Electronic Supplementary Material (ESM)
Transfer Function Performance
The 2nd component for WA-PLS did not significantly improve the reconstruction (% Change in RMSEP = +7%, p = 0.78), so only WA regression was used for the reconstructions. For the transfer function, DI-TP concentrations were correlated with observed TP concentrations (ESM Fig. 5A; R2 = 0.70). There was a strong bias in the residuals, indicating a tendency for the transfer function to under-predict TP at high concentrations (TP > 150 µg/l [log10 TP > 2.2]; ESM Fig. 5B). After cross-validation by boot-strapping, the R2 boot was reduced to 0.61 (ESM Fig. 5C) and showed the same bias towards under-prediction at high TP in the residuals (ESM Fig. 5D). The root mean squared error of prediction (RMSEP) for the transfer function was 0.30 log10 µg/l TP.
ESM Fig. 1 Core profiles for dry density, % organic matter, and % carbonate (shown for select sites with high carbonate content). Dashed lines indicate likely marker horizon for Euro-American settlement
ESM Fig. 2 Total 210Pb activity plots for sediment cores. Supported 210Pb indicated by dashed line. Error bars represent ± 1 sd
ESM Fig. 3 Sediment accumulation rates vs. sediment age as determined by 210Pb dating for each of the study lakes. Error bars represent ± 1 sd
ESM Fig. 4 Box-plots of square-root-transformed relative abundances of diatom species in historic (H; grey) and modern (M; white) sediment samples for the 32 lakes in this study. Species shown had a relative abundance ≥ 5% and occurred in at least two sections
ESM Fig. 5 (A) Diatom Inferred log10 TP (DI-TP) vs Observed log10 TP for the training set lakes. Dashed line represents a 1:1 line. (B) Residual plot of prediction error in observed log10 TP. (C) log10 DI-TP vs observed log10 TP after cross-validation. Dashed line represents a 1:1 line. (D) Residual plot of prediction error in observed log10 TP after cross-validation
1