ENERGY-EFFICIENT TINYML ARCHITECTURES FOR REMOTE CLIMATE MONITORING

Remote climate monitoring is essential for understanding environmental change and supporting climateresilient planning, especially in geographically isolated and infrastructure-constrained regions.Conventional Internet of Things systems typically rely on cloud-centric analytics and frequent data transmission, resulting in high energy consumption, increased latency, and limited device lifetime.Tiny Machine Learning, or TinyML, offers a transformative alternative by enabling intelligent data processing directly on ultra-low-power embedded devices.This paper presents an energy-efficient TinyML architecture for remote climate monitoring applications, targeting key environmental parameters including temperature, humidity, rainfall, air quality, and soil moisture.The proposed design integrates lightweight neural network models with hardware-aware optimization techniques to operate under strict constraints on memory, computation, and power.Model compression strategies such as quantization, pruning, and feature reduction are employed to minimize inference cost while preserving predictive accuracy.In addition, adaptive sampling, duty-cycled sensing, and edge-level decision making are incorporated to reduce unnecessary communication and extend operational lifetime.