Paleoclimate Sampling as a Sensor Placement Problem

Abstract

This study formulates the design of optimal observing networks for past surface climate conditions as the solution to a data assimilation problem, given a realistic proxy system model (PSM), paleoclimate observational uncertainties, and a network of current and proposed observing sites. We illustrate the method with the design of optimal networks of coral δ 18 O records, chosen amongst candidate sites, and used to jointly infer sea surface temperature (SST) and sea surface salinity (SSS) fields from the Community Climate System Model 4.0 Last Millennium simulation over the [1850, 2005] period. We show that an existing paleo-observing network accounts for ∼ 20% of the SST variance, and that adding 25 to 100 optimal pseudo-coral sites would boost this fraction to 35 to 52%. Characterizing the SST variance alone, or jointly with the SSS, leads to similar optimal networks, which justifies using coral δ 18 O records for SST reconstructions. In contrast, the network design for reconstructing SSS alone is fundamentally different, emphasizing the hydroclimatic centers of action of the El Niño Southern Oscillation. In all cases, network design depends strongly on the amplitude of the observational error, so replicates may be more beneficial than the exploration of new sites; these replicates tend to be chosen where proxies are already informative of the large-scale climate field(s). Finally, we discuss extensions to other types of paleoclimatic observations, and outline a path to operationalization to reducing uncertainty in our knowledge of past climates? Similar questions were ex-54 plored by Bradley (1996), who considered an existing proxy network and examined how 55 well it predicted global mean annual temperature series. Evans et al. (1998) explored the 56 3 more general problem of optimal proxy sampling for minimizing theoretical error in re-57 constructed fields, but for the more specific estimation of the SST field. To that end, the 58 authors used a reduced-space optimal interpolation (OI) based reconstruction framework 59 (Gandin 1965; Kaplan et al. 1997), and assumed hypothetical SST observations made with 60 random error might be available at a predetermined, quasi-realistic, coral-atoll-like sam-61 pling network. For small paleoclimate observational errors, they found that relatively 62 small networks were able to resolve the most important large scale patterns of SST vari-63 ance (e.g. Evans et al. 1998, Fig. 4, 5 and 6) and reduce the analysis error by up to 30% 64 regionally. In the presence of relatively large paleoclimate observational errors, resam-65 pling (replication) was often preferred over sampling additional locations. Their itera-66 tive, "next best" approach did not, however, guarantee that the resulting proxy network 67 will be optimal; nor did they consider that the most common paleoclimatic observation 68 made in coral aragonite is oxygen isotopic composition (δ 18 O), which depends not only 69 on temperature but also seawater δ 18 O at time of calcification (Evans et al. 1998). 70 Recent...

North Atlantic-Fennoscandian Holocene climate trends and mechanisms

“…For the other eighteen sites, it was not possible to use ensembles created in Bacon for one or more of the following reasons: 1) the chronology was developed by varve counting, and is not suitable for Bacon, 2) the original chronological data are not available, or 3) depths corresponding to past temperature estimates are not available, disallowing the application of a new model ensembles to the temperature estimates. For these eighteen sites, we estimate age uncertainty using the "Banded Age Model" (BAM) algorithm of Comboul et al (2015). This algorithm was designed to create ageuncertain ensembles for layer-counted chronologies, and is thus well-suited for the varve records in the dataset, but can also be applied to simulate uncertainty for other records.…”

A data assimilation approach to last millennium temperature field reconstruction using a limited high-sensitivity proxy network

“…Interestingly, the temperature response in 1601 CE is relatively small over much of central Europe and reconstruction uncertainty is relatively low, which suggests this feature may be a robust feature of the post-eruption climate anomaly. In addition to supporting analysis of reconstructed climate features, these uncertainty estimates can help identify regions that would benefit from increased network density, as in Comboul et al (2015). In particular, we observe that northern North America and eastern Siberia would benefit from the development of new millennial-length temperature-sensitive tree-ring records.…”

The revolution of crossdating in marine palaeoecology and palaeoclimatology

“…The quantification and characterization of these uncertainties coupled with the general improvement in our understanding of the forcing mechanisms that drive the coupled ocean -atmosphere climate system will ultimately facilitate the continued improvement of the individual GCMs, enhancing the ability of the numerical models to provide robust simulations of likely future climate change. Numerical models can also be used to identify and guide selection of sites where new chronologies likely have maximum palaeoclimatic significance [49,50]. Finally, crossdated marine chronologies can constrain quasi/multi-decadal climate variability over the past few centuries to millennia [9].…”