We describe an alternative method of climate field reconstruction and test it against an existing set of dendroclimatic reconstructions of summer drought patterns over the conterminous US back to AD 1700. The new reconstructions are based on a set of 483 drought-sensitive tree-ring chronologies available across the continental US. In contrast with the 'point-by-point' (PPR) local regression technique used previously, the tree-ring data were calibrated against the instrumental record of summer drought (June-August Palmer Drought Severity Index (PDSI)) based on application of the 'Regularized Expectation Maximization' ('RegEM') algorithm to relate large-scale patterns of variation in proxy and instrumental data over a common (twentieth century) interval. A screening procedure was first used to select an optimal subset of candidate tree-ring drought predictors, and the predictors (tree-ring data) and predictand (instrumental PDSI) were prewhitened prior to calibration, with serial correlation added back into the reconstruction at the end of the procedure. The PDSI field was separated into eight relatively homogenous regions of summer drought through a cluster analysis, and three distinct calibration schemes were investigated: (i) 'global' (i.e., entire conterminous US domain) proxy data calibrated against 'global' PDSI; (ii) regional proxy data calibrated against regional PDSI; and (iii) global proxy data calibrated against regional PDSI. The greatest cross-validated skill was evident for case (iii), suggesting the existence of useful non-local information in the tree-ring predictor set. Cross-validation results based on withheld late nineteenth/early twentieth-century instrumental data, as well as a regionally limited extension of cross-validation results back to the mid-nineteenth century based on long available instrumental series, indicate a modest improvement in reconstructive skill over the PPR approach. At the continental scale, the 1930s 'Dust Bowl' remains the most severe drought event since 1700 within the context of the estimated uncertainties, but more severe episodes may have occurred at regional scales in past centuries.
All Science Journal Classification (ASJC) codes
- Global and Planetary Change
- Earth-Surface Processes