This study aims to reconstruct total water storage anomalies (TWSa) derived from GRACE satellite data using the LightGBM algorithm. It integrates hydroclimatic and environmental covariates including precipitation, land surface temperature (LST), evapotranspiration (ET), and vegetation cover along with topographical factors such as elevation and slope. This study investigates the long-term impacts of these variables on TWSa and examines potential delayed effects of GRACE signals. Guided by a robust theoretical framework that considers the intricate interplay of climatic and environmental factors on water storage, the research design employs a comparative modeling approach. LightGBM, random forest (RF), and support vector machine (SVM) models were implemented using GRACE and GRACE-Follow On (GRACE-FO) data from 2002 to 2022 in Iran. Key findings reveal that all three models achieved similar accuracy (RMSE ≈ 1.39 cm, R-squared ≈ 0.94, and NSE ≈ 0.89). However, LightGBM demonstrated superior computational efficiency, operating several hundred times faster than SVM and RF, making it advantageous for large-scale studies. Further, incorporating the time variable significantly enhanced predictive accuracy, surpassing the influence of ET and LST. The study also found that lagged effects of GRACE signals had a negligible impact on reconstruction accuracy. These findings suggest that LightGBM is a promising algorithm for efficiently and accurately reconstructing TWSa, with potential applications in large-scale hydrological studies.