Date of Award

2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Complex Systems and Data Science

First Advisor

Chris Danforth

Abstract

High-resolution, bias-corrected climate data is necessary for climate impact studies and modeling efforts at local scales. General circulation models (GCMs) provide important information about historical and future larger-scale climate trends, but their spatial resolution is too coarse to investigate localized effects of climate processes. Additionally, raw GCM output is characterized by some degree of bias. Two post-processing procedures known as downscaling and bias-correction are typically applied to raw climate model output prior to its use in further modeling applications. Downscaling is the process in which data at a coarse spatial scale is transformed to a fine spatial scale. Bias-correction refers to a collection of methods in which climate model output is adjusted such that its statistical properties (e.g. mean, variance, and potentially higher moments) resemble those of observations in a common climatological period. Bias-correction is a challenge, due to relatively short calibration and long future time periods and potential spatial misalignment issues between griddedclimate model output and observed data. Issues that warrant further research are 1) spatially-coherent bias-correction, 2), processing of extremes, 3) temporally coherent bias-correction, and 4) balancing the bias-correction of future model output with the preservation of the climate change signal. Performing spatially-coherent bias-correction is particularly difficult, as model and observed data must be present in the same location where bias-correction is applied. Depending on the type of observed data used, this may not be the case. Extremes are challenging to represent accurately during bias-correction, because extreme values in both observed and model data are highly variable, limited, and there is greater uncertainty regarding their correction. Finally, very few bias-correction methods explicitly correct temporal dependence structures of model output. However, it is important that the temporal dependence of model data resembles that of observed data, as climate variability is closely linked to temporal dependence. In this body of work, I developed methodological workflows to generate high-resolution climate data products in which 1) bias-correction is carried out in a spatially-coherent manner, and 2) precipitation extremes are accurately represented. I also created a new, two-step bias-correction approach in which the temporal dependence and distributional properties of model output are corrected. This method allows for sensible bias-correction in both historical and future time periods and minimizes distortion to the future climate change signal.

Language

en

Number of Pages

347 p.

Share

COinS