Validation of this multi-method composite reconstruction against the
published Kaufman et al. (2020) Temperature 12k curves
(doi:10.1038/s41597-020-0530-7),
bundled from the NOAA archive under reference_data/published/.
Both curves are anchored to 0 at 100 yr BP (NOAA archival convention) and
metrics computed on absolute anomalies so amplitude and offset
errors are visible — unlike r/CE on recentered series, which hide them.
Per-method and consensus comparison against the published curves. Both
anchored to 0 at 100 yr BP. n pts ages 0–12 ka common to both.
r: Pearson correlation of anchored medians.
bias / RMSE / maxΔ: of (mine − published), °C.
amp: std(mine)/std(pub) — 1.0 = matched amplitude.
midHol: mean over 5.5–6.5 ka — mine | published | paper Table 1.
12 ka: mean over 11.5–12 ka — mine | published.
spread: mean(q95−q05) mine/published — 1.0 = matched uncertainty.
flat: longest run of identical consecutive median values (clipping flag).
| Method | r | bias | RMSE | maxΔ | amp | midHol (m | p | T1) | 12 ka (m | p) | spread | flat |
|---|---|---|---|---|---|---|---|---|---|
| SCC | 0.982 | -0.085 | 0.102 | 0.194 | 1.01 | 0.35 | 0.49 | 0.50 | -0.88 | -0.77 | 0.94 | 1 |
| DCC | 0.990 | -0.032 | 0.054 | 0.125 | 1.00 | 0.43 | 0.50 | 0.50 | -0.79 | -0.77 | 0.92 | 1 |
| GAM | 0.965 | -0.101 | 0.121 | 0.278 | 1.11 | 0.31 | 0.45 | 0.44 | -0.81 | -0.70 | 0.92 | 1 |
| CPS | 0.994 | +0.040 | 0.126 | 0.312 | 1.02 | 1.11 | 1.09 | 1.08 | -3.25 | -3.36 | 1.04 | 1 |
| PaiCo | 0.987 | -0.028 | 0.069 | 0.179 | 1.14 | 0.40 | 0.42 | 0.42 | -0.87 | -0.72 | 1.02 | 1 |
| Consensus | 0.986 | -0.048 | 0.072 | 0.175 | 1.02 | 0.42 | 0.51 | 0.70 | -0.89 | -0.83 | 1.08 | 1 |
Reconstruction (coloured line + shaded 5–95% spread) vs published curve (dotted + grey shaded). X-axis: oldest age left, present right.
Pooled multi-method ensemble: 2,500 members. Age range in the comparison: 0–12,000 yr BP.