Enhancing water body detection with shadow awareness in remote sensing

In mountainous or highly varied terrains, shadows can appear spectrally similar to water due to their low reflectance. This means standard water detection algorithms may incorrectly identify shadowed slopes as water bodies, leading to overestimations and inaccuracies.

To address this challenge, our approach leverages a unique advantage of Sentinel-2 satellite imagery. Sentinel-2 consistently passes over the same location at the same local time (10:30 AM UTC), allowing us to accurately predict the positions of the sun and the distribution of shadow coverage for each scene. By integrating this shadow awareness into our detection pipeline, we are able to significantly reduce the occurrence of false positives in water body detection, ensuring more accurate results.

Our method incorporates COPDEM GLO-10, a high-quality Digital Elevation Model (DEM), to calculate static shadow maps. By inputting a given date and time, we can model:

• Solar position relative to topography
• Shadow distribution across a Sentinel-2 tile
• Seasonal variations in shadow length

A striking example comes from Cuevas de Almanzora reservoir in Southern Spain. By calculating shadows for 10:30 UTC on December 21st (Winter Solstice), our predicted shadow areas aligned almost perfectly with actual Sentinel-2 observations, confirming the accuracy of our model. The video below shows how shadows pre-computed from COPDEM GLO-10 at a given date and time on a Sentinel-2 image result in a near-perfect match with actual satellite imagery.

By incorporating shadow awareness into our ensemble-based water detection approach, we are already seeing significant improvements in algorithm performance.

• More accurate water body classification
• Reduced overestimation in shadowed areas
• Better environmental monitoring and decision-making

This breakthrough will soon be integrated into our next-generation water body detection model, further enhancing precision in remote sensing applications. Stay tuned for more updates as we refine and scale this innovation!

Learn more about our work on Improved water bodies mapping.

To stay up-to-date with the results of EvoLand and the future developments on this candidate prototype, sign up for our newsletter (here below or via this form), and follow us on LinkedIn, Mastodon or X (Twitter).

Questions about the article or the project? E-mail us at contact@evo-land.eu.