# **Comprehensive Summary: Coastal Erosion Risk Prediction Using AI**
*Presented by Lily Pinault, Data Scientist & Founder of AquaAdapt*
---
## **1. Introduction: The Case of Le Signal**
- **Le Signal**, a residential building in Soulac-sur-Mer (France), was constructed **200 meters from the ocean** and **demolished in 2023** due to coastal erosion.
- **78 families lost their homes**—a predictable outcome given the building’s location in a high-risk zone.
- **Key takeaway**: Coastal erosion is **slow but sudden**, with devastating consequences for vulnerable communities.
---
## **2. The Global Scale of Coastal Risk**
- **1 billion people** (nearly **1 in 8 globally**) will live in **low-elevation coastal zones** by **2050**, exposed to rising sea levels.
- **France alone** has **371 local governments** legally required to map coastal risks, yet **most lack the tools** to do so effectively.
- **Le Signal was not an accident**—it was a **failure of foresight**, and similar risks exist in coastal areas worldwide.
---
## **3. The Problem: Inadequate Tools for Decision-Makers**
### **3.1 Current Limitations**
- **Decision-makers (mayors, insurers, developers, planners)** receive **incomplete or unusable data**:
- **Spreadsheets with rough estimates** (no timelines, scenarios, or location-specific details).
- **No actionable insights**—e.g., which parcels, streets, or infrastructure are at risk and when.
- **Existing tools are designed for researchers, not policymakers**:
- **Slow** (months of processing).
- **Expensive** (requires specialized teams to interpret).
- **Static** (no forward-looking scenarios or risk scores).
### **3.2 The Consequences**
- **Decisions are made in the dark**, leading to **poor planning, financial losses, and displacement**.
- **Example**: A mayor knows their coast is retreating but **cannot predict which specific plots will be at risk in 30 years**.
---
## **4. The Solution: AI-Powered Coastal Risk Prediction**
### **4.1 Leveraging Satellite Data & AI**
- **Satellites have documented coastlines since 1972**, providing **50+ years of high-resolution imagery**.
- **AI can analyze this data** alongside **climate projections and coastal physics** to predict erosion risks at **parcel-level resolution**.
- **AquaAdapt’s approach**:
- **Input**: A specific parcel (e.g., a building, road, or land plot).
- **Output**: **Possible futures** (10, 30, 50, 100 years) under **multiple climate scenarios** (low, medium, high emissions).
- **Confidence intervals** for each projection (tighter for near-term, wider for long-term).
### **4.2 Example: Soulac-sur-Mer (Site of Le Signal)**
| **Time Horizon** | **Low Emission Scenario** | **High Emission Scenario** |
|------------------|--------------------------|---------------------------|
| **10 years** | 18 meters lost | 30 meters lost |
| **50 years** | 58 meters lost | 118 meters lost |
| **100 years** | 100+ meters lost | **Parcel completely gone** |
- **Real-world impact**:
- A **single storm in 2013-2014** eroded **40 meters**—exceeding projections for **2040**.
- **Current erosion rate in Nouvelle-Aquitaine**: **8 meters/year** (up from **4.3 meters/year** between 1997-2021).
### **4.3 Key Features of AquaAdapt’s Model**
- **No GIS expertise required**—accessible to **mayors, insurers, developers, and planners**.
- **Uncertainty as a primary output**:
- **Not deterministic** (unlike tools like ChatGPT, which provide single answers).
- **Explicit confidence intervals** (e.g., "This projection is 70% likely").
- **Transparency**: Clearly states where predictions are **less reliable** (e.g., long-term forecasts).
- **Goal**: **Inform decisions, not replace judgment**.
---
## **5. The Challenge: Trust in Tools vs. Reality**
- **Risk of overconfident AI**:
- **Bad AI**: "This **will** happen." (False certainty)
- **Good AI (AquaAdapt)**: "Prepare for **these possible futures**, with **X% confidence**." (Actionable uncertainty)
- **Why uncertainty matters**:
- **Decision-makers often assume predictions are 100% accurate**—leading to **poor planning**.
- **AquaAdapt’s model** makes uncertainty **visible and quantifiable**.
---
## **6. Adaptation Strategies: What Can Be Done?**
### **6.1 Gray Solutions (Engineered Defenses)**
- **Seawalls, dikes, and artificial barriers** to **slow erosion**.
- **Limitations**: Expensive, temporary, and can **disrupt ecosystems**.
### **6.2 Green & Blue Solutions (Nature-Based)**
- **Replanting mangroves, dunes, or wetlands** to **absorb wave energy**.
- **Example**: Mangroves in **Vanuatu** (not feasible in metropolitan France).
- **Limitations**: **Climate-dependent** (not all regions support these ecosystems).
### **6.3 Planned Retreat (Managed Relocation)**
- **Gradual relocation** of communities and infrastructure **away from high-risk zones**.
- **Challenges**: **Political, financial, and social resistance**.
### **6.4 AquaAdapt’s Future Vision**
- **Integrated decision-support platform**:
- **Maps + reports** generated from **user-defined zones and scenarios**.
- **Goal**: Help policymakers **choose the best adaptation strategy** for their context.
---
## **7. The Team & Technical Approach**
### **7.1 Team Background**
- **Lily Pinault (Founder)**:
- **Earth sciences** (oceanography, climatology).
- **Satellite data analysis** (remote sensing).
- **AI/ML** (current master’s studies).
- **Co-founder**: Legal/finance background (VC/investment expertise).
- **Team**: Urbanist, AI researchers, and a **web developer** (Pinault’s brother).
### **7.2 Current Development Stage**
- **Data collection**: Satellite imagery, climate models, coastal physics.
- **Model experimentation**:
- **Testing different AI architectures** (parameters, hyperparameters).
- **Team debates** on best approaches—**open to external insights**.
- **Next steps**:
- **Refining predictions**.
- **Building a user-friendly platform** (draw a zone → select scenario → get risk report).
---
## **8. Conclusion & Call to Action**
- **Coastal risks are a defining challenge of the 21st century**, but **the science and data already exist**.
- **The gap**: **Bridging data and decision-makers** with **fast, accessible, and transparent tools**.
- **AquaAdapt’s mission**: Provide **parcel-level risk predictions** to **prevent future Le Signals**.
- **Invitation**:
- **QR code** links to **bilingual slides** (French/English).
- **Follow-up discussions** encouraged—**contact Lily Pinault for collaboration**.
---
### **Key Takeaways**
✅ **Coastal erosion is accelerating**—**1 billion people at risk by 2050**.
✅ **Current tools are inadequate**—**slow, expensive, and not actionable**.
✅ **AI can bridge the gap**—**predicting risks at scale with quantified uncertainty**.
✅ **Adaptation is possible**—**gray, green, and retreat strategies**, but **no one-size-fits-all solution**.
✅ **AquaAdapt is building the future of coastal risk assessment**—**transparent, fast, and decision-ready**.