Comparison of Machine Learning Models to Predict Lake Area in an Arid Area

Machine learning (ML)-based models are popular for complex physical system simulation and prediction. Lake is the important indicator in arid semi-arid areas, to achieve proper management of water resources a lake basin, it crucial estimate predict dynamics, based on hydro-meteorological variations anthropogenic disturbances. This task particularly challenging regions, where scarcity poses significant threat human life. In this study, typical area China was selected as study area, performances eight widely used ML (i.e., Bayesian Ridge (BR), K-Nearest Neighbor (KNN), Gradient Boosting Decision Tree (GBDT), Extra Trees (ET), Random Forest (RF), Adaptive (AB), Bootstrap aggregating (Bagging), eXtreme (XGB)) were evaluated predicting area. Monthly determined by meteorological (precipitation, air temperature, Standardised Precipitation Evapotranspiration Index (SPEI)) factors (ETc, NDVI, LUCC). Landsat satellite image classification 2000–2020 analysed side-by-side with (SPEI) 9 12-month time scales. With evaluation six input variables algorithms, found that RF performed best when using SPEI-9 index, R2 = 0.88, RMSE 1.37, LCCC 0.95, PRD 1331.4 test samples. Furthermore, performance model constructed 9-month scale SPEI (SPEI-9) an variable (MLSPEI-9) depended seasonal variations, average relative errors up 0.62 spring minimum 0.12 summer. Overall, provides valuable insights into effectiveness different demonstrating right inputs can lead remarkable increase 13.89%. These findings have implications future research prediction zones demonstrate power advancing scientific understanding natural systems.