Team #06 // Njord Smart Fishing
Our developed concept is Njord Smart Fishing that delivers a platform to complement fishermen in their decision making, which helps to maximize their yield while minimizing the time wasted in searching for a good fishing spot. The platform is based on the processing of satellite data, fish migration simulation models and machine learning.
Through experience fishermen build intuition and knowledge of where to go when planning their fishing trip. There is often no way for the fishermen to know for sure if their planned locations will have enough fish for them to catch and hereby cover the expenses of the trip. This is a major issue, as it consequently means that the fishermen sometimes would have been better off not leaving the harbour at all. This problem is increasing due to the decreasing populations of fish as a consequence of overfishing.
Target user / Customer
Our target user is the Industrial fishermen. The fishing industry is struggling to catch enough fish to cover their expenses. This should make fishermen an easily obtainable customer if we can convince them that our platform can increase their revenue. This is on the other hand also a conservative industry. It is therefore essential that our solution is easy to understand but also the platform should be easy and intuitive to use in order for the fishermen to care. We will start by targeting Danish fishermen. After success in Denmark, we will expand to the rest of the nordic countries, before going to Europe. In Denmark, there are approximately 2400 fishing vessels and a total of 11000 in the Nordic countries.
Our solution and how the concept is feasible
Use of satellite data in the fishing industry is an established practice. A few companies offer the fishing industry satellite data. Nevertheless the data are largely interpreted analytically by the fishers themselves. This is a time consuming and challenging process, and the results might not be as good as wished. We are therefore building a platform that provide trawler crews with user friendly data interpretation. The platform will be built and based on three main technologies;
1) Processing of satellite data
Schools of fish have been tracked since the ’90s. Sea surface temperature, sea currents, and chlorophyll measurements can be used in predicting the location and concentration of fish, by for example the prediction of algal blooms.
Fish migration has been studied for years. Our team has the expertise to develop simulation models that can predict the short term future migration of fish.
3) Machine learning
Using the abovementioned satellite data, we can train a neural network to predict locations of high concentrations of specific fish types. A major issue in the fishing industry is overfishing. If a population of fish gets under a critical level it will not be able to recover fast enough, as the breeding rate is proportional to the population size. If most of the trawlers are using our platform, we can distribute the fisheries such that overfishing is prevented. This means that the fishermen have a monetary incentive to use our platform but also to make an investment in the future of their occupation. The way that Njord Smart Fishing differentiates is by utilizing a wide array of satellite data to create reliable prediction models along with a user-friendly interface.
The amount of emissions are directly proportional to key design parameters for trip-planning in the fishing industry. With the Njord SF platform it is intended, along with the above mentioned technology, to as well incorporate other relatable key parameters, such as total emissions, into a user-friendly interface. These parameters are not crucial for machine learning to function, nevertheless they are rather included in the platform, such that the consumer in any case will experience familiarity with the product.
Building this tool is a very challenging task. Even though our team has expertise in necessary technical disciplines, the team does not have a lot of knowledge about fish behavior, nevertheless, a lot of research has previously been done. Therefore to accurately develop models for fish migration, it also requires allocation of resources for research in the area. We estimate a development phase of 2 to 3 years before having built a functional platform that can be released to the market.
Technical development areas:
Process appropriate satellite data
- Use freely available satellite data in training for the neural network. Further inclusion of free or purchasable external historical training data.
- Categorize optimal formats of processed data for best performance of our neural network
- Research fish behavior and migrations. Take subset in well-defined models for the determination of variables needed to be taken into account in building a reliable prediction model.
- Research/investigate which sort of data makes for the best performance of the neural network. Build a training set that can be used to train a highly accurate network. Building good neural networks requires a lot of fine-tuning. How to do this fine-tuning is a sort of pseudoscience which requires a lot of trial and error. It can require a substantial amount of time to train these networks if we do not have access to high-performance computers.
Known challenges: In order to develop highly accurate neural networks, very large training sets are needed. An application of machine learning for the prediction of fish migration from space seems not to be readily available. This means that there is no ready-to-use training set for this application. Consequently, training sets consisting of a substantial amount of labeled inputs must be generated. As previously mentioned the information required is well established in terms of prior research. This leads to a fundamentally feasible path in determining proper training data, which can be labeled in an automated fashion.
