Transition from Research to Operations: ARKTOS - A Knowledge-Based Sea Ice Classification System

ARKTOS is a fully automated intelligent system that classifies sea ice and that is now being used by the U.S. National Ice Center (NIC) for daily operations related to the NIC’s task of mapping the ice covered oceans. In this paper we describe the process of taking a research project and transitioning it to an operational environment. We discuss the theoretical methodologies implemented in ARKTOS, and how ARKTOS was developed, tested, and finally moved to operations. HISTORY The U.S. National Ice Center (NIC), under sponsorship of the U.S. Navy, U.S. Coast Guard, and National Oceanic and Atmospheric Administration (NOAA), is tasked with mapping the ice covered oceans of the world using both remotely sensed and in situ observations. Synthetic Aperture Radar (SAR) data became a major input to the program of the NIC soon after the November, 1995 launch of the Canadian RADARSAT satellite [1]. The Alaska SAR Facility (ASF) at Fairbanks provides RADARSAT data to the NIC. The NIC also acquires limited amounts of imagery under contract from Tromso, Norway and West Freugh, Scotland, and from the Gatineau station mask through a bilateral data exchange agreement with the Canadian Ice Service (CIS). In order to make efficient use of the 0.8 GB per day of data currently received, the NIC has been actively involved in the development of systems that can assist in analysis of ice conditions using SAR. The University of Kansas (KU) began to study the use of expert systems in sea ice classification from SAR under a NASA graduate student fellowship in 1990. The work was extended in 1992 and 1993 under a NASA Mission to Planet Earth grant and produced various techniques to measure and identify sea ice features in SAR imagery [2,3]. Of particular interest to the NIC was a system to classify winter sea ice into three major classes using dynamic thresholding and expert, rule-based systems. A research to operations development cycle was identified, and appropriate contributions from both the research and operational community were identified.