Effective pavement management relies on timely, accurate data to assess road conditions. However, traditional monitoring methods often provide infrequent data, hindering the development of robust short-term forecasting models for road degradation. This study evaluates how high-frequency data from computer vision condition assessment can be used to predict the change in crack count over winter months. A hybrid forecasting model combining Long Short-Term Memory (LSTM) and dense neural network structures with static features was developed, leveraging key distress indicators such as crack count, pothole count, and the Pavement Condition Index (PCI), supplemented with openly available static road information to enhance prediction accuracy. Additionally, qualitative insights from semi-structured interviews with industry stakeholders, including transport agencies, contractors, and researchers, were integrated to assess the practical applicability and operational challenges associated with predictive maintenance methods. The results show that combining high-frequency image data with ML techniques has the possibility to predict the formation of cracks over the winter, while limitations in the size of the dataset warrant further research on the subject. By enabling forecasting, the proposed approach, with more robust and diverse data, has the ability to support data-driven decision-making, leading to more timely, cost-effective, and efficient maintenance planning, which in turn could lead to safer roads. Ultimately, these advancements have the potential to extend pavement lifespan and optimize resource utilization across infrastructure networks.