Adapt#

Real-time detection, projection, and tracking of convective storms in weather radar data — built for operational adaptive scanning at the DOE ARM User Facility, and for large-scale generation of value-added products and AI-ready storm datasets.

Health

Install

Project

Note. Adapt is under active development. External pull requests are not being accepted at this time; contribution guidelines will follow the first stable release. APIs, configuration files, database schemas, and output formats are subject to change without notice.

Documentation: arm-doe.github.io/Adapt — Installation · User Guide · CLI Reference

Overview#

Adapt is a modular framework for real-time convective cell detection, motion projection, and storm lifecycle analysis. It processes a stream of radar volume scans and, for each cell, produces a tracked identity, a set of measured properties, and a record of lifecycle events — initiation, growth, splitting, merging, and decay. Results are written to a structured, queryable repository (NetCDF, Parquet, and SQLite) exposed through a read-only Python API, rather than as loose output files.

The processing pipeline is composed of independently registered modules — ingest, detection, projection, analysis, and tracking — assembled automatically at run time. New algorithms and data sources are added by registering a module, not by modifying the pipeline. Every run is deterministic and reproducible: runs are registered, derived quantities are traceable to their inputs, and identical inputs produce identical tracks.

Motivation#

Adapt is developed to support adaptive scanning at the U.S. Department of Energy’s Atmospheric Radiation Measurement (ARM) User Facility, where the C-Band Scanning ARM Precipitation Radar (CSAPR2) requires a real-time source of high-value scan targets, identified and prioritized automatically rather than by a fixed volume-scan schedule. The same pipeline runs unchanged on historical data, supporting research applications such as storm lifecycle studies, radar–lightning correlation analysis, and the construction of analysis-ready, multi-sensor datasets combining radar, satellite, and Lightning Mapping Array observations.

From operations to large-scale analysis#

The pipeline that selects targets in real time also reprocesses years of archived data, with identical, traceable results. Its tracked cells, lifecycle events and decision logs make ready-made value-added products and consistent datasets for AI-driven research.

Current status#

Adapt is at an alpha stage of development.

The reference NEXRAD pipeline (ingest, detection, projection, analysis, tracking) runs end to end in real-time and historical modes.
A read-only query API and GUI dashboard provide access to pipeline output.
Post-processing modules enrich completed storm tracks with observations from other instruments.
Rule-based target selection for adaptive scanning is under active development.
Supported platforms: macOS, Linux, and Windows. See Installation.

No backward compatibility is guaranteed for APIs, configuration, or generated data products until the first stable release.

Roadmap#

Planned work extends Adapt toward a general, multi-sensor storm-analysis framework: pluggable tracking and segmentation backends, a data-agnostic variable layer so new instruments are a configuration change rather than a code change, and integration with ARM radar-control systems to close the loop between detection and adaptive scanning. See Vision for details.

Funding#

Adapt is supported by the U.S. Department of Energy as part of the Atmospheric Radiation Measurement (ARM) User Facility, within the Office of Science.