Abstract Geologic CO2 Sequestration (GCS) is a promising engineering technology to reduce global greenhouse emissions. Real-time forecasting of CO2 leakage rates is an essential aspect of large-scale GCS deployment. This work introduces a data-driven, physics-featuring surrogate model based on deep-learning technique for CO2 leakage rate forecasting. The workflow for the development of data-driven, physics-featuring surrogate model includes three steps: 1) Datasets Generation: We first identify uncertainty parameters that affect the objective of interests (i.e., CO2 leakage rates). For the identified uncertainty parameters, various realizations are then generated based on Latin Hypercube Sampling (LHS). High-fidelity simulations based on a two-phase black-oil solver within MRST are performed to generate the objective functions. Datasets including inputs (i.e., the uncertainty parameters) and outputs (CO2 leakage rates) are collected. 2) Surrogate Development: In this step, a time-series surrogate model using long short-term memory (LSTM) is constructed to map the nonlinear relationship between these uncertainty parameters as inputs and CO2 leakage rates as outputs. We perform Bayesian optimization to automate the tuning of hyperparameters and network architecture instead of the traditional trial-error tuning process. 3) Uncertainty Analysis: This step aims to perform Monte Carlo (MC) simulations using the successfully trained surrogate model to explore uncertainty propagation. The sampled realizations are collected in the form of distributions from which the probabilistic forecast of percentiles, P10, P50, and P50, are evaluated. We propose a data-driven, physics-featuring surrogate model based on LSTM for CO2 leakage rate forecasting. We demonstrate its performance in terms of accuracy and efficiency by comparing it with ground-truth solutions. The proposed deep-learning workflow shows promising potential and could be readily implemented in commercial-scale GCS for real-time monitoring applications.