Bridging Past and Future Climate across Paleoclimatic Reconstructions, Observations, and Models: A Hydroclimate Case Study

摘要

Potential biases in tree-ring reconstructed Palmer drought severity index (PDSI) are evaluated using Thornthwaite (TH), Penman-Monteith (PM), and self-calibrating Penman-Monteith (SC) PDSI in three diverse regions of the United States and tree-ring chronologies from the North American drought atlas (NADA). Minimal differences are found between the three PDSI reconstructions and all compare favorably to independently reconstructed Thornthwaite-based PDSI from the NADA. Reconstructions are bridged with model-derived PDSI_TH and PDSI_PM, which both closely track modeled soil moisture (near surface and full column) during the twentieth century. Differences between modeled moisture-balance metrics only emerge in twenty-first-century projections. These differences confirm the tendency of PDSI_TH to overestimate drying when temperatures exceed the range of the normalization interval; the more physical accounting of PDSI_PM compares well with modeled soil moisture in the projection interval. Remaining regional differences in the secular behavior of projected soil moisture and PDSI_PM are interpreted in terms of underlying physical processes and temporal sampling. Results demonstrate the continued utility of PDSI as a metric of surface moisture balance while additionally providing two recommendations for future work: 1) PDSI_PM (or similar moisture-balance metrics) compare well to modeled soil moisture and are an appropriate means of representing soil-moisture balance in model simulations and 2) although PDSI_PMis more physically appropriate than PDSI_TH, the latter metric does not bias tree-ring reconstructions of past hydroclimate variability and, as such, reconstructions targeting PDSI_TH can be used with confidence in data-model comparisons. These recommendations and the collective results of this study thus provide a framework for comparing hydroclimate variability within paleoclimatic, observational, and modeled data.