Bias correcting discharge simulations from the GEOGloWS global hydrologic model

Abstract

The Group on Earth Observations (GEO) Global Water Sustainability (GEOGloWS) hydrologic model provides global river discharge hindcasts and daily forecasts at approximately one million subbasins worldwide. The model is meant to sustainably provide discharge data during emergency situations and to underdeveloped countries which do not have sufficient local capacity. The primary model error is biased flow magnitudes which reduce the usefulness of the results. We applied a revised implementation of the SABER bias correction method to correct GEOGloWS model results. SABER uses a combination of watershed clustering with machine learning, geospatial analysis, and statistics to generalize bias patterns in gauged basins so they can also be applied to ungauged basins. We validated the bias corrected data created using the improved SABER method at 12,965 gauges globally and showed that this method reduced the mean error at 90% of gauges. We present an analysis of the improved SABER method using several metrics including mean error, root mean squared error, and Kling Gupta Efficiency. We found that the GEOGloWS model is usually biased high but our results indicate a reduction in the bias of the GEOGloWS model worldwide. We evaluate the varied performance of the bias correction procedure and significance of the improvements which vary based on stream order the watershed classifications derived in our analysis. We provide guidance on the use of bias corrected global data to local scale applications and discuss implications for the GEOGloWS model in the future.