Reference evapotranspiration (ET0) is a key parameter in hydrological modeling, agricultural planning, and water resource management. Deep learning methods have shown promising capabilities in improving the accuracy of ET0 estimation by utilizing diverse datasets. So, this study investigates the application of three advanced deep learning models including Long Short-Term Memory (LSTM), Sequence-to-Sequence LSTM (SS-LSTM), and Meta-LSTM, for ET0 modeling in Tabriz, Iran. The models were calibrated and validated using meteorological data and remote sensing inputs at the time period of 2003 to 2023, employing 17 different combinations of variables to evaluate performance under varying input scenarios. The evaluation metrics used include Coefficient of Determination (R2), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Bias Error (MBE), and Willmott's Index (WI). Results indicated that Meta-LSTM outperformed other models, achieving exceptional accuracy with R2 values approaching 0.997, RMSE as low as 0.165 mm/day, and MAPE under 4.5% during both calibration and validation phases. Incorporating remote sensing data, particularly land surface temperature and vegetation indices, further enhanced prediction accuracy. Comparative analysis demonstrated the robustness of Meta-LSTM, with consistent and reliable predictions across all statistical metrics and visual assessments. These findings prove the potential of incorporating deep learning techniques with multi-source data for precise ET0 modeling. The proposed approach provides a reliable tool for improving water management strategies and agricultural planning in semi-arid and arid regions, addressing challenges posed by climate variability and water resource constraints.