Assimilating Citizen-Based Observations from Low-Cost Sensors in Hydrological Models to Improve Flood Prediction
Abstract:The main goal of this study is to demonstrate how integration of citizen-based observations coming from low-cost sensors (having variable uncertainty and intermittent characteristics) into hydrological models can be used to improve flood prediction.
The methodology is applied in the Brue basin, located in the South-West part of UK. In order to estimate the response of the catchment to a given flood event, a conceptual hydrological model is implemented. The measured precipitation values are used as perfect forecast input in the hydrological models. Then, a Kalman filter is implemented and adapted to account for asynchronous streamflow observations coming at irregular time steps having random uncertainty. Synthetic streamflow values are used in this study due to the fact that citizen-based observations are not available.
The results show how streamflow observations having variable uncertainty can improve the flood prediction. In particular, increasing the number of observations from low-cost sensors within two model time steps can improve the model accuracy leading to a better flood forecast. Observations uncertainty influences the model accuracy more than the irregular moments in which the streamflow observations are assimilated into the hydrological model. This study is part of the FP7 European Project WeSenseIt Citizen Water Observatory (www.http://wesenseit.eu/).