Paul Switzer: Reconstructing climate trends using proxy data
Prof. Paul Switzer (Stanford, Dept. of Statistics) will give a seminar in the lunch area, 8th floor N.H. Abel's House at 14:15 October 20th.
Reconstructing climate trends using proxy data
Concurrent time series commonly arise in environmental applications where the goal is to extract their shared time trend information. An application that we are studying involves time series of tree ring data that reflect to a greater or lesser degree their shared regional climate time trend. The common time trend extracted from these tree ring time series can be calibrated to instrumented weather time series, and the resulting calibration is then used to infer pre-instrumental weather time trends from the much longer sequence of tree ring data. The commonly used method to extract shared trend information from environmental time series has been PCA, principal component analysis. We consider an alternative method MAF, maximum autocorrelation factors [MAF] and demonstrate optimality properties of MAF in the context of signal+noise time series models. MAF is compared to PCA both theoretically and through examples, as well as in the application to tree ring time series in western US. We also illustrate the calibration to instrumented weather data and obtain reconstructions of regional annual temperature time series for the last millennium, together with a quantification of statistical uncertainty.