Mapping benthic biodiversity using georeferenced environmental data and predictive modeling

• Published in: Marine biodiversity 2019-02, Vol.49 (1), p.131-146
• Main Authors: Peterson, Anneliis, Herkül, Kristjan
• Format: Article
• Language: English
• Subjects: Additives; Algorithms; Animal Systematics/Taxonomy/Biogeography; Aquatic plants; Benthos; Biodiversity; Biomass; Biomedical and Life Sciences; Climate change; Decision trees; Ecological function; Ecosystems; Environmental management; Environmental protection; Freshwater & Marine Ecology; Geographic information systems; Geographical distribution; Independent variables; Life Sciences; Macroinvertebrates; Macrophytes; Mapping; Marine ecosystems; Marine environment; Mathematical models; Model accuracy; Modelling; Ocean floor; Original Paper; Plant Systematics/Taxonomy/Biogeography; Prediction models; Predictions; Regression analysis; Remote sensing; Salinity; Sampling; Scuba & skin diving; Sediments; Spatial distribution; Species richness; Stabilizing; Variables; Zoobenthos
• ISSN: 1867-1616; 1867-1624
• DOI: 10.1007/s12526-017-0765-5

Biodiversity is critical for maintaining and stabilizing ecosystem processes. There is a need for high-resolution biodiversity maps that cover large sea areas in order to address ecological questions related to biodiversity-ecosystem functioning relationships and to provide data for marine environmental protection and management decisions. However, traditional sampling-point-wise field work is not suitable for covering extensive areas in high detail. Spatial predictive modeling using biodiversity data from sampling points and georeferenced environmental data layers covering the whole study area is a potential way to create biodiversity maps for large spatial extents. Random forest (RF), generalized additive models (GAM), and boosted regression trees (BRT) were used in this study to produce benthic (macroinvertebrates, macrophytes) biodiversity maps in the northern Baltic Sea. Environmental raster layers (wave exposure, salinity, temperature, etc.) were used as independent variables in the models to predict the spatial distribution of species richness. A validation dataset containing data that was not included in model calibration was used to compare the prediction accuracy of the models. Each model was also evaluated visually to check for possible modeling artifacts that are not revealed by mathematical validation. All three models proved to have high predictive ability. RF and BRT predictions had higher correlations with validation data and lower mean absolute error than those of GAM. Both mathematically and visually, the predictions by RF and BRT were very similar. Depth and seabed sediments were the most influential abiotic variables in predicting the spatial patterns of biodiversity.