Machine Learning Predicts Carbon Emissions and Economy-Land Linkage in IORA Countries: A New Tool for Sustainable Development

When Development and Environment Intertwine on the Indian Ocean Rim

The Indian Ocean rim countries face a classic dilemma: how to promote economic growth without sacrificing the environment? For the member states of the Indian Ocean Rim Association (IORA), especially island or coastal economies such as Malaysia, Mauritius, Sri Lanka, and Madagascar, climate vulnerability forces policymakers to quantify the relationship between economic development and ecological footprint.

A study published in *Scientific Reports* attempts to provide a new analytical framework for this challenge using machine learning. The study constructed a predictive model called XOS-ELM-GA (Xavier Online Sequential Extreme Learning Machine with Genetic Algorithm), using historical CO₂ emission data as key input to predict the annual GDP and agricultural land area changes of these countries.

Nonlinear Capture Beyond Traditional Regression

Traditional economic models often assume linear relationships, but the interactions among carbon emissions, economic output, and land use are inherently nonlinear and dynamic. The researchers first conducted statistical analysis on historical data from 1960–2020, confirming significant correlations between CO₂ emissions, GDP, and agricultural land. However, simple regression cannot efficiently handle high-dimensional, non-stationary time series.

The core innovation of XOS-ELM-GA lies in reducing random fluctuations through Xavier weight initialization and combining genetic algorithm to optimize parameters, overcoming the shortcomings of traditional Extreme Learning Machine (ELM) such as high prediction uncertainty, parameter sensitivity, and low computational efficiency. Tests on four countries show that the model achieves an average SMAPE between 10.13% and 12.13% in GDP prediction and as low as 2.99% to 3.77% in agricultural land prediction, significantly outperforming benchmark models such as ELM and OS-ELM.

Four Cases, One Structural Logic

Malaysia, Mauritius, Sri Lanka, and Madagascar represent different development stages and industrial structures within the IORA. Malaysia, as a middle-income industrial country, has carbon emissions closely tied to manufacturing growth; Mauritius has undergone a transition from agriculture to services, with its carbon intensity changes paralleling economic structure upgrades; Sri Lanka oscillates between tourism and agriculture; Madagascar relies on agriculture but is constrained by low industrialization. Despite these differences, the machine learning model captures a commonality: CO₂ emissions can serve as a reliable proxy variable for predicting economic activity and land change.

This finding suggests that, given high availability of environmental data, emission trajectories can provide signals to policymakers about the direction of GDP and cropland changes several years in advance. For developing countries lacking detailed economic statistics, this may be a pragmatic approach to quickly assess policy impacts.

From Prediction to Policy: A Bridge to SDGsThis study explicitly does not aim to establish causal relationships, but rather provides "correlation-based predictions." However, the predictive value should not be underestimated. The advancement of UN Sustainable Development Goals (SDGs) 8 (Decent Work and Economic Growth), 13 (Climate Action), and 15 (Life on Land) in IORA countries is often constrained by data lags and model complexity. The simplicity and online learning capability of XOS-ELM-GA allow it to continuously absorb new data and output near-real-time predictions, thereby supporting dynamic adjustments in emission reduction and land-use planning.

For example, the model achieves minimal prediction errors for agricultural land area in Sri Lanka and Madagascar, which means climate policymakers can more confidently assess the long-term impact of reforestation or agricultural intensification on the carbon budget.

AI-Driven Geoeconomic Paradigm Shift

This study is a microcosm of a global trend: artificial intelligence is redefining the analytical boundaries of environmental economics. With the development of technologies such as extreme learning machines and genetic algorithms, the computational bottlenecks of traditional quantitative methods are being broken. More importantly, such "data-driven" models lower the barrier for cross-disciplinary research—economists do not need to be climate science experts, and ecologists do not need to be algorithm specialists to use predictive tools to dialogue on shared challenges.

IORA covers nearly 3 billion people, accounting for one-third of global trade. If similar models can be extended to other member states, they would not only aid sustainable planning in individual countries but could also spawn regional coordination mechanisms. For instance, green investments targeting marine plastic pollution or mangrove protection could first use models to simulate economic returns and land trade-offs under different carbon emission scenarios, before reaching consensus.

Limitations and Future Directions

The study also acknowledges its limitations: current predictions are based on historical correlations and cannot extrapolate disruptive technologies or policy shifts (e.g., large-scale deployment of carbon capture). Moreover, using only CO₂ as a single environmental variable overlooks other sustainability dimensions such as biodiversity and water resources. Future work could integrate more satellite remote sensing data (e.g., nighttime lights, vegetation indices) and introduce multi-task learning frameworks to simultaneously predict multiple SDG indicators.

In the long run, machine learning will not replace structural analysis, but it gives decision-makers an additional "operational time window." When IORA countries negotiate for funding in climate finance talks, a peer-reviewed predictive model based on local data may be more persuasive than abstract commitments.

> Source: Xu, X., et al. (2026). Machine learning-based forecasting of CO₂-related economic growth and agricultural land change in IORA countries. *Scientific Reports*, 16, 21411. https://www.nature.com/articles/s41598-026-51807-1