In contemporary society, the reverse osmosis (RO) process is important for water treatment and reuse industry. However, membrane fouling remains a challenging issue during operation. To date, RO system operation mainly relies on the operator's knowledge, and the maintenance is carried out based on schedule or pre-defined criteria due to our limited understanding of RO fouling in real applications. To better understand the process and enable system optimization, an integrated data-driven coupled with adsorption model has been proposed in this study. The data-driven model calibrates the mechanistic model and predicts future values, while the adsorption model provides predictions on transmembrane pressure (TMP). By integrating these two models, an accurate prediction with high robustness and an insight into the detailed fouling mechanisms in real plants were achieved. In addition to the model itself, multiple regression analysis (MRA) had also been applied to identify the dominant fouling mechanisms. This analysis confirms that it is highly possible that membrane fouling is developed from an intermediate pore blockage to cake filtration.