We present results of the optimization of near-real time on-board sampling strategy in the Iceland-Faroes oceanic frontal area, based on the outputs of a mesoscale 3D operational data assimilation forecasting experiment. By minimizing a root mean square error cost function, we show that in this example an along-front sampling strategy, i.e. with transects parallel to the front, produces smaller errors in temperature, salinity, nitrate, phytoplankton, and zooplankton fields, as a result of a combination of the direction of the sampling of the front and errors associated with the asynopticy of observations (Doppler effect). This is contrary to the classic across-front sampling strategies that are used in most field experiments reported in the literature, i.e. where transects are perpendicular to the front. A control model shows that at these spatio-temporal scales, the along front sampling strategy is optimal when the frontal instability has sufficiently developed.