Optimizing light gradient boosting machine with the slime mould algorithm for reference evapotranspiration estimation

Accurate estimation of reference evapotranspiration (ETo) is crucial for achieving precision irrigation and sustainable water resource management, particularly in regions with limited meteorological observation data. This study employed the Gradient Boosting Decision Tree (GBDT) feature selection method to quantify the influence of meteorological factors on ETo and constructed estimation models. The key hyperparameters of the Light Gradient Boosting (LGB) model were optimized using the Slime Mould Algorithm (SMA), thereby establishing the novel hybrid optimization model (SMA-LGB). Using 30 years of daily meteorological data from the Songliao Plain, the performance of different ETo models was systematically evaluated. The results indicated that the ETo estimation accuracy improved as the number of input meteorological factors increased. Significant regional differences in ETo estimation accuracy were observed, with lower RMSE values in the southeastern regions and higher values in the northern areas. Compared to traditional machine learning models, the SMA-LGB model, which incorporates bio-inspired optimization algorithms, demonstrated higher accuracy and more stable performance under different factor combinations and spatial contexts. Notably, when data were missing at the target station, the SMA-LGB model still maintained high accuracy (R² > 0.9, RMSE < 0.5 mm/d) by relying on data from nearby stations. Moreover, under identical conditions with limited meteorological inputs, SMA-LGB consistently outperformed the locally calibrated Hargreaves-Samani (H-S) empirical model, achieving higher accuracy and more stable performance in ETo estimation. Overall, the SMA-LGB model significantly enhances both the accuracy and robustness of ETo estimation across the Songliao Plain, providing a reliable reference for crop water requirement assessment, precision irrigation decision-making, and refined agricultural water resources management. • Quantified the importance of meteorological factors in ETo estimation. • A hybrid model was proposed for robust hyperparameter optimization and estimation. • Thirty years of daily data in the Songliao Plain were systematically evaluated. • The model maintained high estimation accuracy even under cross-station conditions.