Extended Data Fig. 10: Proxy network sensitivity tests.
From: Mid-latitude net precipitation decreased with Arctic warming during the Holocene
a–e, Scatterplots showing the relation between decadal mean temperature at the proxy locations versus the average of the entire 20° latitudinal zone using gridded instrumental CRU TS4.01 temperature observations15. In the instrumental dataset, the mean temperature at the proxy locations explain between 77% and 96% of the variance in the latitudinal bands. The spread in data represents the overall temperature trend over the twentieth century. f, Mid-Holocene minus preindustrial (MH − PI) temperature averaged for the proxy locations (y-axis) versus the latitudinal averages (x-axis) from 12 PMIP3 climate models (symbols) across five latitudinal bands (colours). The models suggest that the proxy network captures the same mid-Holocene temperature anomalies as the latitudinal averages. g, Using gridded CRU TS4.01 precipitation observations15 to test the representativeness of the hydroclimate proxy network and the influence of standardization on the magnitude and variability in the composite time series. The standardized (mean = 0; variance = ±1 s.d.) decadally averaged composite is shown for the entire latitudinal zone 30° N to 50° N (orange squares) and for the proxy locations (green diamonds), along with the latitudinal average in native units (purple circles). The standardized mean from the proxy locations closely tracks the variability of the standardized and native unit latitudinal averages. h, Testing the number of proxy records needed to characterize Holocene hydroclimate changes in the mid-latitudes. Proxy record composites with iteratively smaller sample sizes were correlated with the final mid-latitude composite. The solid and dashed lines show the strength of correlation (R and P values, respectively) between the mid-latitude hydroclimate composite based on the full dataset of 72 records and composites using a randomly selected subset of fewer records. The error envelope shows the 95% bootstrapped random sampling confidence intervals. Correlation with the final composite begins to plateau near 40 records, suggesting our sample size of 72 is sufficient to represent the region.
