Brace Yourself — Seasonal Forecasts Are Coming!
7 months of seasonal weather forecast based on open-data ECMWF SEAS5 and EC46
TL;DR: The Seasonal Forecast API is now live 🎉. It combines ECMWF SEAS5 seasonal forecasts and ECMWF EC46 sub-seasonal forecasts to local forecasts at 36 km spatial resolution from weeks to months ahead. Users can access temperature, precipitation, and climate anomaly data in a single, easy-to-use API.
When planning for the future, a daily weather forecast is often not enough. That is where sub-seasonal and seasonal forecasts come in.
Seasonal forecasts help farmers plan crops, energy companies estimate heating or cooling needs, ski resorts manage snow, and communities prepare for floods or droughts by showing likely weather trends for the coming weeks and months.
Sub-seasonal forecasts cover roughly two to six weeks and help anticipate significant changes in weather patterns, such as cold spells, heatwaves, or periods of heavy rainfall. They bridge the gap between short-term weather predictions and longer-term climate outlooks.
Seasonal forecasts look even further ahead, spanning months. Instead of predicting specific daily conditions, they indicate the likelihood of broader trends, such as warmer or cooler temperatures than normal or above or below average rainfall. These forecasts rely on slowly changing aspects of the climate system, like ocean temperatures, soil moisture, and large-scale atmospheric patterns.
Seasonal forecasts are hard to come by because they are complex to produce, require massive computational resources, and are often only available in large, raw datasets. Ingesting this data can be difficult since it comes in specialized formats, covers multiple variables and ensemble members, and often requires processing and analysis before it can be used for local applications.
The European Centre for Medium-Range Weather Forecasts (ECMWF) runs both seasonal and subseasonal forecast models. Seasonal models, like SEAS5, provide outlooks for several months ahead, showing likely temperature and precipitation trends. Subseasonal models, such as EC46, fill the gap between weather forecasts and seasonal predictions, providing forecasts for up to six weeks ahead. Together, these models give a continuous view from the near-term to the seasonal scale.
The great news is that as of October 1, 2025, both EC46 and SEAS5 datasets are open data, meaning anyone can use them for research, applications, and services. With these datasets now publicly available, we developed a Seasonal Forecast API that combines the strengths of EC46 and SEAS5 and offer easy access to seasonal forecasts for everyone.
ECMWF EC46:
46 days forecasts with 100 perturbed members (50 accessible through Open-Meteo)
36-km spatial resolution with 6-hourly data
Updates daily
400GB of data per update
ECMWF SEAS5:
7 months forecasts with 50 perturbed members
36-km spatial resolution with 6-hourly data
Updates once per month on the 5th
800 GB of data per update
Model Updates
SEAS5 is updated only once per month, specifically on the 5th. While this provides a consistent seasonal outlook, it means that for the current month, the forecast may quickly become slightly outdated as new observations arrive.
EC46, on the other hand, is updated daily, providing more up-to-date information for the near term. By integrating EC46 with SEAS5, we can combine the long-range stability of the seasonal forecast with the higher temporal resolution of the sub-seasonal forecast. This ensures that forecasts for the current month reflect the latest observed conditions while still extending several months ahead.
Starting in 2026, seasonal forecast updates are expected twice per month.
Spatial Resolution
Both EC46 and SEAS5 use the O320 grid, which has a horizontal resolution of approximately 36 kilometres. For comparison, typical seasonal forecasts are only available at about 1° resolution, roughly 100 kilometres, so the 36 km resolution is a significant improvement in detail.
Seasonal and sub-seasonal forecasts are much more computationally expensive than short-term weather forecasts, which is why their resolution remains lower than the IFS O1280 grid used for high-resolution weather forecasts at 9 kilometres. While 36 km is generally sufficient for capturing broad-scale patterns, it is less reliable in areas with complex terrain or along coastlines, where local topography strongly affects the weather.
Open-Meteo provides data on ECMWF’s native O1280 and O320 grids without resampling to regular grids at 0.1° or 0.4° spacing, preserving the original model data as accurately as possible.
Ensemble Forecasts and Anomalies
Both EC46 and SEAS5 are ensemble forecasts, meaning they produce multiple simulations, or members, from slightly perturbed initial conditions. Each member represents a plausible evolution of the atmosphere, accounting for the inherent uncertainty in weather and climate systems.
The ensemble members are then compared against a model-climate, which is constructed from hindcasts over the past ten or more years. By doing this, we can calculate anomalies for upcoming weeks and months. For example, if a location is forecast to be +1 K warmer than the model-climate, this indicates a temperature that is above what the model typically simulates for that time of year.
Because anomalies are calculated relative to the model-climate rather than the real-world observed climate, they are removes systematic biases allowing the forecast to focus on deviations from expected conditions rather than absolute values, which can vary due to model limitations or resolution.
Examining the monthly mean temperature and anomaly for Berlin, Germany, December 2026 is forecast to have a mean of 1.5 °C. Since this is an absolute value, it may include a systematic bias, for example around 0.5 °C, so it should be interpreted with a grain of salt. The anomaly provides a clearer picture: at +0.5 K, December 2026 is expected to be slightly warmer than the model climate based on past years. This suggests that heating requirements at the start of winter may be typical or slightly lower than usual. For February and March, anomalies of +1 to +1.5 K indicate a reduced heating demand. Naturally, these values may change with each update of the seasonal forecast.
Looking at the weekly data from the ECMWF EC46 forecast provides more detail, but only for the next 46 days. For example, the week starting 10th November is predicted to be about 1.4 K colder than the model-climate, while the following weeks are expected to be roughly 1 K warmer.
