The Arctic is warming at approximately twice the global rate in response to anthropogenic climate change, resulting in disappearing sea ice, increased open water area, and a longer growing season (IPCC, 2013). This loss of sea ice has resulted in a 30% increase in annual net primary production (NPP) by Arctic Ocean phytoplankton between 1998 and 2012 (Arrigo and van Dijken, 2015). To quantify NPP, many algorithms require input of chlorophyll a (Chl a) concentration, which serves as a biomass proxy for phytoplankton. While satellites provide temporally and spatially extensive data, including Chl a, the standard global ocean color algorithms are prone to errors in Arctic Ocean waters due to higher than average phytoplankton pigment-packaging and chromophoric dissolved organic matter (CDOM) concentrations. Here, we evaluate retrievals of Chl a using existing ocean color algorithms, test and develop a new empirical ocean color algorithm for use in the Chukchi Sea, and evaluate the effect of using different satellite Chl a products as input to an NPP algorithm. Our results show that in the Chukchi Sea, Chl a was overestimated by the global algorithm (MODIS OC3Mv6) at concentrations lower than 0.9mgm−3 because of contamination by CDOM absorption, but underestimated at higher concentrations because of pigment packaging. Only within the in situ Chl a range of 0.6–2mgam−3 was the satellite retrieval error by the OC3Mv6 algorithm below the ocean color community goal of <35%. Using coincident in situ Chl a concentrations and optical data, a new linear empirical algorithm is developed (OC3L) that yields the lowest statistical error when estimating Chl a in the Chukchi Sea, compared to existing ocean color algorithms (OC3Mv6, OC4L, OC4P). When we estimated regional NPP using different Chl a satellite products as input, three distinct bio-optical provinces within the Arctic Ocean emerged. These provinces correspond to the inflow shelves, interior shelves, and outflow shelves+deep basin as defined by Carmack et al. (2006). Eleven sub-regions within the Arctic Ocean were grouped into each of these three provinces based on their mean value for R, the ratio of blue to green remote sensing reflectance (RRS). Our results suggest that three algorithms tuned to each of the three bio-optical provinces may be sufficient to capture the bio-optical heterogeneity within the Arctic Ocean. Currently, only within the inflow shelf province do we feel confident that Chl a and NPP can be accurately estimated by satellite using the OC3L algorithm. The interior and outflow shelf+basin provinces require development of ocean color algorithms specific to their respective bio-optical conditions.