Maritime and Coastguard Agency: Themis
Themis is an Electronic Maritime Drift modelling system used to calculate predicted drift of leeway objects using deterministic and probabilistic modelling.
Tier 1 Information
1 - Name
Themis Search Planning Tool
2 - Description
The tool is used to predict the drift of objects, it is being used by HMCG to calculate areas to search for objects at sea. The tool uses Mobidrift particle modelling (Monte Carlo modelling); and IAMSAR Modelling (deterministic modelling) to make these predictions.
3 - Website URL
N/A
4 - Contact email
HMCG Technology Team Service Owner - hmcgtech@mcga.gov.uk
Tier 2 - Owner and Responsibility
1.1 - Organisation or department
His Majesty’s Coastguard
1.2 - Team
HMCG Technology Team
1.3 - Senior responsible owner
Director of HM Coastguard
1.4 - External supplier involvement
Yes
1.4.1 - External supplier
COLLECTE LOCALISATION SATELLITES (CLS) UK LIMITED
1.4.2 - Companies House Number
06635456
1.4.3 - External supplier role
Developer and supplier of the tool provided as a fully managed service to HMCG.
1.4.4 - Procurement procedure type
Competitive tender
1.4.5 - Data access terms
Under the terms of procurement rules. The supplier does not have any access to government data sets. The MCA has purchased data sets which can be digested by the System, such as data from the UK Hydrographic Office, but this agreement is between the MCA and UK HO and not the THEMIS Supplier. The supplier does have the ability to access the data stored within the system, but this is for archiving or fault finding and personal data isn’t accessed, only the basic system info.
Tier 2 - Description and Rationale
2.1 - Detailed description
Themis is a system used in maritime and aeronautical Search and Rescue (SAR) operations. It helps predict the drift of objects at sea, such as vessels, debris, or individuals, by calculating their probable movement over time. This is achieved by combining data on environmental factors like wind, currents, and tides with advanced modelling techniques.
Themis uses two primary modelling approaches to predict drift:
- Monte Carlo Particle Modelling (Mobidrift):
- This method involves simulating the movement of numerous “particles” from a starting point (e.g., the last known position of an object). Each particle represents a possible trajectory based on environmental conditions.
- Monte Carlo modelling is probabilistic, meaning it accounts for uncertainties and variations in the data. By running multiple simulations, it generates a range of possible outcomes, which are then used to identify the most likely drift patterns.
- The advantage of Monte Carlo methods lies in their ability to handle complex, multi-dimensional scenarios, making them ideal for SAR operations where conditions can vary widely.
- IAMSAR Deterministic Modelling:
- Based on the International Aeronautical and Maritime Search and Rescue (IAMSAR) guidelines, this approach uses fixed inputs like wind speed, current direction, and leeway factors to calculate a single, deterministic drift path.
- Deterministic models are straightforward and provide a clear prediction, but they may not account for the full range of uncertainties in real-world conditions.
The Maritime and Coastguard Agency (MCA) employs Themis because it integrates both probabilistic (Monte Carlo) and deterministic (IAMSAR) models, offering a comprehensive toolkit for SAR operations that include:
- Search Area Evaluation: Themis calculates the most probable search areas by combining drift predictions with error margins, ensuring resources are deployed effectively.
- Flexibility: Depending on the scenario, operators can choose between Monte Carlo modelling for probabilistic insights or deterministic modelling for precise predictions.
- Compliance: Themis aligns with international SAR standards, making it a reliable choice for global operations.
The reasons for using the Monte Carlo and Deterministic Models are: - Monte Carlo Modelling: This approach is invaluable for handling uncertainty. By simulating multiple scenarios, it provides a probability map of where the object might be, helping prioritize search efforts. - Deterministic Modelling: While less flexible, deterministic models are faster and simpler, making them useful for straightforward cases or as a baseline for comparison.
Together, these tools enable SAR teams to make informed decisions, optimize search efforts, and increase the likelihood of successful rescues.
2.2 - Scope
Themis is primarily employed in maritime and aeronautical Search and rescue (SAR) operations when there is a need to predict the drift of objects or people at sea to locate them efficiently. It is used in scenarios like:
- Distress Signals: When a vessel, aircraft, or person has gone missing, and their last known position needs to be analysed.
- Debris Tracking: For identifying likely locations of floating debris.
- Search Planning: Determining the most probable areas for SAR efforts, optimising resource allocation.
- Incident Analysis: Supporting investigations by modelling movement patterns of objects over time.
Themis may not be used in situations where: - Environmental Data is Insufficient: Accurate drift modelling depends on reliable data about currents, winds, and tides. - Small-scale or localised incidents: For example, when the object’s position can be determined visually without extensive modelling. - Non-drift Related Cases: If an incident doesn’t involve movement influenced by environmental factors (e.g., structural failures unrelated to drift).
