Author(s): Sumit Sinha; Simon Waller

Linked Author(s): Sumit Sinha

Keywords: Hydrological model; Hydrodynamic model; GPU

Abstract: Damages caused by floods are all pervasive and impacts almost every corner of the world. Global loss attributed to floods are typically in billions of dollars (Bloschl, 2017). Floods also accounted for the highest percentage (45%), by virtue of occurrence, of natural disasters between 1995 and 2015 causing severe damages to life and property. This disastrous impact of a flooding event has been further exacerbated by climate change, there is a consensus among researchers about the increasing frequency of peak flows and rainfall events on catchment level and broader scale as a result of climate change. Given all this there is a strong need for robust numerical models that can aid in decision making and be a crucial component in designing critical flood warning systems. Ideally, it is desired that these numerical models are physically based, easy to run and computationally inexpensive. These demands are often inversely related i.e. physically based models are often computationally expensive on the other hand easier to run models are hidden in layers of abstraction distancing it from the fundamentals of flow. However, the significant advances in desktop computing as witnessed by emergence of multi core central processing unit (CPU) and graphical processing unit (GPU) enabled machines, has led to more frequent application of physics-based model for flood risk assessment on catchment and broader scale. This research presents an integrated modelling approach for flood risk assessment in the Thames river basin. The basin under consideration drains an area of 10000 km2 encompassing both heavily urban and rural landscape overlaying a complex geology which is instrumented with multiple monitoring gauges providing long term stream flow record. Taking advantage of the widely available data in the basin under consideration, we first configure a fully distributed hydrological model at a coarse spatial grid resolution. The fully distributed hydrological model utilizes Muskingum-Cunge scheme for routing the stream flow through the catchment and accepts spatially distributed rainfall, temperature and potential evapotranspiration at daily time step. The input data is obtained from the Centre of Ecology and Hydrology datasets (Lewis et. al., 2014). The applied hydrological model is calibrated and validated using model performance metrics of Nash-Sutcliffe efficiency (NSE) and Kling Gupta efficiency (KGE). Adequately validated hydrological model is then used for spatiotemporal identification of peak flow events along the drainage network. These peak flow events are then used for higher-resolution flood inundation modelling in a reach wise fashion using two-dimensional hydrodynamic model JFlow based on dynamic shallow water equations. The numerical kernel of JFlow is based on finite volume methodology is extremely efficient and runs on GPU enabled machines. Conjunctive use of hydrological and hydrodynamic model helps in identifying the flood prone areas in the basin under consideration.

DOI: https://doi.org/10.3850/IAHR-39WC252171192022414

Year: 2022