable communities

Common Thread: Same Dual-Benefit Score algorithm, localized data inputs, culturally appropriate implementation—universal methodology, regional execution.

Built With

Share this project:

Updates