Time domain reflectometry waveform interpretation with convolutional neural networks

Authors: WANG, ZHUANGJI - University Of Maryland; HUA, SHAN - Zhejiang Academy Of Agricultural Sciences; Timlin, Dennis; KOJIMA, YUKI - Gifu University; LU, SONGTAO - International Business Machines Corporation (IBM); SUN, WENGUANG - University Of Nebraska; Fleisher, David; HORTON, ROBERT - Iowa State University; Reddy, Vangimalla; TULLY, KATHERINE - University Of Maryland

Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/23/2023
Publication Date: 2/16/2023
Citation: Wang, Z., Hua, S., Timlin, D.J., Kojima, Y., Lu, S., Sun, W., Fleisher, D.H., Horton, R., Reddy, V., Tully, K. 2023. Time domain reflectometry waveform interpretation with convolutional neural networks. Water Resources Research. 59(2). Article e2022WR033895. https://doi.org/10.1029/2022wr033895.
DOI: https://doi.org/10.1029/2022wr033895

Interpretive Summary: Time domain reflectometry (TDR) is widely used in measuring soil water content. Improved methods to determine the spatial variation of soil water content along the TDR sensor rods, especially long sensor rods, can significantly increase the measurement accuracy and data use efficiency. In this study, we developed a deep learning model that can reveal the variations of soil water content (soil relative permittivity) along the sensor rods, and evaluated the model performance using simulated TDR data. The model presented in this study is important for soil scientists and agricultural engineers.

Technical Abstract: Time domain reflectometry (TDR) sensors are commonly used in determining the average soil water content in the soil. We propose a new TDR waveform interpretation model based on convolutional neural networks (CNNs), which can analyze waveforms measured from soils with non-uniform water content distributions along a TDR sensor. The proposed model, namely TDR-CNN, is constructed with three modules. First, the geometrical features of the TDR waveforms are extracted with a simplified version of VGG16 network (Simonyan and Zisserman, 2015). Second, the reflection positions in a TDR waveform are traced using a 1D version of the region proposal network (Ren et al., 2015). Finally, the soil relative permittivity values are estimated via a CNN regression network. The three modules are developed in Python using Google TensorFlow and Keras API, and then stacked together to formulate the TDR-CNN architecture. TDR-CNN is evaluated using simulated TDR waveforms with varying relative permittivity, and the accuracy and stability of the TDR-CNN model are confirmed. The proposed TDR-CNN model is simple to implement, and each module in TDR-CNN can be updated or fine-tuned individually with new datasets. In conclusion, the proposed TDR-CNN can effectively interpret TDR waveforms obtained in soil with a heterogeneous water content distribution.