Four unique value propositions
What makes our solution/concept unique?
- Improved pre-trip planning for the fishing industry
- Saving fishermen time and fuel as the suggested destinations have larger yield probability.
- Increased yield for fishermen
- Sustaining a healthy fish population, securing the future of the fishing industry
The largest trawlers cost around 300M DKK to purchase. When such a trawler leaves the harbour it has a daily average operational cost of 250k DKK. This covers crew wages, visas/travel permits, maintenance and repair, registrations etc.
If our platform can make for marginal improvement of 2-4%. The trawler can stay in the harbour for 7-14 days more in a year, which translates to a saving of 1.8 – 3.6 mill DKK. If our platform can suggest a trawler crew to wait 2 days for sailing out, we would have saved them 500k DKK. These numbers are gathered from a meeting with Flemming Thorbjørn Hansen, Section for Oceans and arctic, special consultant (see annex).
As our initial target markets are the nordic countries, it is of relevance to present the value of the fishing industry herein.
Next we turn the focus to the Business Model Canvas. Starting with the central part of the BMC, reveals the most prominent value propositions of Njord Smart fishing. A key aspect of the business model is the ability to differentiate from what is already on the market. Furthermore, a key driving force for the company is the strong connection between a marginal increase in yield and resulting savings in operational costs. Finally with regards to sustainability, as more companies become a customer of the Njord SF platform, the vessels can in larger-scale be distributed such that overfishing is prevented whilst still achieving a marginal increase in yield, thus also less fuel consumption.
Moreover, the company shall be driven with a variable cost-driven structure, due to a tailored buyer-supplier relationship, i.e. a specific sales are always done as a tailored manifestation of the Njord SF platform, namely to meet specific trawler demands. This model also supports the revenue stream being subscription-based.
Key activities within the company include platform development (production), software testing (platform/network) and neural network training (ongoing development). Through these key activities, Njord SF seeks to deliver the aforementioned value propositions. The customer segment is B2B, as fishing vessels most often are a part of a company with continuous revenue generation. To reach these businesses, several channels shall be established. First of all the product can be sold directly through our website (link below for more information). For product awareness, we furthermore seek to have affiliate programs and business representatives to achieve sales.
The main goal of Njord SF is to create a platform that directly saves labor and cuts fuel costs for fishing companies. This is done by providing optimal fishing routes and hereby save the environment for large amounts of inefficient fuel emissions. For the fishers, this means improved turnover and better profits. As the business grows it will be easier to control and distribute vessels around the ocean. In this way, overfishing can be prevented, whilst still providing fishers with a marginal increase in yield. At this stage (long term) it can be beneficial to partner up with regulation makers (DTU-AQUA) to accommodate sustainability goals. With Njord SF platform we envision the following 2015 global goals to be addressed
We are a team with diverse competences, ranging from business experience to highly technical know-how, equipped to lift the task of making this business a reality. We care deeply about the fishing industry as it has been core to the Nordic industry for centuries. We are motivated in helping this industry in a sustainable manner, namely one that preserves the aquatic ecosystem and ensures a rich fishing foundation for years to come.
Link to our Njord SF website
Meeting summary 30/10-2019
With Jens Olaf Pepke Pedersen, senior researcher, Innovation and research based counseling, at DTU – Space, Denmark’s national space institute
Goal of meeting: Confirmation of idea, and type of training data to include
After explanation of the idea, Jens Olaf stated: “…Fundamentally the idea will work”
We were further advised to make use of
- Sea surface temperatures (SST)
- Ocean current measurements ~50 km resolution
- Ocean color measurements
- External data to the extent available: GPS, catch volume
Meeting summary 01/11-2019
With Flemming Thorbjørn Hansen, Section for Oceans and arctic, special consultant. Asbjørn Christensen, Section for Marine Living Resources, senior researcher.
Goal of meeting: View the idea from a biological (marine) stand-point
The idea is interesting and surely feasible. Research is already being done, and at DTU Aqua they are currently building machine learning algorithms to predict fish migration, this however on a research basis and not commercial. After a technical discussion on the algorithms themselves, we were given:
- Key numbers for large trawler leading to yearly savings of 3,6 mill dkk.
- Companies that already sell satellite data to the fishing industry
These companies are listed below and should be viewed as competitors, nevertheless we stand out with differentiation by providing the Njord SF platform