TimeLagged Recurrent Network for Forecasting River Suspended Sediment Load in Southern Taiwan

Y.-M. Wang, S. Traore, and W.-G. Chung (Taiwan)


Neural network, performance, hydro-meteorological data, river sediments load, water quality management


Forecasting suspended sediment load is crucial for river water quality continuous management. This paper investigates the accuracy of a time-lagged recurrent network (TLRN) for forecasting suspended sediment load (SSL) in Kaoping River basin located in Southern Taiwan. For this study, two major stations of Kaoping River basin; Liukwei and Lao-Nung are taken into account where important data have been collected between years 1984 to 2005. The ability of TLRN in SSL estimation is assessed by using hydro-meteorological data such as rainfall, water level and discharge as input sets. The network accuracy was evaluated with the goodness-of-fit measures of normalized mean square error, mean absolute error and coefficient of correlation between estimated and observed data. The results showed that the TLRN has a good performance in SSL forecasting when using only water discharge variable as the network input for both stations. However, Liukwei station presented a better statistical performance than Lao-Nung. It was found that among the input variables considered in this study, water discharge is the most effective for sediment load forecasting in the two stations. Finally, TLRN can be successfully employed in Southern Taiwan for modeling river sedimentation if the other factors related to SSL are apprehended.

Important Links:

Go Back