What’s Falling: Hydrometeor Classification

A Hydrometeor is any water or ice particle in the atmosphere: cloud droplets, raindrops, snowflakes, graupel, hail. Dual-polarization gives the radar several measurements for every gate. The Hydrometeor Classification Algorithm (HCA) combines them into a single best guess at what kind of target fills each gate. In LightningWX that product is called What's Falling.

What Data Does Hydrometeor Classification Use?

The WSR-88D HCA uses:

• Reflectivity (Z), related to size and number.
• Differential reflectivity (ZDR), related to shape.
• Correlation coefficient (CC), related to how diverse the targets are.
• Specific differential phase (KDP), related to liquid water content.
• Texture of Z and differential phase, measuring how noisy the fields are from gate to gate. Ground clutter and birds are noisy; rain is smooth.
• The melting layer, estimated from dual-pol data and model temperatures, which tells the algorithm whether a gate is above, in, or below the level where snow melts.

How Does Fuzzy Logic Classify Precipitation?

Real categories overlap. A 55 dBZ echo might be heavy rain or rain mixed with hail. Instead of hard yes-or-no thresholds, the HCA uses Fuzzy Logic:

1. For each category, each input has a membership function: a curve describing how well a given value fits that category, from 0 (does not fit) to 1 (fits perfectly). For example, a CC of 0.99 fits "rain" very well but "biological" poorly.
2. Each input is given a weight reflecting how useful it is for that category.
3. The algorithm adds the weighted scores for every category.
4. The category with the highest total score wins for that gate.

The melting-layer information then restricts which categories are allowed. Rain is not allowed far above the melting layer, for instance, and dry snow is not allowed well below it.

What Precipitation Types Can Radar Identify?

The operational WSR-88D HCA assigns one of these classes:

• Biological (BI): birds, insects, bats. Low CC, noisy ZDR.
• Ground Clutter / Anomalous Propagation (GC): very noisy, often low CC.
• Ice Crystals (IC): small crystals, usually high in clouds, low Z.
• Dry Snow (DS): low to moderate Z, low ZDR, high CC, above the melting layer.
• Wet Snow (WS): found in the melting layer, enhanced Z and ZDR, reduced CC.
• Light and/or Moderate Rain (RA).
• Heavy Rain (HR): high Z and KDP.
• Big Drops (BD): high ZDR relative to Z, often at storm edges or in updrafts, where size sorting leaves few but large drops.
• **Graupel (GR):** soft ice pellets formed by riming, moderate Z, low ZDR.
• Hail, possibly with rain (HA): high Z, low ZDR, somewhat lowered CC.
• Large Hail and Giant Hail categories were added in later upgrades using a separate hail size discrimination step.
• Unknown (UK): no category fit well enough.
• Range Folded (RF): data could not be assigned to a range.

Where HCA Shines

• Hail detection: identifying hail cores and, with the hail size step, separating large and giant hail.
• Melting layer and winter transitions: showing where wet snow gives way to rain.
• Non-weather filtering: marking birds, insects, clutter and AP so they are not mistaken for rain.
• Quality control: feeding better rainfall estimates by knowing where hail or biological targets would bias rain rates.

How Accurate Is Radar Precipitation Type?

The HCA is a powerful tool, but it is not ground truth:

• Beam height: at long range the beam samples high in the cloud, so the class reflects conditions far above the surface.
• Melting layer errors: if the estimated melting level is wrong, the algorithm may call snow rain or rain snow, especially in complex winter storms with warm noses.
• Mixtures: a gate can contain rain, graupel and hail together, but the algorithm must pick one label.
• Beam broadening and partial blockage degrade ZDR and CC, confusing the inputs.
• Noise in weak echoes: dual-pol data are noisy at low signal strength, so light precipitation and distant echoes are harder to classify.
• Freezing rain and sleet are not directly classified, because they occur below the beam; separate winter surface-type products try to address this.
• Tornado debris may be labeled biological, clutter, or unknown. Look at CC and velocity directly for debris.

Reading The Display

Look for coherent patterns rather than single pixels. A solid region of hail in the core of a storm with rain around it is meaningful; a scattered pixel of hail in light rain probably is not. A ring of wet snow around the radar marks the melting layer. Big drops along a storm's inflow edge can hint at a strong updraft.