Using a Random Forest model to predict the distribution of benthic biomass in the Bering Sea

Marine benthic invertebrates provide a critical resource base for several higher trophic level consumers, such as seabirds and marine mammals. Exploring the distribution and movements of higher level consumers requires maps of benthic resources at appropriately large scales.

Logistic constraints render it improbable that a spatially continuous map of benthic biomass can be provided by sampling alone, and predictive modeling offers a valuable alternative to create such maps. Here, we describe how to use an algorithmic model that overcomes many weaknesses of traditional data models to predict benthic biomass at large spatial scales.

We use a decision-tree modeling approach (RandomForest) to link benthic biomass to chlorophyll a concentration, sea surface temperature, sea ice cover, depth, distance to coastline, sea bottom temperature and sea bottom salinity, and present a digital map of predicted benthic biomass across the Bering Sea.