Accurate prediction of pan evaporation is critical for water resource management, agricultural planning, and climate adaptation. This study presents a novel hybrid framework integrating the Gaussian Mutation Bat Optimization Algorithm (GMBOA), Complementary Ensemble Empirical Mode Decomposition (CEEMD), and Adaptive Neuro-Fuzzy Infer ence System (ANFIS) for daily pan evaporation prediction. The approach is evaluated in two contrasting Iranian basins (Sefidrood and Kashafrood) to assess its generalizability. GMBOA simultaneously optimizes the parameters of CEEMD and ANFIS, while CEEMD reduces data nonstationarity by decomposing meteorological time series into intrinsic mode functions. The performance of the proposed model is compared with several classical, hybrid, and deep learning mod els, including RF, ANN, RNN, XGBoost, and LSTM, as well as their optimized versions. The model's performance is evaluated in testing phase using mean absolute percentage error (MAPE), index of agreement (IA), standard deviation of residual error (STDRE), and 95% uncertainty level (U95). The proposed model achieves the highest predictive accuracy across all settings, with MAPE ranging from 5% (single-day) to 8% (seven-day-ahead forecasting), IA up to 0.95, and U95 as low as 4. Results show superior performance compared to all benchmark models in both basins. Statistical sig nificance of the improvements is confirmed by the Wilcoxon signed-rank test (p < 0.05). The framework also demonstrates efficient computation time despite its complexity. These findings confirm the robustness, accuracy, and practical utility of the GMBOA–CEEMD–ANFIS model for pan evaporation forecasting across diverse climatic conditions and multiple forecasting horizons.
