Explainable SAR Measurements for Wind Assessment with Artificial Intelligence (ESAWAAI)

The Explainable SAR Measurements for Wind Assessment with Artificial Intelligence (ESAWAAI) project, funded by the European Space Agency (ESA) Phi-Lab, is dedicated to advancing the retrieval of sea surface wind fields using a diverse set of observables derived from multi-polarization Single-Look Complex (SLC) Synthetic Aperture Radar (SAR) data, with a primary focus on Sentinel-1 sensors. By integrating advanced SAR processing techniques with Artificial Intelligence (AI) and Explainable AI (XAI) methodologies, the project is developing a comprehensive legacy dataset to support both data-driven and physics-informed Deep Learning (DL) models for wind estimation.

A central objective of ESAWAAI is to enhance the interpretability of the Geophysical Model Function (GMF) used in SAR-based wind retrieval. This includes improving transparency in the relationships between radar observables and geophysical wind parameters, proposing wind retrieval schemes that avoid the use of a priori ancillary wind information, and achieving improved performance compared to state-of-the-art wind estimation algorithms.

Dataset Overview

Our legacy dataset incorporates features derived from the NORA3 and CERRA numerical weather prediction and regional reanalysis models, corresponding respectively to the North Sea and Mediterranean Sea domains. These model-derived variables, along with official ESA Level-2 Ocean (L2 OCN) products—such as sea surface wind fields and rain flags predicted by the AI algorithm developed by Collins et al. (2025)—are embedded directly within the Level-1B (L1B) product as complementary features. The dataset aggregates more than three years of Sentinel-1 SLC SAR data, processed using the SAR WAVE framework.

Only Sentinel-1 Interferometric Wide (IW) mode data in VV polarization are included in this release.

Data Structure and Processing

The dataset preserves the structure of the original Level-1 SLC products:

• Each subswath (IW1, IW2, IW3) is processed independently
• Within each subswath, data are segmented by burst
• Each burst is further divided into smaller tiles, enabling wind and wave-related analyses

Spatial Grids

Two spatial grids are defined to organize the data:

• Intra-burst grid: tiles extracted within individual bursts
• Inter-burst grid: tiles corresponding to overlap regions between successive bursts

This structured tiling strategy ensures flexibility and consistency for downstream processing tasks, including machine learning model training and evaluation.

Consortium and Applications

The ESAWAAI project is developed by a consortium of leading institutions:

• ESA Phi-Lab
• CLS – Collecte Localisation Satellites Group, France
• IFREMER– Institut Français de Recherche pour l'Exploitation de la Mer, France
• DTU – Technical University of Denmark
• UPB – University Politehnica of Bucharest, Romania

Beyond dataset development, ESAWAAI aims to deepen scientific understanding of SAR observables across a wide range of metocean and geometric conditions. The resulting products support applications in:

• Meteorology
• Climate science
• Wind energy resource assessment
• Joint wind and wave retrieval