SEEA-AI: A New Model for Natural Capital Quantification Combining AI and Remote Sensing Technology

SEEA-AI is an innovative open-source natural capital accounting model that combines artificial intelligence (e.g., random forest algorithm) and remote sensing technology to quantify the environmental and economic value of ecosystems. The model has been validated in Australia's Murray-Darling Basin (temperate climate) and Colombia's Sinú River Basin (tropical climate), demonstrating transfer learning capabilities across climate zones.

Developed and maintained by Vilar Ramírez, González, and Bastons Prat, the source code is hosted on GitHub (link: https://github.com/ivilaruic/SEEA_AI), and the related paper was submitted to the Ecosystem Services journal in 2026. Its core goal is to address the high cost and scalability challenges of field measurements in traditional natural capital accounting, as well as the data acquisition and model accuracy challenges in the practical application of the UN SEEA-EA framework.

Ecosystems provide indispensable services for humans such as clean water, carbon storage, and agricultural support. However, converting these 'natural capitals' into quantifiable economic values has long been a major challenge in environmental science and policy-making. Traditional accounting methods rely on expensive field measurements, making large-scale promotion difficult.

The SEEA-EA (System of Environmental-Economic Accounting—Ecosystem Accounting) framework developed by the United Nations Statistical Commission provides guidance for standardized ecosystem accounting, but it still faces challenges of data acquisition difficulties and insufficient model accuracy in practical applications, which are the core issues that the SEEA-AI project aims to solve.

Multi-source Remote Sensing Data Fusion

SEEA-AI integrates multiple satellite data sources to comprehensively assess ecosystem conditions:

• Landsat series: Used for land cover change and calculation of vegetation indices (NDVI/NBR)
• Hansen Global Forest Change Data: Monitors forest cover changes
• RADD Tropical Deforestation Alerts: Real-time deforestation monitoring
• MODIS products: Net Primary Productivity (NPP) and vegetation indices
• Soil Erosion Model (RUSLE): Evaluates soil conservation services

Machine Learning-driven Biomass Estimation

A random forest algorithm is used to build a biomass prediction model, with NASA/ORNL carbon monitoring data as training labels. After training in the Murray-Darling Basin, the model was validated in the Sinú River Basin. Key indicators:

• Coefficient of determination (R²) = 0.77
• Root Mean Square Error (RMSE) = 13.2 tons of carbon per hectare

SEEA-AI estimates the carbon sequestration rate at 12.6 tons of CO₂ per hectare per year through remote sensing data inversion, which is highly consistent with the reference value of 9.9 tons of CO₂ per hectare per year based on field measurements by Smith et al. (2025). This result is further confirmed through cross-validation of Landsat, Hansen, and MODIS data.

The reliability of SEEA-AI is cross-validated with independent datasets from Australia's Commonwealth Scientific and Industrial Research Organisation (CSIRO). The project documentation provides complete DOI links to ensure transparency and traceability of data sources.

Validation strategies include:

• Comparison with values from published literature
• Consistency check of multi-source remote sensing data
• Performance evaluation of cross-regional model transfer

Value for Policy Makers

Provides a cost-effective tool that can quickly establish ecosystem accounts in areas lacking dense ground monitoring networks, which is particularly significant for developing countries with rich natural resources but insufficient monitoring infrastructure.

Value for Researchers

Demonstrates the combination of cutting-edge machine learning technology and mature ecological methods, opening up new possibilities for ecosystem service assessment. The open-source licenses (code MIT/data CC-BY) ensure the accessibility and scalability of the results.

SEEA-AI represents an important advancement in the field of environmental accounting, proving that the combination of artificial intelligence and remote sensing technology can produce reliable natural capital assessment results. Cross-regional validation shows that properly calibrated models can be transferred across different ecological zones, significantly lowering the cost threshold for large-scale implementation.

With the extension of satellite data time series and the improvement of machine learning algorithms, methods like SEEA-AI are expected to become standard tools for global ecosystem monitoring, providing technical support for achieving the 'natural capital accounting' goal in the UN Sustainable Development Goals.