Evaluating Flood Model Performance: A Multi-Criteria Approach to Assessing Effectiveness During the 2023 Auckland Anniversary Weekend Floods

Abstract

Hydrological-hydraulic models remain the predominant feature of flood inundation modelling around the world. However, it is difficult to apply these traditional methods to evaluate the efficacy of flood modelling given that the actual flooding process is extremely ephemeral, and therefore, it is difficult to evaluate which areas have been inundated in a flood. Therefore, approaches are needed to assess the areas inundated during flooding to test the efficacy of hydrological-hydraulic models. This thesis uses a multi-criteria approach to attempt to fill this gap. This research incorporates high-resolution synthetic aperture radar (SAR) imagery and geolocated social media data that serve as a real-time validation tool for flood events. These methods are tested against the accuracy and validity of Tonkin & Taylor’s (T&T) Te Ararata Stream sub-catchment hydrological-hydraulic model during the Auckland Anniversary Weekend flooding in 2023. The T&T model is designed to simulate the extent and depth of inundation across the sub-catchment during a 1- in-100-year flood event for existing developments. Flooded areas confirmed by SAR and social media data were then combined with the broader T&T model to assess socio-economic vulnerability to flooding in the Te-Ararata Stream sub-catchment. The K-Means Nearest Neighbours classifier applied to the SAR imagery produced the most accurate output. SAR and social media data were integrated with T&T’s model to evaluate the effectiveness of the current flood model in forecasting flood events in the Te Ararata Stream sub-catchment. The analysis revealed that the T&T Te Ararata Stream flood model missed capturing three per cent of the total subcatchment area that experienced observed flooding during the Auckland Anniversary event. The Index of Multiple Deprivation (2018) explored deprivation levels between different socio-economic mesh blocks. The Te Ararata Stream sub-catchment encompasses highly to extremely deprived mesh blocks. A total of 53 per cent of predicted and observed flooding occurs in extremely deprived mesh blocks. This study demonstrates that using a multi-criteria approach could be considered valuable to flood modelling research, particularly the testing of the accuracy of flood hazard maps. However, hydrological-hydraulic models and SAR are often susceptible to topographic and human errors. These errors must be mitigated to produce optimal outcomes for flood inundation research.