Naslov
Using artificial neural networks (ANN) and self organizing maps (SOM) for better understanding of recent benthic foraminifera distribution

Autori
Borčić, Adriana ; Bogner, Danijela ; Popadić, Siniša

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
FORAMS 2010, International Symposium on Foraminifera, Abstracts Volume with Program / Organizing Committee in Bonn - Bon : Rheinische Friedrich-Wilhelms-Universität Bonn, 2010, 60-60

Skup
FORAMS 2010, International Symposium on Foraminifera

Mjesto i datum
Bonn, Njemačka, 05.09.2010. - 10.09.2010

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
benthic foraminifera ; neural networks ; self organizing maps

Sažetak
Data set used in ANN-SOM approach has been obtained from 48 sediment samples taken from the River Cetina estuary (eastern coast of the Adriatic Sea). The sediment cores (up to 12 cm long) were collected at two stations (P1 and P2) in front of the Cetina River mouth over the period of one year (four sampling campaigns: May, August, November 2003 and April 2004). As ANN work with numerical data sets only, a quantitative analysis of benthic foraminifera was performed on the fraction >63 µm. All the parameters have been scaled down using normalization methods which ensured that they all have the same impact on the ANN. As opposed to standard data analysis and low dimensional statistical observations ANN are able to comprehend today’s modern environments with numerous characteristics (sediment type, organic matter, etc.) in a more complex manner by correlating all the parameters of the data set. This was the core idea for this experimental approach. The data set was used to prepare two record sets (two tables) that could be used for cluster forming analysis. Five tests were performed to determine the clusters in two record sets using Self Organizing Maps (SOMs). In the first test a relative abundance of foraminifera per sample was used (the first record set) which pointed out dominant species of foraminifera. Three clusters were formed. The first and the biggest cluster gathered foraminiferas that occurred rarely on both stations. The second cluster contained foraminiferas with moderate occurrence (Bulimina aculeata, Cribroelphidium decipiens, Cibicides sp., Rosalina bradyi, Ammonia inflata and Miliolinella subrotunda). In the third cluster dominant species were found. The third cluster contained sub-clusters which separated dominant species on station P1 (Elphidium punctatum, Quinqueloculina sp., Q. seminula, Q. bosciana, Globorotalia sp., Globigerina sp. and Ammonia tepida) and dominant species on station P2 (Pseudoparrella exigua, Haynesina depressula and Ammonia tepida). Relative abundance of species, granulometric parameters, organic matter, carbonates, foraminiferal density and relative number of planktonic species were used (the second record set) in the remaining 4 tests which were done by excluding certain parameters from the analysis under suspicion that they had greater influence in cluster forming than the other parameters. The second test was done using the whole second record set but it was suspected that the foraminiferal density and the number of planktonic species were too concrete for the analysis. The third test included the second record set with all the parameters except foraminiferal density and relative number of planktonic species. Generally, the SOM map was divided among the two stations probably because of the parameters that were characteristic for each station (organic matter, carbonates, etc.). Although station P2 was deeper than the station P1, no data regarding depth were included in this analysis. The most distinctive cluster formed in this test contained data from the first three measurements on the station P1. Sub-cluster structure was also noticeable in this cluster between different sediment layers and measuring seasons on that station. Further analysis of this cluster showed that it could be described with lower values of silt, organic matter, clay, sorting, skewness and curtosis, and higher values of gravel, sand, carbonates and mean size. It is interesting how the cluster was clearly formed under influence of all these parameters. The last two tests were done additionally only for better understanding of the cluster forming by analyzing separately the relative abundance of species and granulometric parameters. The test results showed that, indeed, clusters and sub-clusters were formed in a way so that they could be described by common parameter values. Different graphical visualizations of the clusters were able to show how each parameter in the record sets participated in forming certain clusters. These tools could be better utilized if bigger data sets were available where the full potential of this approach would come to the fore.

Izvorni jezik
Engleski

Znanstvena područja
Biologija

POVEZANOST RADA