Enhanced Net Community Production With Sea Ice Loss in the Western Arctic Ocean Uncovered by Machine-Learning-Based Mapping

Abstract

In the Arctic Ocean (AO), net community production ($NCP$) has displayed spatially heterogeneous responses to sea ice reduction and associated environmental changes. Using a random forest machine learning model trained with >42,000 in situ measurements and concurrent, collocated environmental predictors, we reconstructed 19 years of 8-day, 6-km $NCP$ maps. During 2015–2021, the integrated $NCP$ between late-May and early-September (${}_{\mathrm{int}}NCP$) over the western AO was $10.95\pm 3.30\text{Tg}C$ per year, with interannual variations positively tracking open water area. While the relationship between ${}_{\mathrm{int}}NCP$ and open water area was quasi-linear at high latitudes, strong nonlinearity was detected on the inflow shelf. The nonlinearity highlights that the ${}_{\mathrm{int}}NCP$ increase resulted from area gain could be compounded by sea-ice loss induced ecosystem adjustments. Additional retrospective analysis for 2003–2014 suggests a potential long-term increase of export production and efficiency in the western AO with sea ice loss.