Abstract
Ishim-Tobol River Basin, Northern Kazakhstan In this study, an ensemble reconstruction of the June–July streamflow from the Ishim-Tobor River was carried out using random forest (RF), K-nearest neighbor (KNN) and multiple linear regression (MLR) models. The reliability of the ensemble reconstruction was verified by a comparison with other regional reconstructions and historical records. A correlation analysis and vapor fluxes were applied to visualize the significant influence of atmospheric circulation on the study area. The cumulative distribution functions (CDFs) examined the distribution of the high (low) flows highlighted by the reconstruction. Our study analyzes the application of machine learning algorithms and a traditional MLR model to hydrological reconstruction. The single model reconstruction contained information and results on streamflow variability were not sufficient. Consequently, we integrated the three models into the ensemble reconstruction. The extended streamflow record reveals the basin's hydrological changes over the past 229 years. From 1788–2016, the reconstructed streamflow was perennially below the mean value, which indicates more prominent drought and water deficit conditions within the basin. This phenomenon was significantly influenced by water vapor transport from the North Atlantic and Arctic Oceans. If future climate scenarios lead to drought in the basin, then surface water demand will not be satisfied for 7 out of 10 years.