Multi-model approach of data-driven flood forecasting with error correction for large river basins

• Published in: Hydrological Sciences Journal
• Main Author: Lim F.H.; Lee W.-K.; Osman S.; Lee A.S.P.; Khor W.S.; Ruslan N.H.; Ghazali N.H.M.
• Format: Article
• Language: English
• Published: Taylor and Francis Ltd. 2022
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85131575203&doi=10.1080%2f02626667.2022.2064754&partnerID=40&md5=3adf2407386f5b53cfa6fa4ee835fccd

Four data-driven hydrological flood forecasting methods are applied at 20 locations in Pahang River basin, Malaysia (area 30 000 km2). Models are calibrated and validated using historical monsoon flood data. To improve real-time forecast accuracy with 48-h lead time, continuous error correction is applied. An analysis of model performance above the alert water level shows that key forecast points along the main reach are best predicted using a stage regression method, whereas the upstream-most stations are best modelled using rainfall-stage correlation. The unit hydrograph method and Sugawara’s tank model perform well in the intermediate tributaries. Contrary to applying a single model to multiple points of interest or an ensemble model which requires evaluation of multiple models during operation, the multi-model approach allows the practical use of only the best-performing primary or secondary models at different points of interest within a large river basin to produce a reliable overall forecast with equal lead time. © 2022 Drainage and Irrigation Department, Malaysia.