About the S2S API

Climavision's Subseasonal to Seasonal (S2S) API is powered by a cutting-edge machine learning model built for long-range, monthly weather forecasting. Developed using an ensemble of 30 high-quality Earth System Models (ESMs), the S2S AI model learns from global climate simulations to produce skillful, physics-informed predictions across a wide range of variables. The system is designed to capture both large-scale climate behavior and localized variability without relying on real-time observations. Forecast outputs from this model are delivered directly through our API, providing seamless access to globally gridded, probabilistic climate forecasts up to two years in advance.

The model produces full predictive distributions for each forecast, capturing not only the expected value but also the uncertainty, skewness, and extremes. Forecasts are available globally at 1-degree resolution, with output smoothing options over 1, 3, and 12 months, enabling both seasonal and multi-annual planning horizons. Forecast lead times range from 1 to 24 months depending on the variable and time span, with each forecast leveraging a full year of prior global conditions to inform its predictions.

The API currently provides forecasts for key weather variables including 2-meter temperature, 10-meter wind speed, solar radiation, and precipitation. Additionally, derived metrics such as weighted degree days are provided to enable a more direct integration with energy demand modeling. Forecasts are probabilistic, enabling risk-aware decisions, and are updated regularly to reflect the latest model outputs. Forecasts are available in monthly increments and cover the entire globe, with validation metrics included to assess historical skill and reliability. These metrics, including CRPSS, ACC, and MAE, are available back to 1993 and are broken down by month, providing clear transparency into forecast performance over time.

Climavision’s S2S model exhibits especially strong skill over land and consistently outperforms ECMWF’s SEAS5 across leads 1 to 5, with reliable forecasts extending out to two years, well beyond SEAS5’s six-month range. This long-range capability, delivered through a flexible API, makes it well suited for energy planning, climate risk management, and other applications that depend on extended-range forecasts. Comparative verification metrics against SEAS5 are available upon request.

Each forecasted variable in the API is governed by scientific principles that define its inherent predictability and lead-time behavior. These principles are embedded in the modeling approach to ensure physical consistency and realistic climate evolution.