Assessing the impact of subsurface temperature observations from fishing vessels on temperature and heat content estimates in shelf seas: a New Zealand case study using Observing System Simulation Experiments

We know that extremes in ocean temperature often extend below the surface, and when these extremes occur in shelf seas they can significantly impact ecosystems and fisheries. However, a key knowledge gap exists around the accuracy of model estimates of the ocean’s subsurface structure, particularly in continental shelf regions with complex circulation dynamics. It is well known that subsurface observations are crucial for the correct representation of the ocean’s subsurface structure in reanalyses and forecasts. While Argo floats sample the deep waters, subsurface observations of shelf seas are typically very sparse in time and space. A recent initiative to instrument fishing vessels and their equipment with temperature sensors has resulted in a step-change in the availability of in situ data in New Zealand’s shelf seas. In this study we use Observing System Simulation Experiments to quantify the impact of the recently implemented novel observing platform on the representation of temperature and ocean heat content around New Zealand. Using a Regional Ocean Modelling System configuration of the region with 4-Dimensional Variational Data Assimilation, we perform a series of data assimilating experiments to demonstrate the influence of subsurface temperature observations at two different densities and of different data assimilation configurations. The experiment period covers the 3 months during the onset of the 2017-2018 Tasman Sea Marine Heatwave. We show that assimilation of subsurface temperature observations in concert with surface observations results in improvements of 44% and 38% for bottom temperature and heat content in shelf regions (water depths< 400m), compared to improvements of 20% and 28% for surface-only observations. The improvement in ocean heat content estimates is sensitive to the choices of prior observation and background error covariances, highlighting the importance of the careful development of the assimilation system to optimize the way in which the observations inform the numerical model estimates.