Continental Shelf Research 26 (2006) 2454–2468 Multivariate geostatistics for the predictive modelling of the surficial sand distribution in shelf seas Els Verfaillie a , Vera Van Lancker a , Marc Van Meirvenne b a Renard Centre of Marine Geology (RCMG), Ghent University, Krijgslaan 281, S8, 9000 Gent, Belgium b Department of Soil Management and Soil Care, Ghent University, Coupure 653, 9000 Gent, Belgium Received 22 December 2005; received in revised form 13 July 2006; accepted 26 July 2006 Available online 18 September 2006 Abstract Multivariate geostatistics have been used to obtain a detailed and high-quality map of the median grain-size distribution of the sand fraction at the Belgian Continental Shelf. Sandbanks and swales are the dominant geomorphological features and impose a high-spatial seafloor variability. Interpolation over complex seafloors is difficult and as such various models were investigated. In this paper, linear regression and ordinary kriging (OK) were used and compared with kriging with an external drift (KED) that makes use of secondary information to assist in the interpolation. KED proved to be the best technique since a linear correlation was found between the median grain-size and the bathymetry. The resulting map is more realistic and separates clearly the sediment distribution over the sandbanks from the swales. Both techniques were also compared with a simple linear regression of the median grain-size against the bathymetry. An independent validation showed that the linear regression yielded the largest average prediction error (almost twice as large as with KED). Unlike most static sedimentological maps, our approach allows for defining grain-size classes that can be adapted according to the needs of various applications. These relate mainly to the mapping of soft substrata habitats and of the most suitable aggregates for extraction. This information is highly valuable in a marine spatial planning context. r 2006 Elsevier Ltd. All rights reserved. Keywords: Multivariate geostatistics; Median grain-size; Bathymetry; Habitat mapping; Resource maps; Belgian continental shelf 1. Introduction Seabed habitats are subject to increasing pres- sures from human developments such as fisheries, aggregate extraction, dredging/dumping and wind- mill farms. In this context, the mapping of habitats and their prediction becomes crucial, both at the level of baseline studies as during the monitoring and decommitment phase. There is a difference between the physical (or abiotic) and the biological (or biotic) part of a seabed habitat (Fig. 1). However, if a full coverage map of the physical habitat is available and if the relations between the physical and the biological habitat are known, it is possible to create a full coverage map of the biological habitat. Nowadays there is an increasing demand for full coverage information. ‘Filling the gaps’ and ‘predictive modelling’ or the prediction of physical and biological information in areas with gaps, is a hot topic (e.g. ICES, 2005) in the ARTICLE IN PRESS www.elsevier.com/locate/csr 0278-4343/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.csr.2006.07.028 E-mail addresses: Els.Verfaillie@UGent.be (E. Verfaillie), Vera.VanLancker@UGent.be (V. Van Lancker), Marc.VanMeirvenne@UGent.be (M. Van Meirvenne).