Calibration and Evaluation of a Flood Forecasting Model (Modified Tank Model) for Kelantan River Basin
Ho, Kwee Hong (2001) Calibration and Evaluation of a Flood Forecasting Model (Modified Tank Model) for Kelantan River Basin. Masters thesis, Universiti Putra Malaysia.
A distributed lumped conceptual flood forecasting model, namely Modified Tank Model was calibrated in this study for the Kelantan River Basin (12056 km2). Six hours rainfall and flood level data were collected from DID Data Bank and compiled as an input to Modified Tank: Model. Autoregressive corrections were implemented to improve the simulated flood level at Guillemard Bridge (forecasting station). Statistical method and objective functions were applied to evaluate the simulation and forecasting capability of the Modified Tank Model. Four years of flood data (1990, 1991, 1992 and 1994) were used to calibrate the Modified Tank Model and the performance of the model was verified by using 1998 data. A set of tank coefficients that suit tank configuration selected for Kelantan River Basin were determined by trial and error calibrations. Flood levels at Guillemard Bridge were simulated with actual measured catchment rainfall and the Mean Absolute Error (MAE) was found to be 0.59 m (7.8%) and Rl > 0.81. The Modified Tank: Model was found to be able to simulate and forecast the rising limb of flood hydrograph as well as the runoff peak for 6hr, 12hr, 18hr and 24hr lead time forecasts. Good correlation (R2 > 0.97) and average absolute error of 0.16 m were found for the 6hr lead time forecast (with error adjustment module). While average absolute errors of 0.2 m (2.6%), 0.24 m (3.1%) and 0.28 m (3.8%) were obtained for the 12hr, 18hr and 24hr forecasts with their R2 within the range of 0.96 and 0.97. The accuracy of water level forecast depends on the accuracy of the future rainfall forecast. In this study, two assumptions on rainfall quantities were made in order to evaluate the forecasting capability of the Modified Tank Model in actual forecasting operation. First assumption (Case A), assuming similar rainfall persists for the next 24 hours gives a range of errors from 0.26 m - 0.36 m, 0.36 m - 0.49 m and 0.45 m - 0.59 m with respect to 12hr, 18hr and 24hr lead time forecast while errors of the second assumption (Case B), assuming no rainfall for the next 24 hours were found in the range of 0.23 m - 0.35 m, 0.36 m - 0.56 m and 0.49 m - 0.76 m. Both cases show good correlation (R2 > 0.92) established for 12hr lead time forecast. Meanwhile, R2 were found in the range of 0.86 - 0.88 (18hr lead time forecast) and 0.73 - 0.84 (24hr lead time forecast) for both cases. Verification (using 1998 flood data) results indicated that, 0.51 m, 0.12 m, 0.16 m, 0.2 m, 0.23 m were found as the simulation error, 6hr, 12hr, 18hr and 24hr lead time forecast errors, respectively. Good correlation with R2 greater than 0.840, 0.983, 0.981, 0.975 and 0.967 were obtained with respect to the simulation, 6hr, 12hr, 18hr and 24hr forecasts. The flood hydrograph analysis (approximately 3 weeks of analysis period) showed that all errors increased slightly as compared to the overall simulation and forecast (approximately 2 months). Catchment water balance of the Modified Tank Model has been assessed with satisfactory results.
Repository Staff Only: Edit item detail