Ocean Sci., 13, 925–945, 2017 https://doi.org/10.5194/os-13-925-2017 © Author(s) 2017. This work is distributed under the Creative Commons Attribution 3.0 License. Forecast skill score assessment of a relocatable ocean prediction system, using a simplified objective analysis method Reiner Onken Helmholtz-Zentrum Geesthacht (HZG), Max-Planck-Straße 1, 21502 Geesthacht, Germany Correspondence to: Reiner Onken ([email protected]) Received: 5 May 2017 – Discussion started: 23 May 2017 Revised: 26 September 2017 – Accepted: 9 October 2017 – Published: 20 November 2017 Abstract. A relocatable ocean prediction system (ROPS) 1 Introduction was employed to an observational data set which was col- lected in June 2014 in the waters to the west of Sardinia A relocatable ocean prediction system (ROPS) is presented (western Mediterranean) in the framework of the REP14- which enables rapid nowcasts and forecasts of ocean envi- MED experiment. The observational data, comprising more ronmental parameters in limited regions. In this study, ROPS than 6000 temperature and salinity profiles from a fleet of was implemented for the waters west of Sardinia (western underwater gliders and shipborne probes, were assimilated Mediterranean Sea) within the framework of the REP14- in the Regional Ocean Modeling System (ROMS), which is MED experiment (Onken et al., 2014, 2018). the heart of ROPS, and verified against independent observa- The major components of ocean operational systems are tions from ScanFish tows by means of the forecast skill score observations and ocean circulation models coupled with data as defined by Murphy(1993). A simplified objective analy- assimilation systems, to combine the observations with dy- sis (OA) method was utilised for assimilation, taking account namics and issue nowcasts and forecasts which are deliv- of only those profiles which were located within a predeter- ered to the customers. While systems on the global scale mined time window W. As a result of a sensitivity study, are utilised to provide estimates on large-scale circulation the highest skill score was obtained for a correlation length patterns and associated features, regional operational sys- scale C D 12:5 km, W D 24 h, and r D 1, where r is the ratio tems are focused more on societally relevant oceanographic between the error of the observations and the background er- information for, e.g., search and rescue operations, pollu- ror, both for temperature and salinity. Additional ROPS runs tant dispersal, fishery management (Edwards et al., 2015), showed that (i) the skill score of assimilation runs was mostly and military applications. Meanwhile, a number of real-time higher than the score of a control run without assimilation, (i) ocean operational systems are available, spanning the scales the skill score increased with increasing forecast range, and of ocean horizontal circulation patterns from global to coastal (iii) the skill score for temperature was higher than the score (Dombrowsky, 2011; Zhu, 2011). for salinity in the majority of cases. Further on, it is demon- ROPS was developed for military use in the context of strated that the vast number of observations can be managed rapid environmental assessment but it can easily be adapted by the applied OA method without data reduction, enabling to non-military demands. For maritime forces, there is a spe- timely operational forecasts even on a commercially avail- cial need for relocatable operational systems which can be able personal computer or a laptop. moved in any (potential) conflict area in the world ocean on short notice. As in the majority of cases conflict areas are re- gionally limited, the domains of the operational systems must be tailored to the corresponding regions which means that their horizontal extent should be of the order of 100 km. Con- sequently, the domains share always a wet (open) boundary with the open ocean. Relocatable operational systems based Published by Copernicus Publications on behalf of the European Geosciences Union. 926 R. Onken: ROMS OA on the Harvard Ocean Prediction System (HOPS; Robinson, model initialisation, boundary conditions, and a huge set of 1999) may be considered as the pioneering work in this sub- observational data for assimilation are available on the local ject. They became available since the late 1990s, and have computer system and a download from the internet is not re- been applied in various regional studies up to the present quired. The objectives are to demonstrate the following: (De Dominics, 2014). Another line of research based on the Naval Coastal Ocean Model (NCOM; Martin, 2000) can be – Good forecasts can be obtained from a prediction sys- traced back to the first decade of the present century (Row- tem using OA for assimilation (for the definition of ley and Mask, 2014), and recently, Trotta et al.