Comparison of statistical and artificial neural network techniques for estimating past sea surface temperatures from planktonic foraminifer census data

Abstract

Abstract. We present the first detailed and rigorous comparison of six different computational techniques used to reconstruct sea surface temperatures (SST) from planktonic foraminifer census data. These include the Imbrie-Kipp transfer functions (IKTF), the modem analog technique (MAT), the modem analog technique with similarity index (SIMMAX), the revised analog method (RAM), and, for the first time, a set of back propagation artificial neural networks (ANN) trained on a large faunal data set, including a modification where geographical information was added among the input variables (ANND). By training the techniques on an identical database, we were able to explore the differences in SST reconstructions resulting solely from the use of different mathematical methods. The comparison indicates that while the IKTF technique consistently shows the worst performance, ANN and RAM perform slightly better than MAT and that the inclusion of the geographical information into the training database (SIMMAX and ANND) further improves the accuracy of modem SST estimates. However, when applied to an independent validation data set and an additional fossil data set, the results did not conform to this ranking. The largest differences in the reconstructed SST values occurred between groups of techniques with different approaches to SST reconstruction; that is, ANN and ANND produced SST reconstructions significantly different from those produced by RAM, SIMMAX, and MAT. The application of the various techniques to the validation data set, which allowed comparison of SST reconstructions with instrumental records, suggests that artificial neural networks might provide better paleo-SST estimates than the other techniques. IntroductionOne of the most remarkable consequences of Charles Lyell's uniformitarian principle for the field of paleoceanography has been the opportunity to use the relationship between the distribution of modem faunas and floras and present-day physical conditions in the ocean to reconstruct climatic variations in the Quaternary period. In addition, if this relationship were expressed in the form of a mathematical formula, past climatic variations could be quantified in standard physical scales and units. This appealing prospect was discovered early in paleoceanographical studies, and quantitative reconstruction of Quaternary climate change by means of fossil faunas has become a standard and routinely applied procedure.Some The third approach, using artificial neural networks (ANN), a branch of artificial intelligence, relies on the sole assumption that there, indeed, is a relationship between the distribution of modem faunas and the physical properties of the environment. ANNs have the ability to overcome problems of fuzzy and nonlinear relationships between sets of input and output variables. This computer-intensive approach is based on an algorithm that has the ability of autonomous "learning" of a relationship between two groups of numbers [14/assetman, 1989; Beale and Jackson, 1990]. Once trained, the ne...

Strengths and Weaknesses of Quantitative Climate Reconstructions Based on Late-Quaternary Biological Proxies

“…Although extremely different in their numerical computations and algorithms, artificial neural networks (ANN) have identical mathematical properties when applied to palaeoenvironmental reconstructions. Malmgren and Nordlund (1997), Malmgren et al (2001), Burrows et al (2005), and Kucera et al (2005) describe the use of ANN in marine sea-surface temperature reconstructions; Peyron et al (1998Peyron et al ( , 2000Peyron et al ( , 2005, Tarasov et al (1999aTarasov et al ( , 1999b, and Guiot et al (1996) provide examples of their use with terrestrial pollen data; and Racca et al (2001; discuss ANN applications in palaeolimnology. Telford and Birks (2005) show that in crossvalidation tests with a truly independent test-set, MAT and ANN perform poorly as both involve estimating smooth local functions.…”

Modeling Mediterranean Ocean climate of the Last Glacial Maximum

“…7 together with reconstructions derived from marine sediment cores. Temperatures have been estimated from assemblages of planktonic foraminifera with two different methods (Hayes et al, 2005): The revised analog method (RAM, Waelbroeck et al, 1998) and the artificial neural network technique (ANN, Malmgren et al, 2001). For annual mean temperatures, additional estimates derived from alkenones have been used (Lee, 2004, and references therein).…”

Predicting the global distribution of planktonic foraminifera using a dynamic ecosystem model

“…Shells of planktonic foraminifera extracted from marine sediments serve as an archive of chemical and physical signals that can be used to quantify past environmental conditions, such as temperature (e.g., Pflaumann et al, 1996;Malmgren et al, 2001), ocean stratification (e.g., Mulitza et al, 1997), atmospheric CO 2 concentration (Pearson and Palmer, 2000) and biological productivity (Kiefer, 1998). Past sea-surface temperatures can be estimated by either quantifying differences between modern and fossil species assemblages (e.g., CLIMAP, 1976;Pflaumann et al, 1996;Malmgren et al, 2001), or by analyzing the isotopic or trace-element composition of the calcite in the shell (e.g., Rohling and Cooke, 1999;Lea, 1999). In general, all estimation procedures are based on a correlation between modern environmental condition and assemblage composition or shell chemistry.…”