Building Best-in-Class LNG Destination Forecasting Models

At Kpler, we understand how critical LNG vessel destination forecasting is for market professionals. Accurate predictions help traders, charterers, and analysts anticipate supply availability, spot emerging market opportunities, manage risks, optimize logistics, and fine-tune trading strategies — all of which translate into competitive advantages in a volatile LNG landscape. That’s why we’re on a quest to deliver the highest-quality data, constantly challenging and improving our solutions. Leveraging advanced Machine Learning techniques and user feedback, we’ve just released a major upgrade to our LNG destination forecasting, offering sharper accuracy and stronger decision-making support.

Enhancing Forecasts: Adapting to Commodity-Specific Needs

Kpler’s Destination Forecast has been a critical feature for commodity market participants, offering valuable insights into vessel movements , upcoming port calls and the resulting flows. It has empowered clients to assess future S&D balances, optimize operations and make informed decisions.

We have reassessed our approach, incorporating even more our customers’ feedback, performance assessments and industry insights. We are excited to introduce specialized machine learning models tailored to individual markets - starting with LNG - to account for the unique characteristics and dynamics of each, something that is unique among vessel destination forecast providers. These new models ensure our forecasts remain precise, reliable, and actionable for Kpler’s users.

Building Better Predictions: Our Approach to Model Enhancement

Rather than simply tweaking our existing models, we took a more ambitious approach - redesigning the entire forecasting pipeline and developing specialized machine learning models tailored specifically for LNG commodity flows. 

What’s new?

• Specialized Machine Learning models designed for LNG flows, operating at multiple geographical levels.
• A reviewed confidence-driven framework that dynamically updates forecasts as voyages progress.
• Enhanced feature sets across five key dimensions - vessels, routes, durations, visits, and flows - delivering more granular and precise insights.

With this approach, LNG-specific trade patterns are fully embedded into our predictive framework, ensuring more accurate and market-aware forecasts. This enhancement enables the generation of modeled port calls, giving you deeper foresight to capitalize on market opportunities. Furthermore, each future port call will include a confidence score, providing a clearer assessment of prediction reliability and helping you make more informed decisions.

The Road to Accuracy: How We Revamped Our Models

The development of our enhanced LNG Destination Forecast models followed a rigorous and iterative process, ensuring maximum accuracy and reliability. Our team undertook an extensive phase of exploration, model revamp, and validation, training specialized models on historical LNG flows. This process leveraged vast amounts of data from past LNG trades, vessel movements, and market conditions to refine predictions.

Key Steps in the Revamp Process:

• Collaborating with LNG market experts to identify high-impact levers for improving destination forecast accuracy, drawing on deep domain knowledge on LNG trade behavior and routing patterns.
• Running extensive analysis on historical data - including vessel AIS, trades and port calls records - to develop and validate hypotheses around key influencing factors such as vessel status (laden/ballast), LNG routes and flows seasonality.
• Engineering key features from raw data into a feature vector comprising hundreds of variables that the destination forecast models could learn from.
• Training multiple models to predict LNG vessel destinations, starting with statistical models as baseline and progressing to advanced machine learning models capable of capturing temporal and spatial complexities.
• Systematically evaluating model accuracy and feature importance through rigorous backtesting and feature importance analysis, identifying the most influencing factors and measuring predictive performances.
• Conducting iterative experiments and fine-tuning cycles resulting in best-in-class models that achieved significant accuracy gains.

‍

Closing the Loop: Model Validation with Market Experts

Ensuring the reliability and relevance of our Destination Forecast models is a top priority. During development, we conducted a rigorous validation phase in collaboration with Kpler’s LNG analysts, whose deep market expertise allowed us to continuously assess the model’s outputs against actual vessel movements.

This collaboration enabled us to incorporate expert insights, fine-tuning and validating forecasts against real-world scenarios. At the same time, we leveraged and refined our automated Performance Analytics to monitor and track forecast accuracy, completeness, and stability - ensuring continuous improvement and reliability.

‍

‍

The Results Are In: Accuracy Gains

We are proud to announce significant accuracy improvements in our new LNG Destination Forecast models. Through the extensive testing and validation described above, our best-in-class models now deliver substantial accuracy gains across all geographic levels.

‍

Key Results:

• Accuracy improvements of up to 45% in predicting LNG vessel destinations.
• A major step forward in providing the most reliable data for the LNG market.

These enhancements reinforce our commitment to delivering the most precise and actionable insights. With these improvements, we are confident that our LNG Destination Forecast models will empower you with unprecedented precision for critical decision-making.

Experience the difference  - our enhanced forecasts will help you navigate the complexities of LNG flows with greater confidence and seize market opportunities with unparalleled accuracy.