Recursive classification of satellite imaging time-series: An application to land cover mapping

This project developed an advanced approach to surface water mapping using satellite imagery, combining Bayesian recursion with traditional classification methods.
Key Features:
Implemented a Bayesian recursion classifier for accurate surface water mapping
Utilized Sentinel-2 satellite imagery and spectral indices for classification
Achieved comparable performance to deep learning methods while requiring significantly less training data
Impact:
This innovative approach offers a more efficient and data-conservative method for surface water mapping, with potential applications in environmental monitoring, flood prediction, and water resource management. By reducing the data requirements while maintaining high accuracy, this method can be particularly valuable in regions with limited historical data or rapidly changing landscapes.
Technical Highlights:
Leveraged spectral indices such as Normalized Difference Water Index (NDWI) and Modified NDWI (MNDWI)
Compared performance against traditional classifiers and state-of-the-art deep learning models
Demonstrated improved robustness in multitemporal settings without additional computational complexity
This project showcases the ability to develop novel machine learning approaches that balance performance with practical constraints, a valuable skill in the field of geospatial data science and remote sensing.

Calatrava, H. et. al., (2023). Recursive classification of satellite imaging time-series: An application to water mapping, land cover classification and deforestation detection. ISPRS Journal of Photogrammetry and Remote Sensing. DOI: 10.48550/arXiv.2301.01796 (In review)