NEXRAD Mosaic Construction

Overview

Build a custom multi-radar composite from aligned, validated single-site data. Every output pixel must have a traceable source, coverage state, and quality decision. Preserve the distinction between a local analysis product and an official provider mosaic.

Use this skill after products from multiple radars have been resolved. It does not discover, download, or decode source files. Use nexrad-product-access for single-site access and nexrad-mosaic-access when the required result is an existing NOAA MRMS composite rather than a newly constructed analysis.

When to Use This Skill

• Combine two or more radars to cover a region with overlapping or adjacent fields.
• Construct a consistent base-reflectivity, velocity, spectrum-width, or dual-polarization composite.
• Build a research mosaic on a common grid or for a cross-radar comparison.
• Compare a custom mosaic with an official MRMS product.
• Visualize source contributions and coverage gaps in a composite.

Do not use this skill to fetch an official MRMS product, select a single radar, or claim that a simple pixel average is a physically resolved regional field.

Define the Mosaic Contract

Record:

• product and exact units from each source;
• source sites, coordinates, volume start/end times, and sweeps;
• target region, output grid, projection, and cell size;
• time-matching and temporal interpolation policy;
• native range limits, beam width, beam height, and vertical sampling;
• quality-mask, clutter, attenuation, range-folding, and dealiasing policy;
• overlap-resolution and quality-weighting method;
• treatment of coastlines, terrain, blocked sectors, and missing gates;
• output format, metadata, and provenance requirements.

Do not combine a Level II moment and a Level III display product merely because their names resemble one another. Their processing, resolution, and quality semantics must be verified equivalent or the difference must be represented explicitly.

Validate and Align Source Data

Before compositing:

1. verify each site's decoded metadata, time, product, units, and quality;
2. convert all source fields to one declared projection and common output grid;
3. align source times using a documented tolerance and interpolation policy;
4. preserve each source's native spatial support and quality fields;
5. reject a source that is too stale, out of range, or outside its valid product contract;
6. label any resampling or interpolation rather than presenting the output as a native-resolution observation.

When a common time cannot be formed without excessive interpolation, use a nearest time within the stated tolerance, retain the actual time difference, or mark the output unavailable. Do not average distant scans into one pseudo-time without showing the temporal support.

Define the Output Quantity and Quality

For reflectivity, retain dBZ or the product's native calibrated unit and use a quality-aware combination rule. Do not average dBZ as a linear physical quantity without a stated rationale. For velocity, preserve positive and negative radial-velocity conventions and handle dealiasing and folding before compositing. For dual-polarization moments, preserve their physical units and quality masks.

Keep these fields separate:

• measured or decoded value;
• quality or suitability score;
• source site and source time;
• beam-height or range metadata;
• interpolation or resampling state;
• coverage and no-data state.

The output should make it possible to identify which radar contributed each pixel, especially near overlap boundaries.

Select Candidate Data by Quality and Geometry

For each output cell, consider only sources that provide valid coverage at a compatible time and product level. A useful selection score can include:

• lower beam height at the target location;
• shorter range and smaller beam width;
• fewer terrain or clutter artifacts;
• better data quality or signal-to-noise status;
• newer scan or smaller time difference;
• lower attenuation or range-folding penalty;
• product-specific suitability such as dual-pol quality.

The score is a modeling choice, not a universal truth. State its terms and weights, normalize only comparable components, and test sensitivity to the selection rule. Never hide the score behind an unexplained “best radar” rule.

Resolve Overlaps Deterministically

Use a deterministic rule for every target cell:

…