A comparative survey between cascade correlation neural network (CCNN) and feedforward neural network (FFNN) machine...

Area of Science:

• Environmental Science
• Hydrology
• Machine Learning

Background:

• Accurate suspended sediment concentration (SSC) prediction is vital for water resource management, infrastructure design, and ecological health.
• Sheonath basin, India, faces challenges in managing aquatic resources due to sediment transport dynamics.

Purpose of the Study:

• To compare the efficacy of Cascade Correlation Neural Network (CCNN) and Feedforward Neural Network (FFNN) for predicting daily SSC.
• To identify optimal input variable combinations for SSC prediction models.

Main Methods:

• Daily SSC and discharge data from 2010-2015 for Simga and Jondhara stations were utilized.
• CCNN and FFNN models were developed and evaluated using statistical indices (N_ES, RMSE, WI, LM) and graphical methods.
• Nine input combinations with varying lag-times for discharge (Q_t-n) and SSC (S_t-n) were tested.

Main Results:

• The CCNN4 model, using four lagged SSC inputs, achieved the best performance at both stations.
• For Jondhara Station, CCNN4 yielded RMSE=95.02 mg/l, N_ES=0.662, WI=0.890, and LM=0.668.
• For Simga Station, CCNN4 achieved RMSE=53.71 mg/l, N_ES=0.785, WI=0.936, and LM=0.788.

Conclusions:

• CCNN models demonstrated superior performance over FFNN for daily SSC prediction in the Sheonath basin.
• CCNN exhibits significant potential for hydrological forecasting, particularly in scenarios with complex sediment transport relationships.