Performance comparison between genetic programming and sediment rating curve for suspended sediment prediction

Abstract

The accurate prediction of suspended sediment flowing into reservoirs is very important during their design and operation stage. It is also of great significance in environmental engineering, hydroelectric equipment longevity, river aesthetics, pollution and channel navigability. Hence the development of methods that can accurately forecast the suspended sediment loads from historical water data sets and infrequently or non-sampled rivers will greatly help in the management of water resources. One of the widely used methods employing empirical models are rating curves. Therefore, in this study, 12 sediment rating curves (SRCs) were developed from 14 years’ historical data, one for each month of the year, to show the relationship between stream and suspended sediment flowing into Inanda Dam, Durban, South Africa. The developed monthly SRCs were evaluated using two performance assessment criteria, specifically RMSE and R². The performance of these SRC models was compared with those of the corresponding monthly developed genetic programming (GP) models for each month of the year. The results indicate that the predictions from the GP models are better than the predictions from the SRC models, especially in predicting large quantities of suspended sediment load during high streamflow such as during flood events.