ANALYZING QUALITY OF MONITORING DATA ON DISSOLVED OXYGEN AND SEA TEMPERATURE FROM MULTIPLE NORDIC OPEN SEA CAGE FARMERS

 Authors:   ,

Introduction

 The  interest to introduce digital technologies, such as AI and digital twins, into aquaculture practices is growing, for the purpose of automating and streamlining operations, including stabilizing dissolved oxygen (DO) levels and  optimizing feeding [1,2] .  These technologies rely on quality data to provide value.  Ambient sea temperature (T) and DO are central to animal welfare and growth [3,4]  and farmers collect vast amount of data on these parameters. However, due to the high variability of these water quality parameters within and between sites, standardizing methods to collect data is challenging. While reporting on the parameters are regulated, a lack of trust in the quality of the data prevents it from being used operationally. This phenomenon occurs in several industries [5]. For open sea cages, we have found no existing framework on how to collect and verify that the data collected in operations  are of sufficient quality. In this study, we have gained access to production data from 18 open sea cage farmers to investigate the quality of data and potential operational values.

Methods and materials

 In total, the 18 companies have 105 sites listed with 236 sensors. 137 sensors are active and 99 are inactive , and the active and inactive periods range from days to multiple years .  The sites are spread across a large geographical area along the Norwegian coast and the Faroe Islands.  The data is  analyzed  in an exploratory manner,  using systematic data quality frameworks developed for other industries [5] ,  here applied to aquaculture practices.

Results and Discussion

In our preliminary results, w e find a range of quality issues in the data from the majority of the sensors .  Physical granularity is the first main issue we identify, as 85% of the sites have four or less sensors across the space of the  entire site. DO , especially, can vary greatly across short spaces [6] , requiring a high level of granularity. The second pervasive issue we identify is accuracy. The data is noisy, and  often only one sensor is placed in a cage, making validation of abnormal variations difficult, as it could be due to local variations that, albeit unusual, can occur naturally . Two examples of this are visualized by graphs 1 and 2 in Figure 1, with the first example showing unusually large variations in T and the second showing unusually similar variations in DO and T across all depths of the cage, which is often an indication of sensor issues . In this case, however, variations could be real and caused by a homogenous environment. Other sensor measurement errors are easier to detect, with examples  in graphs 3 and 4 in Figure 1 . Graph 3 in this example shows too stable DO values over time, within the optimal range. While we can visually observe an issue in this case , detecting it in real-time can be more challenging, as the values are not identical over time and within the positive range. Sensor 4 shows clear sensor errors in both T and DO. Abnormal values that are not caused by sensor errors can be due to production routines when the sensors are e.g., moved, such as net cleaning, sensor cleaning, disease treatment or counting. For digitalization using this data to provide value to the organization, more information is required . These early results indicate that current methods to monitor  the sea cage environment is insufficient and provide low operational value. Methodical improvements for systematic monitoring are necessary before implementing digital technologies to extract value from this type of data .

Refer ences

[1]  C. Wang, Z. Li, T. Wang, X. Xu, X. Zhang, and D. Li, “Intelligent fish farm—the future of aquaculture,” Aquac Int , vol. 29, no. 6, pp. 2681–2711, 2021, doi: 10.1007/s10499-021-00773-8.

[2] M. Føre et al., “Precision fish farming: A new framework to improve production in aquaculture,” Biosystems Engineering, vol. 173, pp. 176–193, Sep. 2018, doi: 10.1016/j.biosystemseng.2017.10.014.

 [3] M. Remen, F. Oppedal, A. K. Imsland, R. E. Olsen, and T. Torgersen, “Hypoxia tolerance thresholds for post-smolt Atlantic salmon: Dependency of temperature and hypoxia acclimation,” Aquaculture, vol. 416–417, pp. 41–47, Dec. 2013, doi: 10.1016/j.aquaculture.2013.08.024.

 [4] M. Remen, M. Sievers, T. Torgersen, and F. Oppedal, “The oxygen threshold for maximal feed intake of Atlantic salmon post-smolts is highly temperature-dependent,” Aquaculture, vol. 464, pp. 582–592, Nov. 2016, doi: 10.1016/j.aquaculture.2016.07.037.

 [5]  J. Morewood . “Building energy performance monitoring through the lens of data quality: a review ” Energy & Buildings , vol. 279 , pp. 582–592, Jan. 2023, d oi: 10.1016/j.enbuild.2022.112701.

 [6] D. Solstorm et al., “Dissolved oxygen variability in a commercial sea-cage exposes farmed Atlantic salmon to growth limiting conditions,” Aquaculture, vol. 486, pp. 122–129, Feb. 2018, doi: 10.1016/j.aquaculture.2017.12.008.