How it works
From data to decisions
How we accurately forecast the weather on sub-seasonal to seasonal timescales with our data-driven methods, enhanced with explainable AI and climate expertise.
Numerical weather prediction (NWP) model:
A type of model that uses thermodynamic laws to predict future weather conditions by simulating processes between the atmosphere and other climate variables.
- Traditional NWP models work by taking the current state of the atmosphere — with parameters like temperature, humidity, pressure, wind speed, and wind direction —, states of other climate components (such as the ocean) and mathematical equations to simulate how these conditions will evolve over time.
- These models struggle with simulating small-scale and fast processes, but precisely these processes are important for long-range weather predictability. This is especially true for Europe.
- A consequence of these unresolved small-scale processes is an underestimation of the connection-strength of the ocean and land.
Beyond Weather prediction model
Beyond Weather's AI model is built to find and extract teleconnections through a knowledge-guided automated extraction of slow-moving components in our climate system, which significantly outperforms even the most state-of-the-art seasonal NWP models.
- Beyond Weather’s AI-model relies on recognizing patterns and correlations in historical data and using those to predict future conditions.
- The key differences between our AI-enabled, data-driven & response-guided model and traditional NWP models are:
- AI-enhanced: Machine learning identifies complex patterns in climate data that might be missed by traditional methods.
- Climate expertise included: We don't just rely on AI, we combine AI with the climate expertise of our scientists and years of academic research to deliver the best of both worlds.
- Predictive Horizon: The Beyond Weather algorithms significantly outperform the current best-in-class NWP prediction models on (sub-) seasonal timescales.
- AI is no black box to us: we can show where the predictive power is coming from for a particular forecast. We give insight into the sources of our predictability, unlike traditional NWP models.
Why traditional NWP models fail to accurately predict on sub-seasonal to seasonal timescales
Cracking the code of subseasonal-to-seasonal (S2S) weather predictability presents a completely unique challenge compared to forecasts of up to two weeks. A fundamentally different approach is required
Traditional NWP models predict for up to two weeks by taking the current state of the atmosphere — with parameters like temperature, humidity, pressure, wind speed, and wind direction — and using equations to simulate how these conditions will evolve over time.
The further traditional NWP models are trying to predict into the future, the poorer the forecast quality becomes.
The precision of their predictions begins to blur due to the sheer chaos of atmospheric interactions. This happens due to the growth of random small-scale disturbances, affecting larger and larger – finally – synoptic-scale weather systems. The fascinating chaotic chain of events is affectionately referred to as ‘the butterfly effect’.
In order to solve this challenge, we need to step away from using models which are primarily based on physical equations that drive simulations.
How Beyond Weather solves this challenge
Beyond Weather's AI-model relies on recognizing patterns and correlations in historical data and using those to predict future conditions. It's a fundamentally different approach, focusing more on empirical relationships rather than physical laws.
The growth in climate data and artificial intelligence has paved the way to solve this problem. Artificial intelligence allows us to systematically search for and discover connections in our climate system that can boost predictive power on a longer timescale. By learning from observations from more than 70 years of climate data and predicting from observations directly, we can circumvent the issues of numerical weather predictions.
Our model systematically searches for important teleconnections.
Traditional NWP models for weather prediction often fall short, largely due to their inability to fully grasp the significance of climate system interplays, known as teleconnections. These teleconnections aren't just trivia; they are critical catalysts driving our weather patterns over extended periods.
Take, for instance, the well-known phenomenon known as ENSO—El Niño/La Niña oscillation. It’s a climate event occurring in the Pacific Ocean’s equatorial region. Here, the water’s temperature either warms (El Niño), cools (La Niña), or holds steady. And this has an astounding ripple effect on global weather patterns. Certain areas become more humid, while others experience drier conditions. Traditional NWP models, with their grid-based system, often fail to fully grasp the impact of these phenomena.
Our prediction model uses machine learning to accurately seek out hidden teleconnections within a vast matrix of climate variables.
AI, specifically machine learning, identifies complex, nonlinear relationships in data that might be missed by traditional statistical methods. Our AI-driven model recognizes patterns and learns from vast amounts of data, making it ideal for handling the complexities of climate data. The model’s ultimate goal is to pinpoint those regions that play a pivotal role in shaping specific outcomes, such as winter temperatures in a place like the Netherlands.