Abstract

This presentation explores how satellite observations, weather data, soil information and machine learning can support daily soil-moisture estimates for farms. We compare tree-based models and transformers, examine their performance across locations and seasons, and discuss the challenges of applying them to individual fields. Particular attention is given to farms without irrigation, where rainfall and drying patterns help us understand changes in soil moisture. The work also explores how these estimates could support future predictions of soil-moisture changes following rainfall or a specified amount of irrigation. We highlight current progress, sources of uncertainty and the further validation needed to develop practical tools for monitoring field conditions and supporting farm water management.