Predicting stream baseflow using genetic programing

Abstract

Developing reliable methods to estimate stream baseflow has been a subject of research over the past decades due to its importance in catchment response and sustainable watershed management (e.g. ground water recharge vs. extraction). Limitations and complexities of existing methods have been addressed by a number of researchers. For instance, physically based numerical models are complex, requiring substantial computational time and data which may not be always available. Artificial Intelligence (AI) tools such as Genetic Programming (GP) have been used widely to reduce the challenges associated with complex hydrological systems without losing the physical meanings. However, up to date, in the absence of complex numerical models, baseflow is frequently estimated using statistically derived empirical equations without significant physical insights. This study investigates the capability of GP in estimating baseflow for a small monitored semi-urban catchment (0.021 km2) located in Singapore. A Recursive Digital Filter (RDF) is first adopted to separate the baseflow from observed streamflow. GP is then used to derive an empirical equation to relate the filtered baseflow time series particularly with groundwater table fluctuations which are relatively easy to be measured and are physically related to baseflow generation. The equation is then validated with a longer time series of baseflow data from a groundwater numerical model. These results indicate that GP is an effective tool in determining baseflow.