Showcasing the Potential of Crowd-sourced Observations for Flood Model Calibration

Floods are one the costliest natural disasters, having caused global economic losses worth over USD 51 million and >6000 fatalities just in 2020. Hydrodynamic modelling and forecasting of flood inundation requires distributed observations of flood depth and extent to enable effective evaluation and to minimize uncertainties. Given the decline of in situ hydrological monitoring networks, Earth Observation (EO) has emerged as a valuable tool for model calibration and evaluation in data scarce regions, as it provides synoptic observations of flood variables. However, low temporal frequencies and the (currently) instantaneous nature of EO, still limits the ability to characterize fast moving floods. The concurrent rise of smartphones, social media, and internet access has recently led to the emerging discipline of citizen sensing in hydrology, which has the potential to complement real-time EO and in situ flood observations. Despite this, methods to effectively utilise crowd-sourced flood observations to quantitatively assess model performance are yet to be developed. In this study the potential of crowd-sourced flood observations for hydraulic model evaluation is demonstrated for the first time. The channel roughness for the hydraulic model LISFLOOD-FP was calibrated using just 32 distributed high-water marks and wrack marks collected by the community and provided by the Clarence Valley Council for the 2013 flood event. Since the timings of acquisition of these data points were unknown, it was assumed that these provide information on the peak flow. Maximum model simulated and observed water levels were thus compared at observation locations for each model realization, and errors were quantified through the root mean squared error (RMSE) and the mean percentage difference (MPD), respectively. Peak flow information was also extracted from the 11 available hydrometric gauges along the Clarence River and used to constrain the roughness parameter, to enable the quantification of value addition from the citizen sensed observations. Identical calibrated parameter values were obtained for both data types resulting in a mean RMSE value of ∼50 cm for peak flow simulation across all gauges. Outcomes from this study demonstrate the utility of uncertain crowd-sourced flood observations for hydraulic flood model calibration in ungauged catchments.