Fig 1.
Generation of a homogeneous pseudo-population.
From one Scots Pine ring-width series, 1000 pseudo-trees are generated by adding 1000 different white noises to this particular series.
Fig 2.
Generation of a heterogeneous pseudo-population.
Two different signals (red and blue) are mixed along a linear gradient (rainbow coloured arrows) to obtain 1000 RWS with a gradual transition of underlying signals. As for PP1 individual-specific white noises are added.
Fig 3.
Long term signals of the pseudo-populations.
The signals behind PPs 2–4 showing no (PP2) or differing long-term trends (PP3-4). To better visualize the differing long-term trends these were multiplied by factor 3.
Table 1.
Detailed statistics of the PP based evaluations.
Fig 4.
Upper panel: grbarN plotted against along the population gradient for both population margins (red and blue lines). The blue line suggests a slightly higher grbarN along the gradient. Lower panel: HCA dendrogram suggests defining only one responder chronology, however with weak indications of sub-populations, probably caused by noise similarity. HCA labels indicate to which cluster each RWS belongs (here each belonging to cluster 1), while their colours relate to their position within the known gradient. As no gradient exists within PP1, the colours are mixed randomly.
Fig 5.
(for a higher resolved image we refer to supplement S1 Fig). Upper left panel: Loadings of the PCA over RWS coloured according to the PCGA gradient. Axis labels refer to PC relative importance. Three upper right panels: Detected gradients plotted against known gradients for PCGA, GR, and TR, respectively. Headers of these panels reflect the correlation coefficient between detected and true gradient. Mid left: grbarN plotted against along the population gradient for both population margins (red and blue lines). Both grbarN strongly suggest a population gradient as being well above
. Mid right: HCA dendrogram suggested defining two responder chronologies which mostly correctly split RWS along the known gradient. Lower left: PCGA responder chronologies (coloured curves) show a minimum correlation of 0.74 among each other. Two lower right panels: RWS signal correlations (small dots coloured according to PCGA gradient) show a clear relationship with PCGA gradient. PCGA (coloured ‘◊’) and HCA (black ‘X’) responder chronologies show equal signal correlations of which the marginal responder chronologies express higher correlations than the population master (grey ‘+’).
Fig 6.
For detailed explanations we refer to the caption of Fig 5.
Table 2.
Evaluation statistics for the Alaskan data.
Fig 7.
PCGA plots for the Alaskan sites.
PCA loadings coloured according to PCGA gradients for the four Alaskan sites. At each site PCA loadings suggest ecological gradients supported by comparably high explained variances on the first two PCs (values are given for each PC).
Fig 8.
Upper left: HCA dendrogram suggests selection of four responder chronologies. Note, that cluster four only consists of five specimens. Upper right: grbarN clearly indicates signal strength enhancement at the margins of the population. Mid left: PCGA extreme responder chronologies show weak correlations with each other. Mid right: PCGA and HCA responder chronology signal correlations are comparably strong: see also Table 2. Lower left: Same as mid-left but here for the analyses where the minimum sample size was adjusted to match HCA minimum sample size. Lower right: Same as mid right but here for the analyses where the minimum sample size was adjusted to match HCA minimum sample size.