A lightweight soil moisture prediction model based on irrigation cycle segmentation and Kalman filtering.

Efficient soil moisture prediction is essential for improving precision irrigation practices, promoting sustainable water use, and mitigating crop water stress to enhance yields in irrigated farmlands. However, most prediction models rely on large dataset and high computational resources, limiting their applicability in resource-constrained agricultural environments. This study proposes a lightweight and cost-effective soil moisture prediction model tailored for precision irrigation applications. The method integrates discrete wavelet transform (DWT) and autocorrelation analysis to capture temporal dynamics in soil volumetric water content (VWC) time series. Specifically, DWT is employed to identify change points and segment the time series into irrigation-cycle-based subsequences, while autocorrelation features are extracted to characterize temporal dependencies. These features are subsequently incorporated into a Kalman filtering (KF) framework for recursive prediction. To improve adaptability under varying field conditions, a dynamic updating and rolling prediction mechanism based on a sliding window is introduced, allowing continuous incorporation of newly observed irrigation cycles and removal of outdated information. Results demonstrate that the proposed model achieves competitive prediction accuracy with substantially reduced computational cost. Its low complexity and adaptability make it well suited for real-time deployment on low-cost edge devices in precision irrigation systems.