Design of buoy observation network over the Arctic Ocean

In the Arctic, near-surface temperature affects the surface energy balance and the evolution of planetary boundary layer. Particularly in summer, due to sea-ice extent variations, the uncertainties of simulated near -surface temperature in the Arctic Ocean increase. To reduce these uncertainties, high-quality near-surface temperature observations are necessary; however, in situ observations are difficult to operate in the Arctic. Although buoys can monitor various atmospheric variables in the Arctic over longer periods compared to those of other types of in situ monitoring, the number of buoy near-surface temperature observations in the Arctic has decreased since 2018. Near-surface temperature observations by buoys need to be added and the optimal buoy observation network need to be designed to reduce the uncertainties associated with estimating near-surface temperature over the Arctic Ocean in summer. In this study, buoy observation networks to monitor near -surface temperature were designed to minimize the 24-h forecast error variances of 2-m temperature in the Arctic. To design the optimal buoy observation network, an objective algorithm for the observation network design based on an ensemble sensitivity was used. The reductions of the forecast error variances of 2-m tem-perature were estimated quantitatively using the ensemble sensitivity. To calculate the ensemble sensitivity, ensemble members of the initial condition and response function (i.e. forecast aspect of interest) were composed of deterministic analyses and 24-h forecast errors of 2-m temperature, respectively, which were simulated by the Global Forecast System (GFS) during July, August, September, and October from 2016 to 2019. By using long-term GFS simulations, the buoy observation network was designed in a climatological perspective. The determined optimal buoy locations were consistent with regions where both the analysis variances and forecast error variances of 2-m temperature were large. Assuming that 2-m temperature buoy observations were assimilated to ensemble members from GFS simulations at the 10 optimal buoy locations, the analysis variances of 2-m temperature were reduced by 60.48% and the 24-h forecast error variances of the temperature were reduced by 2.83% compared to the respective variances before assimilating buoy observations. Thus, after efficiently using the limited buoy resources (i.e., 10 buoys) to design the observation network, the uncertainties of the 2-m temperature simulated by the numerical weather prediction (NWP) model were greatly reduced in the Arctic. The optimal buoy observation network can contribute to reducing uncertainties associated with the surface energy balance, planetary boundary layer, and sea-ice properties simulated by NWP models.