Comparative Assessment of Linear and Non-Linear Tank Models for Runoff Predictions

This study assessed the applicability of both linear and nonlinear tank models to predict runoff based on the delay time from rainfall to runoff, applying them to severe rain events in a small river with several square kilometers in the Okinawa Island. Suitable parameters for the linear and non-linear tank models were firstly assessed by applying them to three rain event, utilizing the genetic algorithm (GA) for parameter optimization. This optimization highlighted that both the linear and non-linear tank models reproduced well time variations of the water depths. Then, run-off predictions were performed by using average optimized parameters for each rain. For the 10-min runoff prediction, the linear tank model predicted well the water levels while the non-linear tank model, sensitive to small changes in its exponential parameter, was not able to predict well the water levels using average coefficient that differed from the optimized parameters for each rainfall event.