DEEP LEARNING APPLIED TO WATER SEGMENTATION

This study presents a novel AI-based method for tracking and analysing the movement of rock glaciers and landslides, key landforms in high mountain regions. By utilising time-lapse images, our approach generates detailed velocity data, uncovering movement patterns often missed by traditional methods. This cost-effective tool enhances geohazard monitoring, providing insights into environmental drivers, improving process understanding, and contributing to better safety in alpine areas.

The present study offers a unique, nested, high-resolution dataset that captures soil surface changes every 20 seconds during rainfall events, as well as over seasonal variations, across a range of spatial scales – plot, slope, and catchment. These data collected over a period of 3.5 years was facilitated by a camera-based approach. This open-access resource assists scientists in evaluating and refining existing models, improving process understanding, and training artificial intelligence.

This study develops a new method to improve the calibration and evaluation of models that predict soil erosion by water. By using advanced imaging techniques, we can capture detailed changes in the soil surface over time. This helps improve models that forecast erosion, especially as climate change creates new and unpredictable conditions. Our findings highlight the need for more precise tools to better model erosion of our land and environment in the future.

Soil erosion threatens landscapes worldwide, and understanding how surfaces change over time is key to addressing this issue. We developed a new camera-based system that automatically captures and analyzes daily surface changes on a hillside over several years. Triggered by rain and a clock, the system showed how weather and farming impact the land. Our method offers a powerful way to monitor surface changes and can help improve predictions and solutions for soil erosion.

This study presents a low-cost, automated system for monitoring river water levels using cameras and AI. By combining AI-based image analysis with photogrammetry, it accurately measures water levels in real-time, even in challenging conditions. Tested over 2.5 years at four sites, it achieved high accuracy (errors of 1.0–2.3 cm) and processed over 219,000 images. Its resilience makes it ideal for flood detection and water management in remote areas.