74—JOURNAL OF FOOD SCIENCE—Volume 61, No. 1, 1996 Atlantic Salmon Average Fat Content Estimated by Near-Infrared Transmittance Spectroscopy JENS PETTER WOLD, TONE JAKOBSEN, and LEIF KRANE Fig. 1—Drawing of an Atlantic salmon. Grey pattern indicates lo- cation and size of cut used for both chemical and spectral meas- urements. ABSTRACT Near-infrared transmittance spectroscopy was used to determine the av- erage fat content in farmed Atlantic salmon fillets with skin and scales. The fat content was 5.7–17.6% and weight range 1.0–5.4 kg. A partial least square regression including 43 salmon resulted in a multivariate prediction correlation of 0.97 and a root mean square error of cross validation of 0.75%. Regression models using 6 to 9 wavelengths in the 850–1048 nm range gave somewhat lower prediction errors than a model using 100 wavelengths. Results showed that NIR transmittance was suited to determine fat content nondestructively in whole salmon fillets with skin and scales. Key Words: salmon, fat, NIR transmittance, partial least square, non- destructive spectroscopy INTRODUCTION FAT CONTENT is one of the most important quality criteria of Atlantic salmon (Salmo salar), not only from a nutritional point of view, but also due to its sensory and functional properties (Haard, 1992; Skonberg et al., 1993). The European markets are becoming increasingly concerned with quality and demand mainly salmon with low fat content (below about 12%) (Midt- vedt, 1995). Chemical methods for fat determination are highly destructive and time-consuming, and require use of hazardous chemicals which may be harmful to analysts and the environ- ment. A rapid, low-cost and nondestructive method of estimat- ing fat content would provide salmon farmers with the oppor- tunity to sort fish with different fat content according to market requirements and product specifications. It may also provide a standard method to solve ‘‘disagreements’’ between buyers and sellers. For most applications an estimation accuracy of 1% is satisfactory. Near-infrared (NIR) spectroscopy is well known for rapid and nondestructive compositional analyses of foods. NIR instru- ments for analyzing grains, flour and oilseeds are common (Os- borne and Fearn, 1988). However, development of NIR methods for fish and fish products has been limited. NIR reflectance de- termination of fat content in freeze-dried (Mathias et al., 1987; Gjerde and Martens, 1987) and intact tissue (Rasco et al., 1991) of rainbow trout (Oncorhyncus mykiss) has been reported. Lee et al. (1992) tried the same method on whole rainbow trout. Reflection measurements were performed through the skin of 52 frozen and thawed fish in the weight range 66.5–883g, the fat range was about 2.0–13%. Prediction error results for fat were in the range 0.7–2.2%, correlation coefficients ranged from 0.73 to 0.90, varying with the measurement locations on the fish body. Recently nondestructive NIR analysis of Atlantic salmon fil- lets was reported by Isaksson et al. (1995). Reflectance meas- urements with a remote fiber optic probe were performed on the skinside of 50 whole skinned salmon fillets from fish weighing between 4 and 5 kg. The fat range was 9.1–20.5%. A prediction error of 1.08% corresponding with a correlation coefficient of 0.96 was achieved averaging spectral readings from nine differ- The authors are affiliated with the Norwegian Inst. of Fisheries & Aquaculture Ltd. P.O. Box 2511, 9002-Tromsø, Norway. ent locations on the fillets. The lowest prediction error for single point readings was 1.39%. Average fat content has been estimated with a prediction error of 0.7% from samples of minced salmon applying NIR trans- mittance (Sollid and Solberg, 1992). Good results have also been reported using NIR transmittance to determine pH and wa- ter and protein content in minced cod (Lauritzen, 1994). Our main objective was to investigate development of a rapid, low-cost and nondestructive method for fat content in whole and live salmon. However, as a first step we focused on nondestruc- tive determination of fat in fresh fillets with skin and scales by NIR transmittance spectroscopy. For products such as smoked and brine-cured salmon it is desirable to keep the skin on the fillets. The experiments were divided into two phases. The first phase was to locate sites of the salmon where spectral measure- ments correlated well with average fat content. We also inves- tigated how skin and scales influenced the spectral readings and transmittance differences between light and dark muscle of salmon. In the second phase, NIR measurements were collected from additional salmon to obtain a more extensive prediction model consisting of 43 species. An additional objective was to compare prediction results by using different numbers of wave- lengths in the model. MATERIALS & METHODS Determination of suitable sites for measurement Fat and water together make up 80% of salmon muscle weight. The fat content, which negatively correlates with water content, varies from 5 to 25%. Protein holds a relatively constant level at 18–20%. The fat is unevenly distributed in the fish and distribution varies between indi- vidual fishes (Hardy and King, 1989). However, in an area behind the dorsal fin (Fig. 1), the fat content approximately equals average fat con- tent of the whole salmon (Norwegian General Standardizing Body, 1994). To reduce work and raw material, only that part of the salmon is (in Norway) normally used for determination of approximate average fat content by chemical methods. To find the site of measurement which gave the best prediction results of average fat content, NIR transmittance spectra of intact tissue from different locations of the shaded area (Fig. 1) were recorded. To study