Streamflow prediction in ungauged river basin using gene expression programming

Abstract

Hydrologic studies are facilitated by abundant and continuous records of streamflows and indirect peak discharge measurements. This serves as the basis for design of hydraulic structures, water resources planning and management, hydropower operation, hydrological disaster risk management as well as in assessing the effects of environmental changes. Precipitation data, temperature, humidity, wind speed are some of the pertinent meteorological data required for appropriate studies. The backbone of hydrologic data for this type of study is continuous records of streamflow gauges. However, where streams are ungauged, recourse has to be made to rainfall – runoff processes competent to simulate the flow scenarios in the catchments of interest. Other pertinent data required include geomorphologic and soil characteristics of the catchments as well as the land use and land cover. The recurrence flooding episodes and the need to have better insight to flow variability in the states of Kelantan and Terengganu (some parts of the east coastal region of Peninsular Malaysia) has been a pointer to the need for the development of models that can serve as tools for flow simulations in any catchment within the study area. The main objective of this study is therefore to predict river discharge in ungauged river basins in the study area. For this purpose, a set of multivariate equations are developed; using Genetic Expression Programing (GEP) model available in soft computing software GeneXProTools 4.0 using 4 – 7 explanatory variables. These are: Rainfall, area, perimeter, main stream length, slope, drainage density and curve number. Thus; available streamflow data along with other catchment characteristics from 15 gauged stations are used to prepare the flow duration curves (FDC). The predictable variables as Qext, Qmax, Q0.05, Q0.10, Q0.25, Q0.50, Q0.75, Q0.90, Q0.95, Qmin and Qmean were extracted from the FDC to develop the models. In order that the models may be reliably used for flow simulations in some other catchments within the study area, the accuracies of the models using standard statistical procedures such as; NSE, RMSE, R2 and goodness of fit from the software were measured for both calibrated and validated flows. These indicate very good performance.