AI-Driven Earth System Intelligence for Ocean Health Trajectory Modelling, Forecasting, and Risk Stratification

Introduction:
Global ocean health is usually assessed through descriptive indicators and spatial score comparisons, yet marine systems evolve through complex longitudinal trajectories shaped by ecological condition, human use, climate-related pressures, and national management capacity. This study develops an interpretable artificial intelligence framework for transforming Ocean Health Index data into an Earth science decision support system for forecasting, risk stratification, and policy diagnostics.

Methods:
The study analyses 341,124 Ocean Health Index score observations covering 220 coastal countries and territories, 19 goals or subgoals, six score dimensions, and annual assessments from 2012 to 2025. The framework integrates data harmonisation, longitudinal trajectory modelling, K means country grouping, risk stratification, Random Forest next-year forecasting, feature importance analysis, and country-specific policy diagnostics. Forecast predictors were constructed from lagged goal and dimension level OHI components, while the contemporaneous overall Index was excluded to avoid circular prediction. The methodological design converts OHI data from a descriptive sustainability index into a predictive and interpretable Earth system intelligence framework.

Results:
The global Ocean Health Index increased from 71.85 in 2012 to 74.94 in 2019, declined to 69.50 in 2024, and partially recovered to 72.31 in 2025, remaining 2.63 points below its 2019 level. Goal-level change was highly uneven: Tourism and Recreation declined by 12.42 points, while Natural Products and Lasting Special Places increased by 7.67 and 7.38 points, respectively. Country trajectory clustering identified two regimes with mean 2025 scores of 65.07 and 74.47. Risk stratification classified 51 countries and territories as low risk, 98 as moderate risk, and 71 as high risk. The Random Forest model predicted next-year OHI scores with R² = 0.723, MAE = 3.07, and RMSE = 4.03, identifying Habitat, Carbon Storage, and Fisheries as the strongest predictors of future ocean health.

Conclusions:
This study demonstrates that longitudinal ocean health data can support AI-enabled Earth science intelligence. By combining trajectory modelling, predictive analytics, risk classification, and interpretable goal importance, the framework provides a scalable pathway for marine sustainability monitoring, coastal system prioritisation, and evidence-based ocean policy diagnostics.