In Figure **F2**
we present the opal values (in weight percent) from 137.52 to 266.13 mcd at Site
1082, the interval analyzed by the two laboratories. The two records show the
same general pattern but show offsets in absolute numbers. OLin values tend to
be higher, averaging 23.6% (*N* = 50), whereas OPer values have an average
of 18.3% (*N* = 50), comparing the closest neighbors (<4 cm offset) in
the overlapping portion of the record.

Reproducibility
in the two procedures was estimated from both duplicate and replicate
determinations. The pooled precision was ±1.37% opal for duplicate
determinations of 23 OLin samples and ±0.6% opal for duplicate determinations
of 42 OPer samples. Replicate determinations of nine OLin samples with an
average content of ~15%–20% opal yielded a standard deviation of ±1.42% opal;
replicate measurements of three OPer samples with ~8%–15% opal gave a standard
deviation of ±0.55% opal. Table **T1**
and Figure **F3**
show the differences between the two techniques for selected OLin and OPer
samples (opal-poor and opal-rich samples). The error is larger for samples with
the lower opal content.

We
tested whether a regression of OPer data points vs. OLin data points was
sufficiently precise to transform OPer to OLin values and vice versa. Because
OLin and OPer opal measurements were not done on aliquots from the same samples,
we followed two approaches: the closest neighbor method and the linear
interpolation method. For the closest neighbor method, only those samples taken
at nearby depth intervals (<4 cm offset) were considered in the correlation (*N*
= 50). Alternatively, estimates of the OPer opal content for depth intervals
corresponding to OLin samples were obtained by linear interpolation from the
OPer values (*N* = 70) (Fig.** F4A,**
**F4B**). The data points outside the 95%
confidence interval were removed for the calculation of the regression. We found
that a linear regression had the highest correlation coefficient of several
possible transforms in both approaches (r^{2}
= 0.84 and r^{2}
= 0.86, respectively).

By
using the linear regression equation given by the linear interpolation method
(Fig. **F4B**),
OPer data were converted to OLin data. Subsequently, the estimated OPer data set
and the measured OLin data set were merged to obtain a more detailed and
extended opal record of Site 1082. The merged data set could then be compared to
the opal record of Site 1084, which was analyzed with only the OLin method.
Figure **F5A**
shows a similar pattern at both sites with a distinct opal maximum during the
late Pliocene and early Quaternary. Because of the location of Site 1084, close
to the Lüderitz upwelling cell, opal values were higher at this site than at
Site 1082.

DAI,
as estimated in Lange et al. (1999), yielded the same patterns at both sites
(Fig. **F5B,**
**F5C**). The diatom maximum is
characterized by high abundances of *Thalassiothrix* and other pelagic
species in addition to the *Chaetoceros* spores indicating coastal
upwelling (Lange et al., 1999).