Jn. Le et Nj. Shackleton, RECONSTRUCTING PALEOENVIRONMENT BY TRANSFER-FUNCTION - MODEL EVALUATION WITH SIMULATED DATA, Marine micropaleontology, 24(2), 1994, pp. 187-199
This study examines Imbrie and Kipp's transfer function method (IKM) o
f estimating past sea surface temperature (SST) using simulated biolog
ical species abundance data (Imbrie and Kipp, 1971). We test the effec
ts on SST estimation of (1) the number of factors in calibration, (2)
regression types, (3) counting errors, (4) calibration ranges, and (5)
sub-surface species. When the number of factors is too small, the res
iduals of SST estimates show highly cyclic patterns in full range cali
bration, and the IKM largely overestimates SST at low temperature rang
e and underestimates SST at high temperature range. The simulations al
so demonstrate that curvilinear equations are more sensitive to noise
associated with faunal counting than linear equations because of the p
ower and crossproduct terms, although curvilinear equations usually ac
hieve a higher accuracy in full range calibrations. The use of a minor
factor may greatly increase accuracy in certain regions by accounting
for local ecological phenomena. In general, the IKM, if used with cau
tion, is a robust working method as demonstrated by its wide usage tha
nks to the restraint imposed by regression methodology. The experiment
s have also shown that regional linear calibrations achieve higher acc
uracy in the relevant region by overcoming the inadequacy of full rang
e linear equations and the sensitivity of curvilinear equations, provi
ded that the region is sufficient to fully cover past variations in te
rms of both species abundances and the SST to be estimated. This study
also provides a method to evaluate the contribution of individual spe
cies to SST estimates in the IKM. One may also quantitatively evaluate
the effect on SST estimation of the inclusion of those species that r
espond to environmental variables other than SST in an equation which
is aimed to estimate SST. In such an equation, the variations in other
environmental variables will be apparently manifested as changes in S
ST, the amount of which depends on: (1) the weight of those species on
the equation, and (2) the degree of non-analog condition in the sampl
e for which SST is to be estimated.