Regulating the microclimate to achieve the desired crop quality and yield demands substantial resource consumption, making it essential to optimize resource use. AI models can be used to forecast future plant development based on microclimate conditions, allowing controllers to preemptively adjust climate settings to optimize growth and resource consumption. However, the current paradigm of microclimate controller lacks AI-assisted feedback to predict how crops respond to dynamic climate conditions (crop × environment interactions). Thus, there is an urgent need to develop an AI-assisted predictive analytics system that can support decision-making processes. This study presents a multimodal deep learning approach for forecasting lettuce growth in CEA using both microclimate (aerial and rootzone) and early-stage plant image data. We employed Long Short-Term Memory (LSTM) networks to model the temporal dependencies of microclimate variables such as temperature, humidity, and light intensity. Further, we integrated image and microclimate data into the multimodal growth predictor to enhance T-days ahead prediction accuracy by capturing visual and temporal cues of plant growth and development. The model effectively predicted the lettuce growth trend using multimodal data, achieving high accuracy in its forecasts for the next few days. The combined use of LSTM and image data provides an efficient framework for forecasting lettuce growth, offering valuable insights for optimizing resource use in CEA.
