Flood Simulations Using a Sensor Network and Support Vector Machine Model

This study aims to couple the support vector machine (SVM) model with a hydrometeorological wireless sensor network simulate different types of flood events in montane basin. The was tested mid-latitude basin Vydra Šumava Mountains, Central Europe, featuring complex physiography, high dynamics processes, and occurrence floods. is equipped operating headwaters along conventional long-term monitoring outlet. trained validated using hydrological observations from 2011 2021, performance assessed metrics such as R2, NSE, KGE, RMSE. run both hourly daily timesteps evaluate effect timestep aggregation. Model setup deployment utilized KNIME software platform, LibSVM library, Python packages. Sensitivity analysis performed determine optimal configuration SVR parameters (C, N, E). Among 125 simulation variants, an parameter identified that resulted improved better fit for peak flows. sensitivity demonstrated robustness model, variations yielded reasonable performances, NSE values ranging 0.791 0.873 year. Simulation results scenarios showed reliability reconstructing accurately captured trend fitting, event timing, peaks, volumes without significant errors. Performance generally higher timestep, mean metric R2 = 0.963 0.880, compared 0.913 0.820 all 12 scenarios. very good even rain-on-snow floods combined fast computation makes this promising approach applications.