Accuracy-Efficiency Trade-Offs in Smart Models for Soil Moisture Estimation

Abstract

Soil moisture (SM) is an important indicator for efficient water management, making irrigation a significant challenge in agriculture on a global scale. A well-foreseen soil moisture status enhances irrigation efficiency and promotes sustainable water usage. There are various models available for soil moisture estimation/prediction ranging from machine learning to deep learning. However, implementing any of these models in practice poses really serious computational challenges that could affect trade-offs concerning computational efficiency versus accuracy. In this paper, an investigation is carried out to compare the most used prediction models while exploiting real data to choose the best one. Gated Recurrent Unit (GRU), out of all algorithms that were compared, was able to beat competing methods while establishing clear and balanced considerations for resource usage and accuracy. The model achieves a root mean square error (RMSE) of 1.46, a mean absolute error (MAE) of 0.72, and a coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) of 89%, indicating strong overall performance. Additionally, it shows a CPU usage of only 5% during the testing phase and moderate memory usage.