AI-Enhanced Virtual LIG–IoT Sensor Framework for Microclimatic Stress Prediction in Vasconcellea stipulata (Toronche) from Southern Ecuador

(Toronche), yet current monitoring approaches rely on sparse measurements and lack real-time predictive capability. This work introduces an AI-enhanced virtual sensing framework based on laser-induced graphene (LIG) designed to emulate the thermoresistive response of an LIG transducer and generate high-resolution environmental indicators for microclimatic analysis. Unlike conventional LIG sensors or standalone IoT systems, the proposed framework integrates experimental calibration, data-driven modeling, and embedded inference into a unified architecture suitable for lightweight deployment on edge devices. A multilayer perceptron (MLP) model trained on laboratory data reproduced the temperature- and humidity-dependent electrical behavior of the transducer with high fidelity, achieving an RMSE of 0.016 kΩ in the calibrated range (10-60 °C) and remaining below 0.09 kΩ under noisy and extrapolated conditions. Sensitivity analysis identified temperature as the dominant driver (71%), followed by solar irradiance (19%) and relative humidity (10%), consistent with the microstructural mechanisms governing LIG's response. The virtual sensor enables continuous, low-cost environmental monitoring and provides quantitative variables that can support downstream ecological interpretation. Overall, the results highlight the potential of AI-enhanced LIG-IoT architectures for advancing real-time microclimatic assessment in resource-limited Andean ecosystems.