radar-satellite-analysis – skill Dữ liệu & Khoa học cho agent AI

Radar and Satellite Analysis

Overview

Analyze radar and satellite weather products as remote-sensing observations,

not as ground truth. Validate scan time, geometry, calibration, quality flags,

and coverage before interpreting a feature. Then separate what the instrument

directly measures from what a meteorological inference suggests.

This skill consumes already retrieved and decoded products. Use it after

noaa-radar-satellite-fetching or an equivalent source-specific skill. It does

not discover buckets, download files, repair outages, or make a product

scientifically suitable by naming it radar or satellite data.

When to Use This Skill

• Interpret NEXRAD reflectivity, velocity, spectrum width, or multi-moment

volumes.

• Track storm growth, decay, splits, mergers, rotation, anvils, or cloud-top

evolution over time.

• Analyze GOES visible, infrared, water-vapor, or derived cloud and convection

products.

• Combine radar structure with satellite cloud-top evidence while preserving

each sensor's limitations.

• Produce a short-term precipitation or severe-weather nowcast from a sequence

of observations.

• Explain why a radar or satellite feature is—or is not—supported by the data.

Do not infer a surface hail size, wind speed, rainfall rate, or lightning count

from a single proxy without stating the retrieval assumptions. Do not use a

single still image to claim a storm's future track or intensity.

Define the Remote-Sensing Analysis Contract

Before interpreting pixels, beams, or profiles, record:

• event, target phenomenon, region, and UTC interval;
• radar site, product level, scan/volume interval, elevation angles, moments,

and resolution—or satellite platform, product, sector, channel, scan mode,

and resolution;

• the measurement height or viewing geometry;
• the quality masks, calibration or scaling, and missing-data policy;
• the temporal cadence and the allowed lag between radar, satellite,

observations, and model guidance;

• the feature definition and the expected output: structure, motion,

intensity proxy, nowcast, or uncertainty map;

• the criteria for a confident, tentative, or inconclusive interpretation.

Keep observation time, scan start/end time, file creation time, retrieval time,

and analysis time separate. A file created quickly after a scan is not a newer

measurement.

Validate the Products Before Interpretation

Common checks

1. Confirm the radar site or physical satellite, product identity, sector,

channel or moment, and expected dimensions from the decoded metadata.

2. Verify the time interval, scan start and end, coordinate reference system,

and geolocation. For GOES fixed grids, use the product's geostationary

projection metadata rather than treating scan x and y as latitude and

longitude.

3. Apply the product's documented scale, offset, fill values, calibration, and

quality flags before computing statistics or thresholds.

4. Inspect missing scans, partial volumes, data gaps, saturated or invalid

values, and spatial coverage. Do not smooth over a gap without labeling it.

5. Record the resolution, footprint, and observation uncertainty. A high pixel

count is not a high spatial resolution if the beam or channel is coarse.

Radar-specific checks

• Confirm site, volume start/end time, sweep count, moments, and elevation

angles. Level II, Level III, and real-time chunk products are not

interchangeable encodings.

• Account for range-dependent beam width, beam height above the radar, terrain

blockage, range folding, attenuation, clutter, anomalous propagation, and

velocity dealiasing where relevant.

• Determine whether a feature is sampled by one low-level sweep, several

elevation angles, or a full vertical column. Do not compare a low-level echo

top with a satellite cloud top as if they were the same physical surface.

• Keep base reflectivity, column-integrated liquid, echo top, and derived

products distinct. Their thresholds and physical meanings are not

interchangeable.

Satellite-specific checks

• Confirm platform, instrument, product short name, channel or RGB product,

sector, scan mode, and scan interval from file attributes.

• Apply channel-specific calibration and quality flags. Brightness

temperature, reflectance, cloud-top temperature, and land-surface

temperature are different quantities.

• Treat cloud-top parallax as a viewing-geometry problem. A high cloud can be

displaced from its surface footprint, especially near the edge of a sector.

• Distinguish visible, infrared, water-vapor, and derived RGB products. Their

interpretation depends on reflectance, emission, atmospheric absorption, and

daylight or scan mode.

Analyze Radar Structure

Reflectivity and precipitation proxies

Use reflectivity to describe the location and organization of precipitation

echoes, then qualify the interpretation with beam height, range, attenuation,

and sampling. Look for features such as:

• high-reflectivity cores and their relationship to weaker surrounding echo;
• echo overhangs, weak-echo regions, bounded weak-echo regions, and vertical

development;

• bow, hook, line, multicell, or training structures when the geometry supports

that description;

• echo-top or vertically integrated liquid patterns that are relevant to the

stated question.

Do not call a reflectivity threshold hail, tornado, or destructive wind. Those

are conditional interpretations requiring velocity, cloud-top, lightning,

surface, or model context. Convert reflectivity to rain rate only with an

explicit relation, sample assumptions, and uncertainty bounds; a reflectivity

lookup is not a universal precipitation truth.

Velocity and rotation

Analyze radial velocity as a radar-relative measurement. State whether the

field is gate-to-gate shear, environmental shear, storm-relative radial flow, or

a derived couplet. Consider dealiasing, range folding, noise, side lobes,

velocity folding artifacts, and the fact that a velocity signature can have

multiple meteorological explanations.

For rotation or mesocyclone interpretation, use the appropriate temporal,

azimuthal, and vertical support and compare the result with neighboring scans.

A single couplet is a candidate feature, not a confirmed tornado. Do not use

radial velocity as a direct measurement of the environmental wind vector

without accounting for viewing geometry.

Three-dimensional organization

When a volume supports it, align elevation angles and analyze vertical

structure: low-level inflow, mid-level rotation, updraft organization, echo

top, overhang, and bounded weak-echo regions. Account for increasing beam

volume and decreasing resolution with height. A feature that appears vertically

stacked may be a sampling or attenuation artifact; check adjacent sites or

scans before asserting a continuous structure.

Analyze Satellite Cloud and Storm Structure

Use satellite data to describe cloud-top and environmental context, not to

replace radar precipitation structure. Depending on the product, examine:

• cloud-top temperature, brightness-temperature gradients, and overshooting

top candidates;

• anvil extent, spreading direction, and relationship to upper-level outflow;
• deep-convective cloud shields and mesoscale convective systems;
• visible-texture changes, cloud-top lowering or warming, and inferred growth or

decay only with a time sequence;

• water-vapor patterns, upper-level moisture, and potential convective

environments when the product and resolution support those claims.

An infrared cold cloud top is not, by itself, proof of severe weather, hail,

or a particular updraft strength. A very cold top can reflect a high cloud

whose emission and viewing geometry differ from a nearby lower cloud. State the

proxy, the threshold, and the alternative explanation.

Track Evolution and Motion

For a time sequence, use a consistent feature-identification and tracking

method. Track cell or cloud-system centers, echoes or anvils, growth and decay,

merges and splits, and changes in shape or intensity proxy. Distinguish motion

from the steering flow and from apparent motion caused by parallax, changing

scan geometry, or inconsistent feature definitions.

Estimate motion over multiple scans and report the method, time interval,

position uncertainty, and whether the feature was occluded or lost. A single

displacement between scans is not a reliable nowcast. If a forecast window is

short, say so and show the persistence or extrapolation assumption.

Combine Radar and Satellite Carefully

Align the products in time and space before joining them. Radar can provide

precipitation structure, radial motion, and vertical echoes; satellite can

provide broad cloud shield, cloud-top context, anvils, and environmental

patterns. Their footprints, resolution, viewing geometry, and quality are not

identical.

Use a joint interpretation only when the evidence is complementary. For

ex

… (nội dung đầy đủ có trong gói tải về)