Authors: Riley Chad Hales, Jim Nelson, Josh Ogden – Brigham Young University
Title: Real time bias correction post processing for large hydrologic models
Abstract: Recent hydrologic models cover larger spatial domains with higher spatial and temporal resolutions. Observed discharge is rarely well distributed spatially along the modeled basins or with consistent temporal coverage compared to the modeled resolutions. This presents serious challenges for calibration, data assimilation in forecasts, and evaluating model performance. SABER is an empirical postprocessor to reduce bias in forecast and hindcast results which extends the comparatively small amount of gauge data to all subunits in the model. All modeled basins are classified by k-means clustering using features from their uncorrected modeled flow duration curve (FDC) and physical properties such as soil composition, land use, and slope. Within each cluster, bias at each gauge is quantified on a monthly basis using the ratio of observed and simulated flows for corresponding exceedance probabilities (e.g. ratios of FDC). Basins which do not have gauges are corrected using bias profiles from gauges within the group. The most suitable intra-cluster gauge is determined using a decision tree that accounts for the spatial proximity of each gauged and ungauged basin, connectivity along the river network, and presence of dams or control structures. This method requires substantial preprocessing but is efficiently applied ad-hoc to correct forecast and hindcast flow at any basin. The method has implications for improving flow initializations and data assimilation in ungauged basins where traditional methods are not possible or computationally prohibitive.