Lack of precision in measurements and accounting of numbers of salmon, average weight and size distribution, is a costly game of guesswork for the industry. A recent impact study, The economical consequences of insufficient biomass control in sea-cage aquaculture, gives an estimate of the total potential losses due to insufficient control on a 900 000 ton yearly production and a typical ±5% error on biomass. The numbers then add up to about NOK 1 billion each year only in suboptimal sales and suboptimal feeding (Aarhus, in prep.). There is no apparent evidence that the worldwide losses should be of less proportions. Biomass estimation using the currently available methods functions reasonably well in smaller cages, but as the cages have grown bigger, the precision of today’s solutions has become less accurate. Marine Harvest Norway presents errors of 5-10 % to be normal (Ragnar Nordtvedt pers. com.).

State-of-the-art biomass estimation is done by a combination of experience-based knowledge and measurements using different instruments and/or manual weight sampling. The technology basis applied has not seen major leaps in development during the last 5-10 years (Sunde et al. 2003). An estimate of the biomass is dependent on knowledge of the number of fish in the cage and their average weight. The initial count is done by pipe-integrated counters at the wellboats as the smolts are transferred to the cages. Numerous incidents of counting errors have been reported and some farmers opt to rather trust the counts from vaccination prior to sea transfer than the well-boat counts (Aarhus, in prep.). During production the number of fish is adjusted according to removed dead fish as well as coarse estimates of other unobserved losses through an evaluation of feed consumption and growth tables supplied by the feed producing companies. If the fish are sorted by well-boat, they are usually simultaneously counted. A fish count obtained this way can be used to reset the initial number for the production planning. It seems like inaccurate counts can be linked to situations where fish are pumped too fast, known as a counter overload. Significant improvements in counting accuracy have been obtained in research projects at ICT by using video cameras and advanced image analysis (Thielemann & Clausen 2000).

Average weights are obtained with different methods, ranging from manual weighing to almost fully automated systems. In vitro manual weighing of individual fish is rarely done in commercial salmon farms these days, and the same is true for “Archimedes weighing” (counting fish and measuring their total volume in a special tank). For both methods, the fish sample is caught by casting a seine in the cage and landing samples by hauling a closed-bottom net through the crowded fish, from bottom to surface. There are mainly two types of automated, in situ biomass measurement technologies available: Various frame systems measure fish as they pass through, or stereo-camera systems where a set of pictures of fish is collected and subsequently analysed for fish weight. Both systems are based on optical principles, and they indirectly yield average weight and weight distribution based on a sample of the population. The technologies are based on ambient light sources and cameras or optical scanners, which are combined to obtain the three-dimensional contour of fish within a small measurement volume or across a plane (Beddow et al. 1996, Waagsbø & Lillerud 2003, Kyrkjebø 2007). Further developments of such instruments are ongoing in the project “Non-disruptive weight estimation” funded through CREATE, Centre for Research Based Innovation in Aquaculture Technology (Clausen et al. 2007).

Within CREATE the project “Bio-Statistical Analysis” aims to further develop empirical models for predicting fish weight and weight distribution based on historical data logged throughout the growth history. Such methods are not very accurate, but can be a valuable addition to direct measurements from instruments. Generally, cage biomass estimation bears close resemblance to a well-known class of engineering problems, where the aim is to determine the true state of a system given a set of observations at the system outputs (measurements) and information about the system’s expected behaviour (model) which are more or less contaminated with noise and uncertainty. This is the basis for development of model-based estimation techniques which have found widespread use in engineering disciplines, such as aerospace, robotics, process control and numerous others (Minkler & Minkler 1993). Model-based estimation has been suggested as a promising concept within the fisheries and aquaculture industry (Balchen 2000), and its feasibility has been demonstrated for marine larvicultures (Alver et al. 2005).

Attempts have been made at measuring biomass acoustically using echo sounders, similar to fish stock assessment at sea (Knudsen et al. 2002, Zhao & Ona 2003, Burczynski et al. 2004, Knudsen 2004, Conti et al. 2006). The results, however, have generally been discouraging due to artefacts of the high biomass density in fish farming cages. Furthermore, such methods generally target the overall biomass rather than size distribution. Echo sounders have shown some potential for monitoring the spatial distribution of fish and could possibly be a useful tool ensuring representative sampling of a fish population (Knudsen & Gjelland 2004). Fish tagging and recognition by RFID or acoustics are other possible supporting technologies, in particular for research and development (Holm et al. 2007). Lower-frequency acoustics, exploiting swim bladder resonance for inferring fish size has been tried without achieving sufficient accuracy in biomass measurement (Nero et al. 2004, Godø et al. 2009). Electrical conductivity has been used for measuring total biomass of fish during pumping primarily within fisheries and so far with less accuracy than what is desirable for the purpose of custody transfer of farmed fish (Lied et al. 2000).

No single instrument has so far been able to give a precise estimate of number of fish and weight distribution in a cage; and thus a combination of different technological solutions is  desirable to achieve more reliable and predictable results for biomass estimation. Continuity of data is important for efficient production management and failure detection. The EXACTUS ambition is to gather information from different sensors, new and existing, based on a well thought-through sampling regime regarding the entire production regime and feed this into an intelligent model system. Through this approach the consortium expects that EXACTUS, during its three years, will provide some necessary tools for a possible quantum leap in biomass control.