A Bayesian framework to assess future local climate risks on small islands using heterogeneous data and knowledge: application on Rangiroa atoll

Technical Summary: A Bayesian Framework to Assess Future Local Climate Risks on Small Islands

Problem Statement
Small tropical islands, particularly atolls, face escalating threats to habitability driven by the complex interplay of climate change, biophysical constraints, and socio-economic drivers. Habitability is defined here not merely as survival, but as the capacity to provide protection from hazards, adequate space, food, freshwater, economic opportunities, and cultural continuity. Assessing these risks is inherently difficult due to data scarcity, the heterogeneity of local conditions (geomorphology, settlement patterns, resource dependence), and the need to integrate global climate scenarios with local socio-economic pathways. Traditional modeling often struggles to combine quantitative data with qualitative expert knowledge or to explicitly quantify uncertainties across these interacting dimensions.

Methodology
To address these challenges, the authors developed a Bayesian Network (BN) framework applied to Rangiroa Atoll in French Polynesia. The methodology integrates:

• Habitability Pillars: The model evaluates five interconnected dimensions: availability of habitable land, food supply, freshwater supply, safety of settlements and infrastructure, and sustainable access to economic activities.
• Scenario Integration: The framework incorporates four Shared Socioeconomic Pathways (SSPs) combined with Representative Concentration Pathways (RCPs) to model future conditions in 2050 and 2100.
• Data Heterogeneity: The BN combines quantitative observational data (e.g., geomorphological trends, sea-level rise projections) with semi-quantitative analysis and structured expert judgment (SEJ). This allows the model to function in contexts where observational data is limited, such as specific local impacts of reef degradation on sediment supply.
• Probabilistic Structure: The model encodes probabilistic relationships between variables, allowing for the explicit quantification of uncertainty. It treats risk drivers (e.g., sea-level rise, reef degradation, import disruption) as conditional probabilities that propagate through the network to determine the risk level (Low, Moderate, High, Very High) for each habitability pillar.
• Adaptation Pathways: The framework was used to test the effectiveness of three adaptation strategies: "do nothing," "accommodation" (e.g., raising housing), and "internal relocation" within the atoll. Effectiveness was defined by the strategy's ability to reduce risk levels across the pillars.

Key Contributions

1. Integrated Risk Assessment: The study provides a unified mathematical framework that links global climate scenarios with local biophysical and socio-economic realities, specifically addressing the "context-specific" nature of atoll risks.
2. Quantification of Uncertainty: Unlike deterministic models, this approach explicitly quantifies the confidence levels associated with risk estimates, distinguishing between scenarios with high uncertainty (e.g., high-emission pathways) and those with more constrained outcomes.
3. Granular Spatial Analysis: The model demonstrates that risk is not uniform across an atoll. By applying the framework to eleven specific islands within Rangiroa, it highlights how local characteristics (e.g., elevation, settlement density, geomorphological stability) create distinct risk profiles.
4. Adaptation Pathway Exploration: The paper moves beyond risk identification to explore the "adaptation solution space," illustrating how the viability of specific measures (like accommodation vs. relocation) narrows as risk intensifies across multiple pillars.

Results

• Temporal and Scenario Trends: Climate risk to habitability increases over time and is significantly higher under high-emission scenarios (SSP5-8.5) compared to sustainable pathways (SSP1-2.6). By 2100, the probability of high risk to habitability reaches 30% under SSP5-8.5, compared to 5% under SSP1-2.6.
• Spatial Variability: Risk levels vary markedly between islands. Under SSP5-8.5 in 2100, the probability of high risk to settlements is 50% in Avatoru (a highly disturbed, low-lying urban island) but only 10% in Tiputa. Islands with higher elevations and less human disturbance, such as Otepipi, show lower risk probabilities.
• Threatened Pillars:
  • Food Supply: Projected to face high risk (65% probability) by 2050 due to coral reef degradation, even before 2100.
  • Settlements and Infrastructure: Avatoru and the airport island (Island 5) face the highest risks (50% and 75% probability of high/very high risk, respectively) due to the concentration of critical infrastructure in low-lying areas.
  • Land: Risk is driven by sea-level rise and the inability of highly disturbed islands to naturally accrete vertically.
• Severe Risk Conditions: The model identifies that severe risk (defined as irreversible or persistent impacts) is most likely to occur under a combination of global warming exceeding 3°C, severe ecosystem damage, and trade disruption. Import disruption alone has a 90% probability of leading to severe risk.
• Adaptation Effectiveness: For Avatoru under SSP5-8.5, "do nothing" strategies become ineffective by 2050 as food risks rise. "Accommodation" remains moderately effective until 2100 but becomes constrained when multiple pillars (land, water, settlements) simultaneously face high risks. "Internal relocation" to higher islands (e.g., Otepipi) emerges as a feasible option by 2100, provided those islands remain flood-free.

Significance and Claims
The authors claim that this Bayesian framework offers a robust tool for researchers and practitioners to assess risks and uncertainties in data-scarce environments. The study suggests that:

• Local Context is Critical: Global scenarios alone are insufficient; local geomorphology and socio-economic dependencies (such as reliance on imports) fundamentally shape risk outcomes.
• Adaptation Limits: Even with high global warming, Rangiroa may not become entirely uninhabitable if internal relocation is utilized. However, relying solely on accommodating the main settlement (Avatoru) is likely insufficient under high-emission scenarios by the end of the century.
• Decision Support: The framework supports "robust decision-making" by visualizing how biophysical constraints limit the adaptation solution space. It allows stakeholders to focus on plausible, consistent storylines of severe risk rather than low-probability extremes.
• Methodological Transparency: By making subjective choices and uncertainties explicit through sensitivity tests, the BN approach provides a transparent alternative to other multi-criteria or expert-judgment frameworks, particularly where quantitative data is incomplete.

The paper concludes that while the method has limitations regarding model structure simplification and data availability, it successfully demonstrates how heterogeneous data and knowledge can be synthesized to explore future habitability and adaptation options for small island communities.