Themis offers several advanced features: - Drift Prediction: Calculates the movement of objects based on environmental conditions like wind and current. - Search Area Estimation: Defines probable areas to focus SAR efforts. - Integration of Models: Combines Monte Carlo particle modelling and deterministic approaches for flexibility. - Visualisation: Provides visual maps for operational planning.
Themis’ limitations include: - Data Dependency: It requires accurate input data to deliver reliable predictions. - Limited Non-drift Applications: It doesn’t address incidents unrelated to drift, such as submerged or stationary objects. - Human Oversight Needed: Decisions rely on trained SAR professionals to interpret outputs effectively.
Themis is used by SAR teams, including maritime authorities, coastguards (like the MCA in the UK), and aeronautical agencies due to: - Efficiency: It optimises the allocation of SAR resources, saving time and increasing the chances of success. - Accuracy: By modelling complex drift patterns, it improves prediction reliability. - Standard Compliance: Aligns with international SAR guidelines (e.g., IAMSAR).
2.3 - Benefit
Search and Rescue (SAR) search planning software offers several key benefits:
Enhanced Efficiency: These tools use advanced algorithms and simulations to optimise search patterns, significantly reducing the time and resources needed to locate missing persons.
Improved Accuracy: By integrating real-time data and historical information, SAR software increases the accuracy of search efforts, leading to higher success rates in finding individuals.
Better Coordination: SAR software facilitates seamless communication and coordination among multiple teams and agencies, ensuring that all efforts are aligned and information is shared promptly.
Resource Management: These tools help in managing and deploying resources effectively, ensuring that personnel and equipment are used where they are most needed.
Real-Time Updates: SAR software provides real-time updates and situational awareness, allowing teams to adapt quickly to changing conditions and new information.
2.4 - Previous process
Themis has replaced a tool called SARIS which was a system provided by BMT (Previous search planning software supplier) used within an electronic maritime search planning tool by MCA staff.
2.5 - Alternatives considered
Open competitive tender was followed. Search planning software to match all requirements is a fairly niche spectrum with market engagement and collaboration having to be used to support procurement.
Tier 2 - Decision making Process
3.1 - Process integration
Themis is used by Martine Coastguard Agency staff members as part of missing object/person search. Staff will need to use Mobidrift which is a drifting algorithm based on the Monte-Carlo method within the Themis application. It is used during the Search Area Evaluation phase i.e. when the SMC has to calculate a search area for a missing object/person based on a last know position or expected route. The Mobidrift service is called thanks to an API by submitting THEMIS Search and Rescue (SAR) input data. The tool then outputs the predicted trajectories. MCA staff can then overlay these trajectories onto a map which they then plan their search and rescue actions.
3.2 - Provided information
The information returned by Mobidrift is data formatted in XML that contains, for each particles, all the points of the their trajectory. The data returned can be heavy as the system allow a maximum of 10000 particles drifting for several hours/days and a customised timestep.
3.3 - Frequency and scale of usage
Only HM Coastguard qualified staff have access to the system, as per training and assessment. We are unable to detail how many times the system is used, as its dependant on a search and rescue mission and not all missions require the tool.
3.4 - Human decisions and review
The operator uses the output from the system to assess against search parameters and external factors to derive a decision on the correct search area. thereafter passing it to search and rescue units to quantify and complete the search, providing feedback and thereafter adjusting any search parameters.
3.5 - Required training
Comprehensive Training package delivered in house with trainers attending train the trainer sessions with the development team and product owners. Operators thereafter sit assessments on their ability to use the product.
3.6 - Appeals and review
There are circumstances whereby the data set may not be as reliable as other areas, for example under a bridge or within a harbour whereby variables can skew the data set. Operators are allowed to use local judgement on search parameters but it has to be justified and documented.
Tier 2 - Tool Specification
4.1.1 - System architecture
The THEMIS SAR solution is a modular web-based system composed of several components that communicates with standard Electronic Data Interchange mechanisms. The system is hosted on a cloud environment offered by Microsoft Azure, deployed across two different Microsoft Azure regions.
THEMIS is in the core system is made up of several services, all interconnected via a private network:
The database service relies on a cluster of PostgreSQL servers
A data processing service layer where all the software components run in Docker images, and orchestration (deployment, lifecycle) is managed by Kubernetes
Mapping Service Architecture Overview
The database service is a PostgreSQL database cluster made of four servers (1 primary, 2 standby and 1 witness), spreads on all availability zones of the region. A failover system is installed to ensure that a standby server is automatically promoted as new primary server if the current primary one went down. Moreover, a load-balancer is deployed into Kubernetes cluster to ensure database connections from application is only sent to available PostgreSQL servers.
Data processing service is deployed as a set of Docker containers in a Kubernetes cluster. Containers are like VMs, but they have relaxed isolation properties to share the Operating System (OS) among the applications.
