AI-DOP | Artificial Intelligence-Direct Observation Prediction | ECMWF

AIFS-DOP

We introduce the Artificial Intelligence Forecasting System for Direct Observation Prediction (AIFS-DOP). AIFS-DOP is trained on

• a 40-year harmonized dataset of gridded observations, without using numerical weather prediction (NWP) reanalysis or model data.

The resulting model is competitive with ECMWF's Integrated Forecasting System (IFS) when scored on a one year period of forecasts across 2021/2022. This progress on Direct Observation Prediction represents the first time that a data-driven model, trained solely on observations, is competitive with the IFS at medium ranges for several key upper-air and surface headline scores, when verified against observation data.

• An update on AI–DOP: skilful weather forecasts produced directly from observations
• Machine learning opens new opportunities for global reanalysis | 20 July 2026
• AIFS-DOP: End-to-End Medium-Range Weather Prediction from Observations Alone with Machine Learning | arxiv
  • The model uses an encoder-processor-decoder architecture, using a graph-based attention encoder/decoder and a transformer processor with sliding window attention as described in Lang et al. [2024].

Figure 1: High-level model schematic: A single encoder is used for all observation types. The processor is as described in Lang et al. [2024] using a residual connection. Then a single decoder to predict observations out onto a full grid.

• A single decoder to predict observations out onto a full grid.
• AIFS-DOP operates on an O96 octahedral reduced Gaussian grid, which has a horizontal resolution of approximately 1-degree (100 km to 112 km).