Short-term forecasting of spring freshet peak flow with the Generalized Additive model

• Published in: Journal of hydrology (Amsterdam) Vol. 612; p. 128089
• Main Authors: Dubos, Véronique, Hani, Ilias, Ouarda, Taha B.M.J., St-Hilaire, André
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2022
• Subjects: air temperature; Boreal hydrology; cold; Distributed hydrological model; drainage; Flood forecast; GAM; hydrologic models; linear models; meteorological data; Peak flow; Quebec; rain; risk; river flow; snow; snowmelt; spring; Spring runoff; stakeholders; uncertainty; Quebec; Peak flow; Spring runoff; GAM; Distributed hydrological model; Flood forecast; Boreal hydrology
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2022.128089

•Spring freshet peak flow can be estimated with few variables, easily available.•Generalized Additive Model (GAM) provided a reliable 24-h forecast of the peak flow value.•GAM model forecasts compared favorably to a more complex deterministic hydrological model.•The forecasted flow can be used as a decision support to launch emergency measures. In cold boreal regions, for rivers with small to medium-sized watersheds under natural hydrological regimes, the risk of spring flooding is determined by peak flow intensity rather than flood volume. Nonetheless, short-term forecasting of peak flow intensity is subject to a lot of uncertainty and depends largely on ongoing specific snowmelt conditions. This study proposes a simple operational model based on the Generalized Additive Model (GAM) to forecast short-term spring freshet peak flow. The model uses hydrological and meteorological data publicly available on a daily basis. The model was tested on five rivers in the Province of Québec (Canada) with drainage basins varying between 350 km2 and 1707 km2. The model results (forecasted peak flows) were compared to those obtained using the Generalized Linear Model (GLM) and a distributed deterministic hydrological model (Hydrotel) currently used for flow forecasting of several rivers in the Province. The peak flow was forecasted accurately with relatively few variables, mainly a combination of river flow and rate of flow increase a few days before peak flow, previous air temperature, rain accumulation and snow accumulation. Nonetheless, the best combinations of predictive variables were river-specific. The GAM model, using an automatic fitting and easily accessible daily data, can be implemented by any stakeholder.