Toward Multi-proxy Reconstructions of Hydroclimate over the Contiguous United States from A.D. 500 to Present

Open Access
Brouillette, Daniel Joseph
Graduate Program:
Master of Science
Document Type:
Master Thesis
Date of Defense:
June 10, 2015
Committee Members:
  • Michael Mann, Thesis Advisor
  • drought
  • climate variability
  • hydroclimate
  • climate
  • reconstruction
  • United States
  • Holocene
  • PDSI
  • proxy
For the reconstruction of two targets, summer drought (by way of the Palmer Drought Severity Index (PDSI)) and winter precipitation, gridded instrumental data were calibrated with differing combinations of proxy data that comprised a network. This proxy network consisted of 797 records derived from multiple indicators intended to represent both high- and low-frequency variability (time-scales of less than and greater than 20 years, respectively). Three experiments, performed for both targets, modified the Regularized Expectation Maximization (RegEM) algorithm to yield high-skill reconstructions that depict low-frequency variability going back farther in time. Experiment 1 involved the RegEM algorithm regularized by ridge regression and using in the calibration only tree-ring records, which represented a much expanded data-set compared to previous studies. Experiment 2 introduced a new hybrid RegEM method in which the same tree-ring-only proxy data-set as in Experiment 1 was used in the calibration for two sub-reconstructions -- one regularized using ridge regression and the other regularized using truncated total least-squares (TTLS) regression – that were added together to make a total reconstruction. Experiment 3 used the same new hybrid method but introduced the low-frequency part of the proxy network to the calibration for the TTLS sub-reconstruction. Experiment 4 used a new regionalized version of the Experiment 3 hybrid method. It was found that, back to 1700, reconstructions arising from the RegEM-ridge (Exp. 1) method exhibited the greatest and most widespread skill for summer PDSI while also minimizing variance loss and was, thus, preferred. However, for reconstructions of that same target that extended farther back in time, the regionalized hybrid RegEM method in which low-frequency records are additionally used in the calibration (Exp. 4) was preferred because it produced reconstructions that depicted low-frequency variability going back in time with moderate to high cross-validation skill though not as high as that realized in some previous work. For winter precipitation, reconstructive skill was much more limited, restricted to the western part of the reconstruction domain. In that part of the domain, a hybrid method was preferred for similar reasons as it was for the longer reconstructions of summer PDSI.