A new framework for interregional comparisons of biogeochemical models across the Arctic Marginal Ice Zone

High-latitude marine food webs are fundamentally shaped by magnitude, timing, and composition of phytoplankton blooms. In the Arctic, the growth environment for phytoplankton is rapidly changing due to the loss of seasonal sea ice in the marginal ice zone. Understanding the response of phytoplankton to changes in sea ice is therefore crucial for estimating potential future changes in the Arctic ecosystem. Coupled physical-biogeochemical, three-dimensional models are useful tools to understand ecosystem dynamics under given environmental conditions. However, these models are often developed and calibrated for specific regions, not necessarily accounting for regional differences in hydrodynamics or biogeochemistry. In addition, they are costly in terms of computing resources and time, complicating their use for interregional multi-model comparisons of, e.g. spring bloom dynamics. Here, we present a new cost-effective ‘sandbox’ framework for interregional comparisons of biogeochemical models for the case of two Arctic shelf seas: the Bering and Barents Seas. Our sandbox consists of a flow-following water-column model mimicking the biogeochemical components of two models developed for the Bering Sea: the Biology-Ice-Ocean Modeling and Assimilation System (BIOMAS) and a nutrient-phytoplankton-zooplankton-detritus (NPZD) model developed based on observations from the Bering Ecosystem Study/Bering Sea Integrated Ecosystem Research Program (BEST/BSIERP). Our sandbox uses physical fields simulated by the three-dimensional models BIOMAS and SINMOD (with the latter developed for the Barents Sea). We compare the two biogeochemical models in both Bering Sea and Barents Sea during the spring bloom period, and discuss both regional differences and differences in model dynamics. In addition, we provide a comparison with high-resolution, season-spanning glider data from the Barents Sea obtained within the Arctic PRIZE project to assess the models’ strengths and limitations, and to discuss potential future improvements.