Bovine Respiratory Disease (BRD) poses a significant challenge in beef fattening due to its complex causes. Relying solely on data-driven sensor methods for early detection may yield false alarms. This paper introduces an innovative approach that integrates a deep learning model with a BRD mechanistic model, utilizing pulmonary ultrasounds and clinical exams as sensor data and ground truth, respectively. By employing reliable clinical diagnostics, three crucial biological parameters were inferred, enabling the forecast of the number of asymptomatic animals up to 30 days. The deep learning model achieves 70% accuracy in diagnosis, and the BRD mechanistic model forecasts disease dynamics with less than 5% error. However, the hybrid method's weakness lies in clinical exams' uncertainty for some animal diagnosis, and improvements to the BRD model have been addressed in existing literature. Future work could explore incorporating biological exams or utilizing a pathogen-specific model for enhanced accuracy.