Comparing traditional hydrological forecasting models with CatBoost algorithm: insights from CMIP6 climate scenarios

ABSTRACT

Hydrological prediction is crucial for managing water resources, and innovations like machine learning (ML) present an opportunity to enhance predictive modeling capabilities. The aim of this study is to compare the usage of ML algorithms, such as CatBoost, with traditional techniques such as ridge regression, support vector machines (SVMs), and gene expression programming (GEP) in climate projection. In order to assess the accuracy of the best models, statistical measures such as root mean squared error (RMSE), mean absolute error (MAE), and Nash–Sutcliffe efficiency (NSE) were used. The investigation found that CatBoost was superior to conventional models in the testing period, with RMSE 3.78 m3/s, MAE 2.613 m3/s, Kling–Gupta efficiency (KGE) 0.650, root mean square error to standard deviation ratio (RSR) 0.611 and NSE 0.626. After it was proven that the best-performing model is CatBoost, future projections according to the NorESM2-MM scenarios were calculated using this model. Climate projections are based on simulations from the Coupled Model Intercomparison Project Phase 6 model, utilizing shared socioeconomic pathway (SSP) scenarios. The results show that SSP3-7.0 and SSP5-8.5 scenarios indicate an increasing trend between 2015 and 2100, while SSP1-2.6 and SSP2-4.5 expect a balancing tendency. This suggests that climate change has little effect on the measuring station and its basin and that the flow is increasing positively.