Anomalies are not limited to temperature and are available for a wide range of weather variables. With the skiing season approaching in about a month, the snowfall anomaly is particularly relevant. In Zermatt, Switzerland, the forecasts for the next few weeks indicate less snow than the model-climate. The first week of December looks slightly more promising, with snowfall close to what is typical for this time of year.
It is important to remember that these forecasts represent statistical means and deviations from the model-climate. Extreme outcomes, such as no snow at all or significantly more snow than predicted, remain possible.
Anomaly Probabilities, Extreme Forecast Index and Shift of Tails
Seasonal forecasts go beyond deterministic weather forecasts and rely entirely on statistical methods. To provide more detailed insights, ECMWF also calculates probabilities of exceeding specific temperature or precipitation anomalies, as well as metrics like the Extreme Forecast Index (EFI) and Shift of Tails (SOT).
EFI shows potential extreme events in temperature or precipitation compared to the expected climate. Instead of providing absolute values, the EFI measures how unusual a forecast is relative to the model-climate.
For temperature:
EFI close to +1 indicates a strong likelihood of much warmer-than-normal conditions.
EFI close to -1 indicates a strong likelihood of much colder-than-normal conditions.
For precipitation:
EFI near +1 points to much wetter-than-normal conditions.
EFI near -1 points to much drier-than-normal conditions.
The Shift of Tails (SOT) is a complementary metric to the Extreme Forecast Index (EFI) that describes how extreme an event could become. While the EFI is bounded by the historical extremes of the model-climate, the SOT focuses on the tails of the forecast distribution — the rare, extreme events that may lie beyond what the model has previously simulated. The SOT is calculated for the 10th and 90th percentiles of all 100 EC46 ensemble members.
For temperature:
A positive SOT90 means that the forecast 90th percentile exceeds the model-climate 90th percentile, indicating a higher potential for exceptionally warm conditions.
A negative SOT90 means that the forecast 90th percentile does not exceed the model-climate 90th percentile, suggesting that the warmest ensemble members are not predicting unusually high temperatures.
Similarly, SOT10 can be used to assess the potential for exceptionally cold conditions.
For precipitation:
A positive SOT90 indicates an increased likelihood of very heavy rainfall or snowfall, beyond what is typical in the model-climate.
A negative SOT90 suggests that even the wettest ensemble members are not exceeding past model-climate extremes.
The EFI forecast for the week of 10 November 2025 indicates a strong likelihood of warmer-than-normal temperatures across the Rocky Mountains and colder-than-normal conditions along the East Coast. In the Rockies, EFI values exceed +0.7, suggesting a high probability of unusually warm temperatures across nearly all ensemble members.
For Denver, Colorado, the average temperature from all 100 ECMWF EC46 ensemble members shows an anomaly of +6.3 K, meaning conditions are forecast to be significantly warmer than the model-climate. A temperature EFI of 0.8 indicates a high likelihood of above-normal temperatures, while a SOT90 of 0.8 K suggests that the warmest 10 % of ensemble members point to the possibility of exceptionally high temperatures for this time of year.
In contrast, the following weeks show temperature anomalies near 0 K and EFI values close to 0, providing good confidence that temperatures are expected to remain near the model-climate average.
Together, EFI and SOT help forecasters assess how unusual or extreme upcoming weather may be and to better understand uncertainty in sub-seasonal forecasts. For further details, refer to the ECMWF documentation, which provides an in-depth explanation of both indices.
Available as API
The Open-Meteo Seasonal Forecast API is now available, providing seamless access to ECMWF EC46 and SEAS5 data for any location on Earth. Both forecast systems are combined into a unified dataset, offering a continuous view from sub-seasonal to seasonal timescales.
The API includes dozens of meteorological variables, available as:
6-hourly forecasts for 50 ensemble members
Weekly and monthly aggregated data for easier weather analysis
Detailed statistics including mean values, anomalies, probabilities, the Extreme Forecast Index (EFI), and Shift of Tails (SOT)
Forecasts can be retrieved for any coordinates, allowing users to bring global-scale seasonal predictions to a local level — for cities, smaller regions, or points of interest anywhere in the world.
Outlook
With ECMWF’s recent policy changes making more weather models available as open data, it is now possible to provide high-quality forecasts through a simple, easy-to-use API. This allows integration into websites, apps, and systems, helping users understand how weather may evolve over the coming months. The applications are diverse: estimating heating needs for local communities, optimizing water use for artificial snow at ski resorts, planning solar and wind energy production, or managing agricultural risks like drought.
ECMWF is also developing the next-generation SEAS6 seasonal forecast, which promises improved accuracy and updates twice per month. This is a massive undertaking, requiring significant computational resources to run the seasonal forecasts twice a month while also generating decades of hindcasts to serve as reference data. It is expected to become available in early 2026 and will be fully integrated into the Open-Meteo Seasonal Forecast API once available.
Artificial intelligence has the potential to play an exciting role in seasonal weather forecasting, primarily because running AI-based models can be significantly less computationally expensive than traditional numerical weather models. However, there are still major challenges to overcome. AI models must be able to accurately capture long-term drivers of seasonal climate, such as changes in sea surface temperatures, soil moisture, and large-scale atmospheric patterns.
With the growing trend toward open data, we look forward to integrating additional seasonal weather forecasts at high-resolution from other national meteorological services in the future.
Subscribe now to receive the latest announcements for free!
Thank you, Patrick, for making such an impressive step forward with the new Seasonal Forecast API. Open-Meteo is an extraordinary contribution to open science and accessible climate data.
My project CLIMA TRICORDER is proudly powered by Open-Meteo — bringing these forecasts to citizens in a simple and visual way.
Open data becomes life.