(2016) pre- “goodness”, see Murphy, 1993). sented and compared the performance of a relocatable pre- – A vast number of observational data can be managed by diction system, using structured and unstructured grids. The OA without data reduction by averaging, sub-sampling, common properties and minimum requirements of any such or creating “super observations” (Lorenc, 1981; Moore system are as follows: et al., 2011b; Oke et al., 2015). – a tool for setup of the model domain, including the spec- – ification of the numerical grid, the bathymetry, and the ROPS is able to provide timely operational forecasts coastline; even on a commercially available personal computer or a laptop. – interfaces for definition of initial, lateral, and surface boundary conditions; The area of the ROPS model domain (Fig.2) is charac- terised by a 20–50 km wide continental shelf. The shelf ends – a numerical model; at water depths between 150 and 200 m, followed by the con- tinental slope which features several canyons. The deep-sea – an interface for the provision of observational data; area belongs to the northern Algerian Basin (also referred to as the Sardo-Balearic Basin) and exhibits water depths of up – a data assimilation module; to 2800 m. According to Millot(1999), the mean surface cir- – software for post-processing of the model output. culation is mainly related to the inflow of “new” Modified Atlantic Water (MAW) from the Strait of Gibraltar by means An additional demand for relocatable operational systems is of anticyclonic eddies originating from the Algerian Current. to provide accurate nowcasts and forecasts of the ocean en- Another branch of “old” MAW, which mixed with the un- vironment in a timely manner, i.e. in near-real time. How- derlying water masses on its large-scale cyclonic circulation ever, the requirements of accuracy and timeliness are incon- through the Tyrrhenian, Ligurian, and Balearic seas, comes sistent with one another: accuracy requires the application of probably from the west via the Balearic Current (García et up-to-date assimilation schemes which presently are ensem- al., 1996). Just below the MAW, Winter Intermediate Water ble or variational methods. As the implementation of these (WIW) follows the path of the MAW along its whole cy- schemes is rather complex and they are computationally ex- clonic path. Levantine Intermediate Water (LIW) originates pensive (Zaron, 2011), timely delivery can only be realised from the eastern Mediterranean and the direct path to the on powerful computers which are frequently not available. ROPS domain is via the Sardinia Channel and then north- As a compromise, sequential data assimilation based on ob- ward around the southern tip of Sardinia. Below the LIW, jective analysis (OA; Bretherton et al., 1976; Thomson and Western Mediterranean Deep Water (WMDW) and Bottom Emery, 2014) is used in ROPS. OA is not as accurate as the Water (BW) are found. ensemble Kalman filter (Evensen, 2006) or 4D-Var (Moore From analyses of the REP14-MED observational data set, et al., 2011a,b), but the computational costs are much less it turned out that the distribution of the water masses and the and the implementation is straightforward. circulation patterns resembled the classical picture described Meanwhile, ROPS has been implemented for various re- above, but there were also significant differences. Accord- gions in the world ocean, and has been running automatically ing to Knoll et al.(2017), the temperature and salinity of without any major interruptions since early 2015. The con- MAW, LIW, and BW had increased compared to the obser- cept (Fig.1) for all realisations is identical: every day, ROPS vations during the last decade. In addition, an anticyclonic is initialised from a restart file of the previous day’s run, and WIW eddy with unusual low temperatures and salinities was it provides a 3-day forecast relative to the present. For each identified which may confirm the existence of a direct route run, data sets for the definition of initial and boundary condi- of WIW from its formation region to the observational site. tions plus observational data for assimilation are downloaded In contrast to previous observations, LIW occupied the whole from the internet, in which the initial conditions are only re- trial area and the predominant direction of the geostrophic quired for re-initialisation of ROPS in the case that it died the flow was to the north with the largest transports in the deep day before. water off the 1000 m depth contour; no LIW vein tied closely For this article, ROPS is slightly modified: it is run in hind- to the Sardinian coast was found south of 40◦ N. The MAW cast mode for a period of time in June 2014. All data for pattern was different: namely, the major northward transport Ocean Sci., 13, 925–945, 2017 www.ocean-sci.net/13/925/2017/ R. Onken: ROMS OA 927 Figure 1. The ROPS concept: web resources are depicted by clouds, blue parallelograms represent data sets on the local host, and processes are indicated by green rectangles.