An Inverse-Problem Approach to Detect Outliers in Rainfall Measurements of Ground Gauges for Robust Reservoir Flood Control Operation

This study proposes an approach to detect and remove systematic outliers in rainfall measurement of ground gauges in the watershed of a reservoir. This method utilizes a conceptual rainfall-runoff model with the downstream observed reservoir inflow to supervise the detection and cleanse of upstream rain measurements. The analysis is based on a nonlinear optimization with the objective designed as only removing outliers which leads to relevant improvement in runoff simulation to satisfy required precision. The minimum acceptable precision in hydrological simulation is specified based on the objective of reservoir flood control operation. This method can be utilized to purify historical rain data for hydrological studies, as well as to estimate the current floodwater detained in the watershed for more robust flood management in real-time. A realistic case study along with synthetically designed experiments are demonstrated to verify the effectiveness of the proposed methods.