Using the NWS flood level at each location, hits are defined as the percentage of time that the observed discharge exceeded flood level and the model's prediction exceeded that threshold. Misses are instances where the model's prediction is below the threshold. The ocurrences of flood events predicted by the model but where observed discharge does not exceed flood level are considered false alarms. View diagram here.
The spatial resolution is approximately 400m cell size and the estimates are available hourly for this analysis, and every 5 minutes on the real-time product. It started to be available since 2009. The product is radar-based only, without correction from rain gages.
Mean Absolute Error
Discharge observations and simulations were converted to stage values by using the gage rating curve, and their mean absolute errors (in feet) were obtained as: where and are the simulated and observed river stage (in feet) for each time step, respectively.
The model structure, used in the experiments, corresponds to the one reported in BAMS paper. A complete description of the model equations can be found here.
The spatial resolution is approximately 1km cell size and the estimates are available hourly. Includes rain gauge bias correction.
Nash Efficiency is obtained as: ,where S are model simulations and O are discharge observations.
The model is based on using a decomposition of landscape into hillslopes and channel links. The setup of the reference model uses a network derived from a 90-meter resolution DEM.
Normalized Root Mean Squared Error
Reflects how large are the average errors in the discharge simulations respect to the median observed discharge. The metric is obtained as: , where is the mean average observed discharge.
Percent bias in discharge volume is obtained as: ,where S are model simulations and O are discharge observations.
The model was forced with three different radar rainfall products (hourly Stage IV rainfall, IFC rainfall, and MRMS). The product is available since 2002. The model is forced every year with data from the warm season, starting April 1, and extended until December 1.
Stage IV rainfall
The spatial resolution is approximately 4km cell size and the estimates are available hourly. Includes rain gage bias correction.
Compares the difference in time between observed and simulated features of the hydrographs (i.e. rising, falling limbs, and top discharge value of individual events). It is obtained by finding the time shifting of the simulated series that would maximize the cross correlation with the observed series.