Virtual Machines (VMs) are run on a single physical server’s CPU. Virtualisation allows applications to be isolated between VMs and provides a level of security as the information of one application cannot be freely accessed by another application.
In deployment, Azure Kubernetes managed service (AKS); is used to manage deployment and lifecycle of containers.
4.1.2 - Phase
Production
4.1.3 - Maintenance
Each time a new feature or bug fixing is undertaken or added to Mobidrift, non-regression tests are undertaken. MCA run a clone of old drifts to ensure the results are in accordance with previous performance.
4.1.4 - Models
Mobidrift (Monte-Carlo method)
Tier 2 - Model Specification
4.2.1 - Model name
Mobidrift
4.2.2 - Model version
2.3.14
4.2.3 - Model task
The model is tasked to predict the route and position of an individual/object based on the last known position and environment.
4.2.4 - Model input
Last know position or expected route (waypoints), date, end of the drift (datum time), drifting object, errors, environmental data, number of particles.
4.2.5 - Model output
A predicted trajectory displayed as a list of particle points predicted during the drift.
4.2.6 - Model architecture
Type : Monte-Carlo - probabilistic Description : Monte Carlo methods, or Monte Carlo experiments, are a broad class of computational algorithms that rely on repeated random sampling to obtain numerical results.
4.2.7 - Model performance
The duration of the model calculation is very variable. It depends on the number of particles, the drift elapsed time, the number of search objects, the number of models running in parallel. user acceptance testing (UAT), scripting, and testing:
- User Acceptance Testing (UAT):**
- Verifies the system meets user requirements and business goals.
- Conducted by end-users to validate functionality in real-world scenarios.
- Focuses on usability, reliability, and compliance with specifications.
-
Involves pre-defined test cases and acceptance criteria.
- Scripting:**
- Creates automated or manual scripts to perform specific test scenarios.
- Defines step-by-step instructions for execution.
- Ensures consistency and repeatability in testing.
-
Often written using testing tools or programming languages like Python or JavaScript.
- Testing:**
- Identifies defects and ensures quality before deployment.
- Includes various levels (e.g., unit, integration, system, UAT).
- Combines manual and automated approaches for thorough coverage.
- Iterative process to refine and improve the product.
All testing was conducted with Subject Matter Experts and were signed off as satisfactory during acceptance of the product. During regular update processes further user acceptance testing is conducted to maintain functionality and output.
4.2.8 - Datasets
Historic scenarios or a hydrographical data set were used for the testing and UAT purposes. It would have been used to support training.
4.2.9 - Dataset purposes
The Dataset part was not applicable for the Mobidrift Algorithm
Tier 2 - Data Specification
4.3.1 - Source data name
Environmental Data source - - Marine Weather Forecasts: Detailed forecasts for coastal and offshore areas, including wind speed, wave height, and visibility.
- Shipping Forecasts: Regular updates for seafarers, covering specific sea areas around the UK.
- Inshore Waters Forecasts: Focused forecasts for areas up to 12 miles offshore, tailored for smaller vessels.
- Warnings: Alerts for severe weather conditions, such as gales or storms, to ensure maritime safety.
- Wave Model Data: Predictions of wave heights and directions, aiding navigation and planning.
- Tidal Information: Data on tidal heights and times for various locations.
These feeds are essential for safe and efficient maritime operations
4.3.2 - Data modality
Multimodal
4.3.3 - Data description
The data set is provided to us by the Met Office. We would be unable to define what data is capture to create these data sets
4.3.4 - Data quantities
Each output is its own individual model, so there is no development or training. Each model is unique based on the unique information entered due to the situation and the empirical data used. Aside from the forecasted environmental models provided to us by a third party; Met Office.
4.3.5 - Sensitive attributes
N/A
4.3.6 - Data completeness and representativeness
N/A
4.3.7 - Source data URL
N/A
4.3.8 - Data collection
Model outputs and environmental data provided by Met Office is the only data collected and stored.
4.3.9 - Data cleaning
N/A
4.3.10 - Data sharing agreements
N/A
4.3.11 - Data access and storage
The data is accessible to any HM Coastguard search planning trained member of staff, MCA ICT, HM Coastguard HQ staff and CLS. Each model that is developed within the system is stored my the supplier ‘CLS’ on behalf of the MCA for the length of the contract. At this point it will be transferred to the MCA. The MCA will retain all operational models for a total of 20 years, from creation, in line with the HM Coastguard incident data policy.
Tier 2 - Risks, Mitigations and Impact Assessments
5.1 - Impact assessment
Screening assessment completed in 2021 - no personal data in system
5.2 - Risks and mitigations
Mitigations: Operators are able to fallback to a manual search planning process, however there is also the ability to peer review or request international assistance from other countries using their search planning software. This would be an extreme case, however there are processes in place to support this. Additionally search and rescue units are capable of planning and delivering a search in a condensed format to sustain operations.