Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Adaptive Cropping Systems Laboratory » Research » Publications at this Location » Publication #409919

Research Project: Sustainable and Resilient Crop Production Systems Based on the Quantification and Modeling of Genetic, Environment, and Management Factors

Location: Adaptive Cropping Systems Laboratory

Title: TDR-Transformer: A transformer neural network model to determine soil relative permittivity variations along a time domain reflectometry sensor waveguide

Author
item WANG, ZHUANGJI - University Of Maryland
item Timlin, Dennis
item GONG, XIAOFEI - Navy Federal Credit Union
item KOJIMA, YUKI - Gifu University
item HUA, SHAN - Zhejiang Academy Of Agricultural Sciences
item Fleisher, David
item SUN, WENGUANG - University Of Nebraska
item SAHILA, BEEGUM - University Of Nebraska
item Reddy, Vangimalla
item TULLY, KATHERINE - University Of Maryland
item HORTON, ROBERT - Iowa State University

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/2/2025
Publication Date: 8/13/2025
Citation: Wang, Z., Timlin, D.J., Gong, X., Kojima, Y., Hua, S., Fleisher, D.H., Sun, W., Sahila, B., Reddy, V., Tully, K., Horton, R. 2025. TDR-Transformer: A transformer neural network model to determine soil relative permittivity variations along a time domain reflectometry sensor waveguide. Computers and Electronics in Agriculture. https://doi.org/10.1016/j.compag.2025.110730.
DOI: https://doi.org/10.1016/j.compag.2025.110730

Interpretive Summary: It is more efficient to measure the soil water content at multiple depths, using one vertically installed long time domain reflectometry (TDR) sensor. The soil water contents as a function of depth will be recorded in the TDR waveforms. Inspired by the generative pre-trained transformer (GPT) data processing pipeline, a new transformer neural network data interpretation model is developed. In the transformer neural network, a TDR waveform is treated as a time series with geometrical information, and the overall effect of the transformer neural network is to translate a TDR waveform into a sequences soil water content values that can represent the spatial variations. Simulated and measured TDR waveforms under varying soil water content but constant electrical conductivity are used to evaluate the model performance. This study presents the first attempt to apply transformer neural network in TDR waveform analyses, and the proposed neural network model structure can be exploited and applied in a range of TDR sensor configurations. This research will be useful to scientists, growers and practioners who are interested in measuring soil water contents.

Technical Abstract: Interpreting soil relative permittivity (er) variations along a time domain reflectometry (TDR) waveguide provides an opportunity to determine soil water content at multiple depths using a vertically installed TDR sensor. Compared to placing sensors at different depths, vertical sensor installation reduces measurement efforts and enhances data-use-efficiency. Revealing er variations includes two aspects: identifying er change positions and determining er values. Traditional inverse analyses are not widely applied due to their high computational demands. Machine learning-based methods, e.g., TDR-CNN, provide a forward computational workflow to track er change positions and reduce computational load, but errors in er values are relatively large. In this study, TDR-Transformer is developed as a new waveform interpretation model to improve er estimation accuracy. Modified from the standard transformer architecture, an encoder with convolutional neural layers is used to extract waveform geometric features, and a decoder generates a sequence of er values to represent er variations. Attention is a mechanism that can dynamically extract and process the relevant information within the data, which processes and integrates the waveform geometric information in the encoder, ensures the causality (time-order) of the waveform data in the decoder, and transfers information from the encoder to the decoder. TDR-Transformer was trained and tested using simulated waveforms where er changes along the waveguides, but soil electrical conductivity (EC) was assumed to be small and stable. The RMSE for er values was within 0.5–1.6 % and the RMSE of er change positions was within 5–8 %. A soil infiltration experiment and a precipitation evaporation experiment illustrated applications of TDR-Transformer to observed waveforms. Consequently, TDR-Transformer is a promising artificial intelligence model to interpret TDR waveforms in soils with nonuniform(er), and fine-tuning TDR-Transformer is recommended for specific commercial TDR sensor designs.