Presentation on our ML SM forecast paper at the DOE-PNNL RemPlex Summit

John Kimball presented the following abstract on our ML SM forecast paper at the  DOE-PNNL RemPlex Summit in November, 2025.

Title: Soil moisture profile monitoring and forecasting using in-situ sensors, satellite remote sensing, and machine learning

Technical Session: From Earth to Orbit: Autonomous Measurements and Remote Sensing

Authors: Jinyang Du, Numerical Terradynamic Simulation Group, University of Montana, USA John S. Kimball, Numerical Terradynamic Simulation Group, University of Montana, USA Christopher J. Jarchow Applied Studies and Technology, RSI EnTech, LLC, Grand Junction, CO, USA.

Abstract Text: Soil moisture (SM) is an essential climate variable that governs land-atmosphere interactions, runoff generation, and vegetation productivity. Timely monitoring and forecasting of SM spatial distribution and vertical profiles are critical for applications such as drought early warning and assessing the performance of engineered covers, including those for uranium mill tailings disposal cells. Arid and semi-arid regions are often selected for waste storage due to low groundwater recharge, yet verifying cover efficacy requires non-invasive and spatially representative SM data. Traditional monitoring methods are invasive, costly, and limited in capturing spatial heterogeneity, particularly over vegetated disposal systems. To address these challenges, this study developed a satellite-driven machine learning (ML) approach to model high-resolution (30-m) daily SM dynamics by integrating multisource remote sensing data with in situ multi-layer SM measurements from the Montana Mesonet. The ML model established robust relationships between diverse predictors and in situ measurements across 4-, 8-, and 20-inch soil depths, achieving strong accuracy (R > 0.91; RMSE ≤ 0.047 cm³/cm³) with 1- to 2-week forecast lead times. The forecast system successfully predicted the onset, progression, and termination of the 2017 Montana flash drought, which was not fully captured by operational monitoring systems. By leveraging non-invasive remote sensing and ML, this approach offers a novel solution for long-term disposal cell monitoring, drought prediction, and water cycle assessments, with potential benefits for water resource management, precision agriculture, and environmental risk mitigation.