Big Data Platform for Smart Aquaculture
Step 4: Data Management Plan
The Data Management Plan (DMP) is a formal document that defines how data will be collected, stored, accessed, and shared.
It addresses privacy, data ownership, and intellectual property from the outset, and translates the stakeholder mapping and visioning outputs into concrete operational requirements. The DMP also specifies the functional and technical capacities the platform must deliver.
Functional capacities refer to what the platform must do for its users.
At a minimum, an effective platform must:
- Provide access to key stakeholders with clearly defined roles and permissions.
- Receive, store, and display data from sensors, manual entries, and external services in near real time.
- Generate alerts and notifications when monitored parameters exceed defined thresholds.
- Produce reports and summaries for regulatory and management purposes.
- Support disease surveillance by integrating sampling results and linking to outbreak databases.
- Enable spatial analysis by combining pond registry data with maps and satellite imagery.
- Facilitate multi-stakeholder collaboration through shared dashboards, communication channels, and data governance tools.
Technical capacities are the underlying infrastructure and analytical tools that enable the platform’s functions. These include hardware, software, connectivity, and AI/ML capabilities. There are many ways to describe the technology, and each platform will have it particular design, but in general, they all consist of layers for user management and governance, data storage and processing, data ingestion, data transformation, business intelligence (BI) and analytics and data observability.
In all layers, the use of AI/ML tools are increasingly used. The most relevant examples for aquaculture are AI/ML models that are trained to offer advanced predictive and generative AI analytics over the platform's data. When preparing for the use of AI/ML, careful consideration is needed to ensure cost-effective and reliable services. AI/ML can offer advantages over traditional approaches, however in many cases, traditional approaches still perform better, and care is needed when selecting a solution. AI/ML options are available across multiple levels of the system:
- At sensor level: AI/ML can be applied directly on edge devices or gateways (edge computing) to ensure signal quality, filter noise, detect sensor drift, and raise alerts when out-of-bounds signals arrive before data is transmitted to the cloud. Lightweight ML models for anomaly detection and fallback logic (e.g. trigger aerators if DO drops below threshold) can run on-site without internet connectivity.
- At farm level: Predictive models trained on historical sensor data can forecast pond conditions (dissolved oxygen, water temperature, pH, ammonia) over time horizons of hours to days. The SAB project demonstrated that hybrid statistical/ML approaches (combining ARIMA or regression with ML models such as Random Forest or LSTM) can achieve cost-effective and accurate forecasting for individual ponds.
- For multi-domain interpretation: Large Language Models (LLMs) and AI-assisted dashboards can help non-technical users interpret complex data, query the platform in natural language, and generate management recommendations. AI models can also integrate environmental, health, and economic data streams to produce composite risk indices. Smart systems will rely increasingly on AI services that can become autonomous agents that manage complex, end-to-end workflows independently.
- Geospatial analysis: Systems that include maps and geospatial analysis can integrate AI services including deep learning models such as Convolutional Neural Networks (with YOLO) applied to Copernicus Sentinel-1 radar data to detect pond boundaries across large areas, or Sentinel-2 optical time-series analysis to reveal pond status and detect changes in production activity.
To host these services, a core technical infrastructure is needed that includes: cloud-based storage and processing; a API gateway for system interoperability; telemetry protocols suited to local conditions (LoRaWAN for remote areas, 4G/5G where available, Wi-Fi/MQTT for farm control rooms); and data preprocessing pipelines with documented QA/QC procedures. Additional considerations must ensure that local legal and administrative needs are met, and that the Smart System can interoperate with other systems. National user authentication systems and alignment with ID providers is not an easy feat, but depending on the data that authorities and farmers can share, these can vastly improve understanding of aquaculture structure, assist with location-specific advice and warnings, and provide an inventory baseline after extreme weather events and other emergencies.
As the platform matures, additional functional and technical capacities may be added, including AI-driven forecasting, automated control systems, and value chain integration. A phased implementation approach is recommended, starting with core functions and expanding as institutional capacity and data quality improve. Countries are encouraged to build on existing systems (farm registries, disease reporting tools, weather services) rather than creating entirely new infrastructure.
Use the checklist below to verify that all key elements of this step have been addressed or refer to the overall checklist to ensure all recommended actions are planned or completed: