Improved spectral comparisons of paleoclimate models and observations via proxy system modeling: Implications for multi-decadal variability

TL;DR: In this paper, a forward proxy modeling approach coupled with an isotope-enabled GCM is proposed to disentangle the various contributions to signals embedded in ice cores, speleothem calcite, coral aragonite, tree-ring width, and tree cellulose, and conclude that the paleoclimate record may exhibit larger low-frequency variability than GCMs currently simulate, indicative of incomplete physics and/or forcings.

About: This article is published in Earth and Planetary Science Letters.The article was published on 2017-10-15 and is currently open access. It has received 39 citations till now. The article focuses on the topics: Climate model.

Summary

2.1. GCM & PSM-Generated Pseudoproxies

• Each proxy type employs its own unique PSM.
• The complicated nature of proxy data (e.g. chronological uncertainties and impacts on phasing) precludes point-to-point comparisons of time series, and thus there is a strong case for comparing simulated proxy to the observations in the frequency domain.

3. Case Studies

• Various approaches including downscaling or bias correction can help to minimize such problems, or paleoclimate data can be aggregated to match GCM grid cell size.
• For each proxy type, the authors attempt to answer whether the mismatch arises from a lack of low-frequency variability simulated by the GCM SPEEDY-IER, or from a data-model comparison strategy problem.
• For completeness, the authors report absolute variance for all case studies and the PAGES2k data in SI Section S3.

3.1. Spectral Fingerprinting of Proxy Systems

• As a first pass, the authors forced each PSM with white noise climate inputs to assess the impact of proxy system processes alone on the shape of the spectra.
• For ice cores, speleothems, and tree ring widths, the white noise +.
• For all proxy types, the spectra revert to the shape of the white input climate signal on decadal and longer timescales.
• Under different PSM formulations these spectra could change significantly, and this non-unicity proves a large source of uncertainty.

3.2.1. Corals

• Shows that the corals are generally strong SST proxies (or, possibly, that the GCM completely underplays salinity variability).
• Testing the effects of parametric uncertainty for the corals provides an example of how PSMs can be used to inform data-model comparison.
• More interestingly, discrepancies exist between the simulated and observed power spectrum on decadal to centennial timescales.
• Further, if the authors instead evaluate both in terms of absolute variance, the Palmyra record exhibits larger σ at the decadal band as compared to the PSM-simulated data (SI Section S3).
• While the PSM-generated pseudo-coral captures interannual SST variability similar to observations, the PSM seems not to account for the larger variance in the observations on longer timescales, and this discrepancy remains even when uncertainties in the coral's sensitivity to salinity and δ 18 O S W are taken into account.

3.2.2. Ice Cores

• On decadal to centennial timescales, differences in the observed vs. simulated spectral slopes are more modest than for interannual, but three of the records tend to increasingly diverge at low frequencies (see Fig. 3 ).
• 18 O PRECIP vs. the observed ice core values exhibit some agreement on multi-decadal frequencies, but the model does not simulate comparable variance in the observations on longer (>centennial) timescales (see Fig. 3 ).
• This suggests that neither the GCM, the water isotope physics in the GCM, nor the PSM can account for observed low frequency variability.

3.2.3. Speleothems

• The speleothem PSM highlights the fact that on interannual to decadal timescales, the authors can essentially obtain a β value in agreement with observations simply as a function of the karst parameters.
• On longer timescales, the simulated spectra tend to flatten while the observed spectra continue to show increased lowfrequency variance, potentially indicative of climate processes resulting in a spectrum similar to what the authors would expect from a power law system (see Fig. 5 ).

3.2.5. Tree Ring Width

• Aggressive detrending methods tend to remove low frequency variability (demonstrated by Table 2 ).
• Table 2 also illustrates the RCS method is most conservative in maintaining low-frequency TRW variability.
• In general, using the same detrending method for both proxy and pseudoproxy is essential.

Figures

Figure 4: Ice Core Model-Data Comparison at NGRIP: effect of diffusion length (σ) on β. Estimated power spectral density for simulated (purple) vs. observed (black) δ18O of ice and the effect of varying diffusion length parameters. The grey shaded region spans experiments resampling the data 1000 times varying the diffusion length from 12 · σ to 2 · σ. In the case where σ = 0 (the royal blue line, δ18OPRECIP), GCM-simulated data is generally in agreement with the observations.

Figure 5: Speleothem Calcite Model-Data Comparison: the effect of Karst Transit Time (τ) on β. Simulated vs. Observed δ18O of speleothem calcite at Cascayunga Cave. PSD is plotted, simulated using an ensemble of plausible values (6 months to 5 years) for the karst transit time parameter, τ. The spectrum reddens as the transit time increases. The orange solid line indicates the mean β slope for the observations. Figure illustrates broadening disagreement between simulated and observed speleothem δ18O approaching decadal-centennial timescales.

Figure 8: Distribution of proxy βI , βD values, GCM+PSM vs. Observed, for the PAGES2k Phase 1 dataset: a. ice cores (blue, left), tree-ring width (green, center), and coral data (orange, right) from PAGES2k (lighter colors) and GCM/PSM (darker colors). (a, b., d., e.): temperature (red), precipitation (dark blue) and PSM ice cores (light blue, right); SPEEDY-IER temperature (red), precipitation (dark blue), and PSM tree-ring width (green, center); and (c., f.): CCSM4 sea-surface temperature (pink), sea-surface salinity (purple), and PSM coral data (orange, right).

Figure 1: Spectra and β values by proxy type and as a function of timescale. Estimated Power Spectral Density for a purely white noise input climate signal, leaving the effects of the proxy system only. Each PSM was forced with white noise climate input 10,000 times. The median spectrum is shown as the solid line, and shading represents the 95% confidence interval. a. Ice Core δ18O, b. Carbonates (Coral and Speleothem δ18O) c. TRW, Tree Ring Cellulose δ18O.

Table 2: βI and βD values for coral, ice core, speleothem, and tree ring cellulose sites, simulated vs. observed for all case studies and the PAGES2k Database. βI is calculated as the mean slope between 2 and 8 years (interannual), and βD is the mean slope between 20 and 200 years (decadal to centennial). Mean PAGES2k vs. PSM simulated overall β values for Coral, Ice Core, and Tree Ring PAGES2k sites with test statistics for difference-of-mean Mann-Whitney/Wilcoxon Rank-Sum tests are given alongside case studies. For TRW, β values were calculated for the four tree ring width sites listed in Table 1, simulated vs. observed, using the Modified Negative Exponential (negex) detrending method and simply calculating the slope after simulating the TRW using the forward model VS-Lite (no detrending). We report difference of means test statistics for the PAGES2k data only, as we have a full distribution of values for these data as opposed to the single-point case studies, where a difference of means test is not appropriate. See SI Table S6 for a comparison of six different detrending methods including RCS, Modified Negative Exponential (negex), Linear, Spline, Hugershoff.

Figure 2: Coral Aragonite Model-Data Comparison: simulated (pink) and measured (black) spectrum for Palmyra coral data, as well as the effects of altering α2 at Palmyra island alongside the climate inputs to the model (SST, SSS). Straight orange lines are the calculated mean slopes across both decadal to centennial (20-200, βD) and interannual (2-8, βI ) timescales. The black-dotted envelope represents the full range of outcomes given the perturbed salinity parameter space. Low frequency variance observed in the coral data (black dashed line) greatly exceeds that of the simulated pseudo coral data (pink), and the discrepancy grows with increasing timescale. Also shown: simulated sensitivity to salinity changes and potential parameter uncertainties in the PSM (α2), effects on β, (faint black dots), which does not help account for the model-data discrepancies. We removed the mean from all fields prior to computed the PSD.

Figure 7: Tree-Ring Width Model-Data Comparison: the effect of detrending method on β. For each site, the original measured TRW is plotted in black, along with the PSM (VS-Lite) inputs of temperature (red), precipitation (blue), and the simple VS-Lite generated pseudoproxy data (dark green dashed), which does not include detrending or growth-curve. The negative-exponential detrending method was applied to both the real proxy data and the VS-Lite generated pseudoproxy data (bright green). To estimate reasonable errors for simulated TRW, we took reported errors for the estimated growth parameters in 100 randomized possibilities (grey).

Table 1: List of sites forward modeled and evaluated in this study. By building a workflow connecting paleoclimate proxy data, a climate model, and proxy system models, we evaluate spectral scaling characteristics for five proxy types in four regional case studies. Corals, Ice Cores, Speleothems, Tree-Ring Width, Tree-Ring Cellulose. ∗ Note that the Palmyra coral record is continuous in the interval 1146-1464: we use the previously published, continuous part of the record in this study. This segment has been updated since (?).

Figure 6: Tree-Ring Cellulose Model-Data Comparison. Estimated Power Spectral Density for Simulated (green) vs. Observed (black) δ18O of tree ring cellulose at (a) Boibar, Pakistan and (b) Lhamcoka, Tibet. We removed the mean from all fields prior to computed the PSD.

Figure 3: Ice Core Model-Data Comparison. Estimated Power Spectral Density for Simulated vs. Observed δ18O of ice at four sites: (a) NGRIP, (b) Quelccaya, (c) Austfonna, (d) Lomonosovfonna. The PSD for each PSM-generated record is shown (red) alongside climate inputs of temperature (dark blue), precipitation (light blue), and δ18OP (purple). In all cases, the observed ice core record is plotted in black (dashed). Straight orange lines are the calculated mean slopes across both decadal to centennial (20-200, βD) and interannual (2-8, βI ) timescales. Diffusion in the ice core PSM causes higher-than-observed spectral slopes at high frequencies, and may suggest that the simulation of diffusion and compaction is overestimated. PSM-simulated ice cores do not capture the observed higher variance at low-frequencies. We removed the mean from all fields prior to computed the PSD.

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "Improved spectral comparisons of paleoclimate models and observations via proxy system modeling: implications for multi-decadal variability" ?

In this paper the authors bridge this gap via a forward modeling approach, coupled to an isotope-enabled GCM. The paper addresses the following questions: ( 1 ) do forward modeled “ pseudoproxies ” exhibit variability comparable to proxy data ? The authors apply their method to representative case studies, and parlay these insights into an analysis of the PAGES2k database ( ? ). The authors conclude that, specific to this set of PSMs and isotope-enabled model, the paleoclimate record may exhibit larger low-frequency variability than GCMs currently simulate, indicative of ∗Corresponding author Email addresses: sylvia 11 dee @ brown. The authors find that current proxy system models ( PSMs ) can help resolve model-data discrepancies on interannual to decadal timescales, but can not account for the mismatch in variance on multi-decadal to centennial timescales.