Accurate crop yield prediction is crucial to global food production and national food security. Traditional methods based on subjective assessments are often inaccurate. Recent advances in data collection, such as high-resolution sensors and deep learning techniques, have significantly improved the accuracy and scope of agricultural data, offering a more reliable insight into crop productivity. This study focuses on predicting the daily raspberry harvest using a deep learning approach that combines layers of 1D convolutional neural network (CNN), temporal convolutional network (TCN) and long-term short-term memory (LSTM). The model exploits comprehensive datasets, including weather conditions, irrigation schedules and greenhouse parameters, to capture the complex interactions between various environmental and management factors. The results demonstrate the significant predictive capabilities of the ConvLSTM-TCN model across different datasets, encompassing both greenhouse sensor data and regional historical climate data. The integration of various climate and plant age data improves the accuracy and robustness of the model. This study highlights the potential of advanced DL architectures to improve crop yield forecasts and intelligent agricultural decision-making for raspberry harvests.