Funds are provided to improve the output from climate field reconstructions (CFRs) by using reduced space optimal interpolation (RSOI), which requires climate proxies to be locally calibrated as opposed to most current approaches which use with non-local calibration.
The researchers will analyze the currently employed non-local multivariate CFR methods and test RSOI-derived CFRs using locally-calibrated proxies. Results will be compared and contrasted both empirically and theoretically and will be used to reduce the uncertainty in estimates of millennial climate variability and produce better estimates of the spatial and temporal variability of climate.
The research has broad appeal to the wider science and policy making communities in that it is attempting to improve the quantitative nature of climate reconstructions.
This project investigated the effectiveness of multiple statistical methods that are used to reconstruct global temperature fields, or maps, on an annual basis spanning the last one to two thousand years. These kinds of reconstructions are known as climate field reconstructions (CFRs) and are based on estimates of the relationships between known temperature fields measured with thermometers and natural recorders of past climate known as climate proxies (e.g. tree rings, ice cores, cave formations, lake sediments, etc.) during the period of overlap between these two sources of information. CFRs are important for characterizing patterns of temperature change over decades to centuries, time scales that are not well represented by the short instrumental record that is only 100-150 years in duration. Our project was the first to comprehensively investigate how well CFR methods performed in terms of their ability to reconstruct the spatial patterns, or local features, of temperature changes in the past. We showed that all of the methods currently used for these purposes perform similarly and all contain important uncertainties in the spatial patterns that they reconstruct (see Figure 1 for an example of how four methods performed in a synthetic experiment that quantified the local correlation coefficients between derived reconstructions and a known model target). We also explored new statistical techniques to compare the patterns in CFRs to more rigorously elucidate the differences and similarities between them. An important result of our research is that methods appear to perform best where many climate proxies exist. Although this result may sound intuitive, it was not necessarily expected based on the methodologies that are used. This insight motivated us to derive regional reconstructions for regions of North America and Europe. Our work collectively described important successes and limitations associated with current state-of-the-art CFR methodologies, particularly with regard to their spatial performance, and has helped elucidate how our techniques need to be improved. This work has been broadly published in the scientific literature across a range of disciplinary specialties including climate and statistics journals. It has also engaged computer scientists in the machine learning and data mining communities. The work has been presented in multiple professional and public settings and the codes and data for many of our results have been provided through online websites, including archival of reconstruction results at the NOAA World Data Center for Paleoclimatology.
