Streamflow Bias Correction for Climate Change Impact Studies: Harmless Correction or Wrecking Ball?

Projections of the hydrologic impacts of climate change rely on a modeling chain that includes estimates of future greenhouse gas emissions, global climate models, and hydrologic models. The resulting streamflow time series are used in turn as input to impact studies. While these flows can sometimes be used directly in these impact studies, many applications require additional post-processing to remove model errors. Water resources models and regulation studies are a prime example of this type of application. These models rely on specific flows and reservoir levels to trigger reservoir releases and diversions and do not function well if the unregulated streamflow inputs are significantly biased in time and/or amount. This post-processing step is typically referred to as bias-correction, even though this step corrects not just the mean but the entire distribution of flows. Various quantile-mapping approaches have been developed that adjust the modeled flows to match a reference distribution for some historic period. Simulations of future flows are then post-processed using this same mapping to remove hydrologic model errors. These streamflow bias-correction methods have received far less scrutiny than the downscaling and bias-correction methods that are used for climate model output, mostly because they are less widely used. However, some of these methods introduce large artifacts in the resulting flow series, in some cases severely distorting the climate change signal that is present in future flows. In this presentation, we discuss our experience with streamflow bias-correction methods as part of a climate change impact study in the Columbia River basin in the Pacific Northwest region of the United States. To support this discussion, we present a novel way to assess whether a streamflow bias-correction method is merely a harmless correction or is more akin to taking a wrecking ball to the climate change signal.