Introduction to VCasT
VCasT (Verification and Forecast Evaluation Tool) is a Python-based library designed to compute, aggregate, and analyze statistical metrics for the evaluation of weather forecasts. It provides a flexible and extensible framework for verifying both deterministic and probabilistic forecasts across various spatial and temporal scales.
Developed at NOAA’s Global Systems Laboratory (GSL) in collaboration with CIRA at Colorado State University, VCasT is intended for research, development, and operational applications in numerical weather prediction.
What VCasT Does
VCasT enables users to:
Compute a wide range of verification metrics such as RMSE, MAE, bias, CSI, POD, FAR, GSS, FBIAS, Brier Score, correlation, and more
Filter forecast-observation pair data by time, variable, model, thresholds, lead time, and domain
Aggregate scores across user-defined groupings (e.g., by forecast lead time or region)
Compare the performance of multiple forecasts through pairwise statistical testing
Generate data-ready outputs for visual analysis or reporting
VCasT can process output from external tools like the Model Evaluation Tools (MET), making it easy to integrate into existing verification workflows.
Supported Formats
VCasT supports input from common meteorological and machine learning data formats, including:
GRIB2: Standard in operational weather modeling
NetCDF: Widely used in research and model development
Zarr: Optimized for cloud and AI workflows, commonly used with ML-based weather models
Why Use VCasT?
Focused: Built specifically for weather forecast verification
Lightweight: YAML-driven configuration with no required boilerplate code
Flexible: Easily integrated into larger Python-based workflows
Transparent: Encourages reproducible, configurable, and interpretable analysis
Use Cases
VCasT supports workflows such as:
Validating new weather model configurations
Comparing multiple ensemble or deterministic forecasts
Generating verification summaries for operational systems
Evaluating forecast skill across spatial domains and thresholds
Getting Started
To begin using VCasT, follow the Quick Start Guide