Date of Award
Master of Science (MS)
Earth System Science & Policy
The climate of the Earth is changing, and is primarily a result of our rampant industrialization over the past two centuries. These changes have manifested themselves in many ways over the whole of the Earth’s surface and sub-systems, leading to the need to understand the changes and predict future outcomes. Coupled climate and general circulation - Earth system models (GCMs) allow for the analysis of dynamically active simulations over the whole of the planet, yet are limited by computational power. The model grids are coarse by design to perform within these computational constraints, which enables them to function and provide information at continental and larger scales, but which limit their ability to offer information for regional and local environments. Dynamical models created with higher resolutions allow for regional climate modeling yet are also limited by computational constraints and require detailed information to run. Statistical downscaling seeks to bridge the gap between coarse GCM grids by utilizing observational data and statistical models to remove the biases from the data at the local level. There have been several types of statistical methods applied to this task over many different regions with some success. The goal of this study is to utilize two methods in particular, bias-corrected spatial disaggregation (BCSD) and redundancy analysis (RDA), to downscale maximum and minimum temperature, as well as precipitation, for the Northern Great Plains (NGP) region. These methods are calibrated over the period 1950 – 1970 using a 1/8 degree gridded dataset for 17 GCMs, then applied to a verification period (1970 – 1999) and compared to observations over that period to assess the downscaled models skill in capturing local NGP variability. These methods are also applied to future model runs forced via the representative concentration pathways (RCPs) low end (2.6), median (4.5) and high end (8.5) 21st Century forcings, which provides possible outlooks for local stakeholders over the coming decades. It is found that BCSD does well in downscaling temperature and precipitation, as well as their various metrics. RDA provides more mixed success, with good skill demonstrated for temperatures but a strong wet bias in precipitation. It is noted, however, that RDA yielded better correlations to the observations. Future scenarios show broad ranges of projected outcomes that, as expected, increase with increasing forcing, though temperature shows stronger changes than precipitation, and BCSD exhibits higher sensitivity than RDA. Future research may help further constrain the results of these downscaling methods, particularly RDA, by adopting further bias correction to the results.
Coburn, Jacob, "Statistical Downscaling For The Northern Great Plains: A Comparison Of Bias Correction And Redundancy Analysis" (2015). Theses and Dissertations. 1756.