Water level prediction using soft computing techniques: A case study in the Malwathu Oya, Sri Lanka.

Area of Science:

• Hydrology
• Computational intelligence
• Environmental modeling

Background:

• Traditional hydrologic models are computationally intensive and data-dependent.
• Unavailability of crucial catchment data (soil, land use, etc.) limits simulation accuracy.
• Soft computing techniques offer efficient alternatives with reduced data requirements.

Purpose of the Study:

• To evaluate the computational capabilities and accuracy of Gradient Boosting Algorithms and Adaptive Network-based Fuzzy Inference System (ANFIS) for river flow simulation.
• To develop and compare prediction models for Malwathu Oya, Sri Lanka, using these soft computing techniques.
• To assess the performance of CatBoost against ANFIS and other gradient boosting variants.

Main Methods:

• Development of river flow prediction models using Gradient Boosting Algorithms (CatBoost, XGBoost, LightGBM) and ANFIS.
• Utilizing catchment rainfall as input for the simulation models.
• Comparison of simulated river flows with ground-measured data using metrics like R, Bias, NSE, MARE, KGE, and RMSE.

Main Results:

• Both CatBoost and ANFIS can effectively simulate river flows based on catchment rainfall.
• CatBoost exhibited a significant computational advantage over ANFIS.
• CatBoost achieved the highest correlation coefficient (0.9934) for the testing dataset, outperforming XGBoost (0.9283), LightGBM (0.9253), and Ensemble models (0.9109).

Conclusions:

• Gradient Boosting Algorithms, particularly CatBoost, show strong potential for accurate and computationally efficient river flow simulation.
• CatBoost offers a superior alternative to ANFIS for hydrologic modeling tasks.
• Further research is recommended to explore broader applications of these advanced algorithms in hydrology.