This work proposes and develops a pipeline for constructing accurate surrogate models for future streamflow predictions of the HBV-SASK hydrological model. The pipeline consists of several components. 
First, a Bayesian filtering step, specifically a Particle Filter, is applied to sequentially improve knowledge of uncertain model parameters and internal states by incorporating observed data as it becomes available. 
Next, a Polynomial Chaos Expansion (PCE)-based surrogate is employed to propagate uncertainty from the model's stochastic parameters to its outputs. 
More precisely, the updated posterior distribution over the seven uncertain parameters is used to build an efficient PCE surrogate for future time-step streamflow predictions with quantified uncertainty.
To ensure numerical stability in the PCE construction, we use transport maps to transform the posterior distribution of the parameters into a standard Gaussian form. 
Finally, both the performance and accuracy of the surrogate will be compared with the original model, using both quantitative metrics and visual analysis.
The proposed pipeline and surrogate model facilitate simpler model evaluation and enable real-time applications such as flood forecasting, drought analysis, and other predictive tasks.