WetSAT-ML: Wetlands flooding extent and trends using SATellite observations and Machine Learning

The tool is expected to benefit two main groups of users:

1. Local and national environmental authorities dealing with wetlands status monitoring and reporting: These users may generate high temporal (12 – 24 days) and high spatial (50 – 100 meters) resolution maps of wetlands flooding extent, and water permanence. The tool will also allow users to produce time series plots of flooded areas and identify intra- and inter-annual trends to support analysis and reporting needs.
2. Academic and scientific users in hydrology and ecology: These users can leverage the tool’s outputs to support calibration or validation of hydrodynamic models and to assess ecosystem functions and connectivity. Additionally, WetSAT can be used as a pedagogical tool for teaching and learning in areas of hydrology, environment, artificial intelligence, among others.
3. Developers and programmers: These users may develop or articulate WetSAT with different algorithms and methods as deep neural networks, support vector machines, long language models, regenerative AI, among others.

The first versions of the WetSAT-ML tool were developed by SEI-LA researchers during 2025, using R and Python programming languages. The tool was developed in response to a global need for reliable information on wetland ecosystems. Wetlands have been recognized as a global priority in several multilateral instruments, where related targets and reporting requirements have been established for the Parties. However, meeting reporting needs requires monitoring systems and tools that provide timely and accurate data on wetlands. According to the UNEP Mid-term status of SDG Indicator 6.6.1 and Acceleration Needs Report 2024, no global observation systems can currently deliver actionable data on wetland ecosystems, posing a challenge to achieving global commitments for wetlands conservation. The tool was developed in three main phases:

1. Algorithm development and model training,
2. Validation and proof of concept,
3. Documentation and dissemination of the tool.

For algorithm development and model training, testing areas in different regions with reference data were used. Specifically, the South Florida Everglades, United States, and the wetlands of the middle Atrato River basin, Colombia, were used. The algorithms are based on the packages and libraries available in Scikit-Learn. Currently, the tool has two pre-trained machine learning models (Random Forest and K-means). We developed a proof of concept in four wetland areas: the Meghna River, Thailand; and the Ayapel, Barbacoa, and Beté wetlands, Colombia. For documentation, please visit the following links, where you will find the complete technical support and guides for WetSAT in its official GitHub repository:

• Design and architecture
• Technical documentation
• User manual

The first version of the tool is available in the GitHub repository here.

The team is currently working on a scientific paper that details the tool’s scientific and technical development. WetSAT-ML is under continuous update and construction, and contributions from the public are welcome. New case studies and applications, proofs of concept, and opportunities for training new models are encouraged.