DOI: 10.5176/2251-189X_SEES16.44
Authors: Pham Quy Giang, Kiyo Kurisu and Keisuke Hanaki
Abstract: Flood inundation models are powerful tools for simulating the spread of flooding water upon cities and agricultural fields, so that damages can be predicted and countermeasures can be proposed properly. However, in order to obtain accurate prediction, every model needs to be calibrated when applied to a specific study area, and the calibrated model needs to be tested by another process, which is often called validation. A wide variety of measures including correlation based and non-correlation based statistics have been recommended to assess the model accuracy through the pairwise comparison of model simulated values with observed values during calibration and validation periods. In this study, both correlation based and non-correlation based statistics, including Coefficient of determination (R2), Nash-Sutcliffe Simulation Efficiency (NSE), Index of Agreement, Percent Bias (PBIAS), Root Mean Square Error – Observation Standard Deviation Ratio (RSR), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) were applied for the MIKE Flood model with the case study of the Lower Ca River Basin in the North Central Vietnam. Observed and simulated data of river discharge and water level at six hydrological monitoring gauges were used for the calculation of the above seven statistics. The study found that, there is a strong agreement among correlation based statistics (R2, NSE, and IA) but this may not apply to the non-correlation ones. As such, at all gauges R2, NSE, and IA resulted in a high correlation between simulated data and observed data (all R2, NSE, and IA values close to 1), which indicates that model prediction is good. Meanwhile, for non-correlation based statistics, although RSR, RMSE and MAPE gave acceptable values for all gauges, it was found that the level of PBIAS at two gauges was unsatisfactory because the model largely underestimated river discharge data. The study also found that correlation based statistics are oversensitive to extreme values and are insensitive to additive and proportional differences between simulated data and observed data so that these statistics can indicate that model prediction is good, even when it is not. It is concluded that beside the use of correlation based statistics, a comprehensive accuracy assessment of flood inundation modeling should include non-correlation based measures.
Keywords: accuracy assessment, correlation, inundation simulation, non-correlation
Updating...