Work

Lightweight physics-informed edge AI: a scalable quantized surrogate framework for smart irrigation and water use efficiency

Year: 2026

Type: article

Abstract: The development of a lightweight edge artificial intelligence framework addresses the primary objective of enhancing water use efficiency within resource-constrained smart irrigation systems. Integration of physics-informed synthetic data generation with autonomous surrogate modeling establishes a phenology-driven architecture. Post-training 8-bit integer quantization minimizes the hardware footprint, while simulated sensor degradation evaluates long-term operational scalability. Simulated evaluations demonstrate a sustained macro F1-score of 0.977 for the quantized surrogate model. Statistical Y-scrambling techniques corroborate algorithmic authenticity and yield a 0.952 baseline retention score, effectively mitigating chance correlation. Cycle-accurate hardware profiling restricts the edge-native framework within an 8-kilobyte execution memory constraint and calculates exactly 630.0 micro-joules per inference, achieving a 73.3 percent reduction in computational energy consumption. Cost-sensitive economic evaluations quantify a 96.23 percent Economic Savings Index relative to heuristic baseline methods. Unsupervised feature drift monitoring protocols support predictive reliability across a simulated sixty-month hardware degradation timeline without requiring physical ground-truth data labels. The proposed system design directly embeds deterministic biophysical guardrails into the hybrid deterministic-AI inference engine, enhancing agricultural decision-making safety completely independent of cloud connectivity. Research limitations stem from the reliance on synthetic data and the rigid execution environment of ultra-low-power microcontrollers, preventing the algorithm from capturing long-term phenotypic plasticity and spatial micro-climate heterogeneity. Practical implications highlight the potential democratization of precision agriculture for smallholder communities through this scalable technological paradigm, supporting global socioeconomic viability and water sustainability goals.

Source: Journal of the Saudi Society of Agricultural Sciences

DOI: 10.1007/s44447-026-00193-1

Authors: Chairote Yaiprasert, Achmad Nizar Hidayanto

Institutions: Walailak University, University of Indonesia

Language: English

Topic: Smart Agriculture and AI

Subfield: Plant Science

Field: Agricultural and Biological Sciences

Domain: Life Sciences

SDGs: Industry, innovation and infrastructure, Zero hunger

Open Access status: diamond

Funders: Universitas Indonesia, Walailak University, Ministrstvo za visoko šolstvo, znanost in tehnologijo, Lembaga Pengelola Dana Pendidikan