Journal of Oceanography, Vol. 65 (No. 1), pp. 81-89, 2009Short Contribution
Taiyo Kobayashi1*, Brian A. King2 and Nobuyuki Shikama1
1Institute of Observational Research for Global Change, Japan Agency for Marine-Earth Science and Technology, Natsushima-cho, Yokosuka, Kanagawa 237-0061, Japan
2National Oceanography Centre, Southampton, Empress Dock, Southampton, SO14 3ZH, United Kingdom
(Received 25 December 2007; in revised form 30 August 2008; accepted 2 September 2008)
Abstract: Estimating the average lifetime of floats is very important for Argo, because the total cost of maintaining the monitoring network largely depends on float lifetime. However, the actual lifetime of floats used in Argo is currently unknown. An estimate can be made by examining past float survival, but this is complicated by floats still operating at sea and continuous improvements in float hardware. Because APEX (Autonomous Profiling Explorer) floats are the most widely deployed type of float in the world oceans, in this study we estimate the lifetime of the latest model of APEX powered by alkaline batteries. The expected lifetime is estimated with a statistical method that allows for floats that are still active and that failed because of a known and now fixed hardware fault that should not cause failure in the latest model of floats. As an example, we analyzed the APEX fleets managed by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC), because we have access to a JAMSTEC database in which the causes of float failure have been carefully correlated to known hardware problems. Analysis of the JAMSTEC fleet (n = 571, as of 7 May 2008) indicated that the expected lifetime of the latest model of APEX is 134.6 (127.6-141.5, considering standard errors) cycles, equivalent to 3.7 years of 10-day cycles. We conclude that the annual deployment of 813 (773-859) APEX floats is needed to maintain the Argo observational network of 3000 floats. Floats with different hardware configurations (e.g., lithium batteries) or different mission programs (e.g., shallower profiling, deeper profiling every several cycles) may be expected to have an even longer lifetime.