Comparative study of TinyML approaches for edge deployment in water quality

Water, a vital resource for sustaining life, is increasingly threatened by population growth and global climate change, emphasising the need for intelligent and efficient water quality management solutions. This study investigates and compares two TinyML-based approaches for water quality monitoring, deployed directly on ARM Cortex microcontrollers to eliminate dependency on cloud-based processing. A publicly available dataset was used for both training and testing to evaluate the system’s feasibility. The first approach employs NanoEdge AI Studio for automatic model generation, while the second involves manual implementation using the X-Cube-AI package. Experimental evaluation shows that the NanoEdge AI Studio–based model achieves an accuracy of 76.76% with a RAM usage of 19 kB, outperforming the manually implemented model, which achieves 74.4% accuracy and 22 kB RAM consumption. The comparative analysis highlights trade-offs in terms of accuracy, memory footprint, and computational efficiency. Overall, the findings confirm the feasibility and effectiveness of embedded artificial intelligence deployment for secure, low-latency, and autonomous water quality monitoring in resource-constrained environments.