Integrating Electrical Features for Simultaneous Prediction of Soil Moisture and Potassium Levels Based on Neural Network Prediction Model
Javad Jafaryahya, Rasool Keshavarz, Mehran Abolhasan, Justin Lipman, Negin Shariati
Abstract
This article presents a method to predict potassium percentage and volumetric water content (VWC) in soil by analyzing its electrical properties. It highlights the role of potassium as an important factor in soil salinity and examines how potassium and VWC affect soil permittivity and electrical conductivity (EC) at different frequencies. A dataset is created with diverse potassium levels and VWC percentages, measuring electrical features from 10 to 295 MHz at 58 discrete frequencies with 5-MHz resolution. Initially constructed with 20 soil samples, the dataset was expanded to 2000 samples through 2-D interpolation for model development, with 90% of the data used in the training phase. Using these measurements, a neural network (NN) model is developed to predict VWC and potassium levels accurately. The model’s prediction error improved significantly when using data from over 30 frequencies, reducing potassium prediction error to less than 0.05 g/kg and VWC prediction error to less than 0.4%. This study makes its measurements using a dielectric assessment kit (DAK) and vector network analyzer (VNA), comparing the outputs of two commercial sensors with actual values and providing valuable insights for the development of advanced soil sensors.