Application of Acoustic Tomography Data in Short-Term Forecasting of Streamflow Using Combinatorial GMDH Algorithm (CGA)

Short-term forecasting of streamflow is one of the most important goals in water resources management and flood control. However, one of the problems that researchers always face in this type of prediction is the Lack of an accurate and high time resolution database. The Fluvial Acoustic Tomography (FAT) is an innovative method of data gathering which has both high accuracy and high-resolution time. Therefore, by using the data collected from this technology with a suitable forecast model, accurate short-term streamflow forecasting can be achieved. In this research, the effect of FAT data on short-term streamflow forecasting by Combinatorial GMDH Algorithm (CGA) has been investigated and compared with one obtained from the Rating Curve method. The k-fold cross-validation criterion has been used to prevent over-fitting. The results showed that the FAT data increases the accuracy of short-term forecasting. As an example, the Nash-Sutcliffe coefficient (ENS) for the 1, 6, 12, 24, 48, and 72 hours forecast were 0.98, 0.96, 0.94, 0.88, 0.73, and 0.54, respectively. While these values for the Rating Curve ones were 0.97, 0.84, 0.61, 0.27, 0.12, and 0.11, respectively.