SBIR-STTR Award

Our proposal addresses the need expressed in subtopic 26.a to measure vertically resolved soil moisture distributions. While soil moisture content is a key state variable in subsurface hydrobiogeochemical processes, spatially resolved soil moisture flux is an equally important process variable. We propose to develop an intelligent sensor that will use machine learning to infer vertically resolved soil moisture flux from changes in vertical moisture distribution as measured using technology we previously developed. Our technical approach will build on technology we previously developed for measuring vertically resolved soil moisture distributions using time domain reflectometry TDR), which resulted in three U.S. patents and ongoing negotiations to license the technology to established manufacturers. The new flux measurement system will be suitable for monitoring spatially resolved moisture flux within the 1-2 meter length of the probe, including surface infiltration, root zone uptake, and deep drainage. Using well-established modeling codes for unsaturated flow such as HYDRUS) to generate a vast model ensemble of soil water dynamics in a variety of soil types in response to variable surface wetting sequences, we will train a deep learning network to identify flux including source and sink terms. This machine intelligence, validated against field data, will be useable with a variety of soil water content measuring devices to simultaneously infer both soil hydrologic behavioral characteristics and time varying flux from time series observations of vertical moisture content distribution in response to time- varying surface wetting. Phase I will include ensemble model generation, architecture, training and validation of the machine intelligence, and testing against two years of in situ water content data from a DOE field site. Commercial applications and other benefits range from agricultural crop management and water conservation to environmental site management and monitoring earth-atmosphere moisture and energy exchange for climate observation and study.

We are developing a machine intelligence (MI) that will infer vertical soil water flux from changes in vertical moisture distribution measured using available moisture sensors. This innovation addresses a serious information deficit that limits water management and conservation efforts, and will have immediate societal impact in high value agricultural regions such as California’s Central Valley, while also addressing the DOE need expressed in subtopic 26.a to measure vertically resolved soil moisture distributions. Soil moisture content is a key state variable in subsurface hydrobiogeochemical processes, but soil moisture flux is an equally important process variable for which there is currently no practical means of measurement. In Phase I, we successfully trained an artificial neural network to predict shallow and deep flux within reasonable error tolerances using only soil moisture as input from a vast ensemble of physics-based modeling results. No soil information was required. This work involved modeling 1326 soil textures under various wetting and root uptake conditions as input to training, optimizing, and validating artificial neural networks. In Phase II, we propose to optimize our discoveries to address a specific high impact target and then generalize for a broader range of applications based on what is learned there. Phase II will also perform comprehensive field validation and develop a commercially viable end-to-end product. The product will be an application programming interface (API) as a service that will provide researchers and growers with operationally useful estimations of deep drainage using input data from soil moisture sensors in real-world environments. We will first develop and release this technology to serve the need to quantify deep drainage under permanent crops in California’s Central Valley. Building on that foothold, we will expand the functionality of the product to capture a broadening market both geographically and with respect to crop diversity, as well as provide compatibility with a wider range of commercially available soil moisture probes. Commercial applications and other benefits range from agricultural crop management and water conservation to environmental site management and monitoring earth-atmosphere moisture and energy exchange for climate observation and study.