THE DEPENDENCE OF SHORT-RANGE OCEAN FORECASTS ON SATELLITE ALTIMETRY Peter R. Oke and Madeleine Cahill CSIRO Marine and Atmospheric Research, GPO Box 1538, Hobart TAS 7001, Australia, Email: peter.oke@csiro.au ABSTRACT Short-range ocean forecast and reanalysis systems rou- tinely combine observations from satellite altimetry, satellite sea surface temperature (SST), and in situ tem- perature and salinity, to initialise global and regional ocean models. The most critical observation type for eddy-resolving applications is arguably satellite altime- try. To quantify the impact of satellite altimetry obser- vations on a data assimilating ocean general circulation model we perform a series of Observing System Experi- ments (OSEs). We perform four OSEs that all assimilate in situ temperature and salinity observations and satellite SST observations. Different OSEs assimilate data from a different number of altimeters, including an OSE using no altimetry data, and data from one, two, and three al- timeters. We show that the neglect of altimetry from a data assimilating model increases the model-observation mis-fits in high-variability regions by almost 40%. 1. INTRODUCTION GODAE OceanView is the successor to GODAE - the Global Ocean Data Assimilation Experiment. One of the goals of GODAE was to demonstrate the feasibility of op- erational ocean forecasting. This goal has been achieved, with many operational centres now producing daily or weekly short-range ocean forecasts. All of the forecast systems assimilate satellite altimetry observations. The operational ocean forecasting community regards altime- ter observations as the most critical observation type for mesoscale ocean initialisation. The purpose of this study is to quantify the importance of satellite altimetry for forecast and reanalysis systems like those developed un- der GODAE and GODAE OceanView. To quantify the impact of satellite altimetry observations on a data-assimilation eddy-resolving ocean general cir- culation model, we perform a series of Observing Sys- tem Experiments (OSEs). In the OSEs reported here, we systematically with-hold data from one, two, and three altimeters. The ocean model and data assimilation system we use is the latest version of the Bluelink system, the pre- decessor of which was described previously [8, 10]. Bluelink represents Australia’s contribution to GODAE and GODAE OceanView; and is a partnership be- tween CSIRO, the Bureau of Meteorology, and the Royal Australian Navy. The Bluelink system has been previously used to perform operational ocean fore- casts (www.bom.gov.au/oceanography/forecasts/; [1]), and multi-year ocean reanalyses [10, 15]. Output from Bluelink applications has been used to explore ocean dy- namics [16, 17, 11], and for other observing system de- sign and assessment [13, 12]. 2. MODEL AND DATA ASSIMILATION 2.1. Model The ocean model used here is a configuration of the GFDL Modular Ocean Model [6] and is called the Ocean Forecasting Australia Model (OFAM). To date, the de- velopments of OFAM, under Bluelink, have focussed on modelling the circulation of the upper ocean in the Aus- tralian region. This is reflected in the OFAM grid, with 5 m grid spacings at the ocean surface and 10 m verti- cal grid spacings over the top 200 m. The horizontal grid spacings are 1/10 ◦ in the 90 ◦ -sector centred on Australia and south of 16 ◦ N, and coarser outside of this region. To accommodate the inhomogeneous resolution, the hor- izontal viscosity is resolution and state-dependent, based on the Smagorinsky-scheme [5]. The bottom topogra- phy for the configuration of OFAM that is used here was constructed from a range of different sources, as docu- mented elsewhere [15]. The turbulence closure model used by OFAM is a version of the hybrid mixed-layer scheme [2], plus implicit tidal mixing [7]. To date, model runs performed using OFAM were forced by six-hourly atmospheric fluxes from ERA-Interim [3]. 2.2. Data Assimilation The data assimilation system used in this study is called the Bluelink Ocean Data Assimilation System (BODAS) [10], and is based on Ensemble Optimal Interpolation (EnOI; [9]). For this study, BODAS is implemented