A Resampling Approach for Correcting Systematic Spatiotemporal Biases for Multiple Variables in a Changing Climate

Systematic biases in climate model simulations are commonly addressed using univariate bias correction algorithms that involve matching of mean, variance, and quantiles. These approaches work well for a single variable and location and effectively mimic the observed temporal structure in the corrected series. The intervariable, interspace, and high- or low-frequency temporal dependencies that characterize observed hydrological records are often left untouched and lead to substantial biases in applications such as catchment modeling where their correct representation is critical. In the approach presented here, changes in the dependence attributes are ascertained by resampling of the historical ranks into what these might resemble in the future. The proposed approach is not limited in terms of the number of variables, grid points in space, and the time scale considered. Most importantly, it maintains the shift in dependence and other attributes between the current and the future climate as ascertained by a climate model. The approach is illustrated using daily time series of temperature, precipitation, relative humidity, and wind speed simulated by a regional climate model at 8,910 grid points over Australia. Spatial, temporal, and cross-variable dependence attributes of the corrected simulations at daily and aggregated time scales are compared against quantile mapping and substantial improvements in performance identified. Resampling of corrected ranks offers a very simple, flexible, and effective general purpose multivariate, multitime, and multilocation bias correction alternative for current and future climate. As the approach works in three dimensions, space, time, and variables, it is denoted as 3DBC, or three-dimensional bias correction. Plain Language Summary This article follows on from the considerable work our group and others have done to define approaches for correcting systematic biases in climate model simulations of the future. In the course of our investigation we realized that part of the problem in existing bias correction alternatives was the overreliance on the model simulations, specifically challenging when the models were unstable and simulated unrealistic values for extreme hydrological states of interest. As an alternative, we propose here an approach that relies on relevant statistical attributes of such simulations instead of the full simulations themselves. The proposed resampling strategy simulates future sequences with such attributes, thereby maintaining the hydrological nature of the sequences that are derived. The approach used is a modification of the popular Schaake Shuffle but tailored to work in the context of bias correction here. We demonstrate that the proposed approach is not limited in terms of the number of variables, grid points in space, and the time scale considered. Most importantly, it maintains the shift in dependence and other attributes between the current and the future climate as ascertained using a climate model.