A Crop Digital Twin System for Predictive Growth Monitoring and Adaptive Light Control in Controlled Environment Agriculture

• Integrated digital-twin system enables predictive crop monitoring in CEA. • Edge-AI models achieve high-accuracy growth prediction for lettuce plants. • LoRaWAN sensor network provides low-latency, near real-time microclimate data. • Adaptive lighting control maintains high-precision illumination for optimized growth. Controlled Environment Agriculture (CEA) offers advantages, including improved yields, enhanced quality, and reduced waste compared to conventional farming. However, achieving optimized growth strategies remains a challenge. This study introduces a comprehensive framework for a Crop Digital Twin System (CDTS) to aggregate data pipelines for microclimate and crop growth, enabling predictive monitoring and control for resource-use optimization in CEA. We explored the fundamental construction process of CDTS, covering stages of data acquisition, transmission, model development, control, optimization, and refinement. The proposed CDTS framework establishes a closed-loop system that monitors microclimate and predicts lettuce plant growth, while adaptively regulating artificial illumination to compensate for variations in natural sunlight. Experimental evaluation demonstrates the computational efficiency of the edge-AI system, with average CPU utilization, GPU utilization, CPU temperature, and power consumption of 10.2%, 14.4%, 33°C, and 15.6 W, respectively, during fresh-weight and leaf-area inference tasks. CDTS-enabled plant growth prediction achieves R 2 of 0.928 and 0.903 for the studied phenotypic inferences, respectively. Sensor network performance further validates the framework, with microclimate sensors exhibiting an average daily latency of 165.24 ms (±13.20 ms) and nutrient sensors showing a slightly higher latency of 190.86 ms (±27.70 ms), highlighting the effectiveness of the LoRaWAN infrastructure for near real-time data acquisition. Moreover, the light control demonstrated high precision, maintaining illumination within a 2% error margin for 94.88% of the operational period, 2–4% for 3.71%, 4–10% for 1.29%, and exceeding 10% error for only 0.12%. Overall, the proposed framework enables precise environmental control, supports crop growth optimization, and improves management practices within the CEA.