Why Your Forecast Says Rain While the Radar Is Empty: The Gap Between Model Output and What's Overhead

Have you ever pulled up the radar to check a forecast, found an empty screen, and decided the app was wrong? Most people have, and it is the single most common complaint we hear about weather software.

The reasoning feels airtight. Radar shows something real and physical, the forecast shows a guess, so when they disagree the real thing should win.

The trouble is that radar is not showing you what is falling on your street. It is showing you what a tilted microwave beam intercepted somewhere above your street, several thousand feet up, a few minutes ago.

What Radar Actually Measures

A weather radar does not take a picture of the sky. It emits a pulse of microwave energy and listens for the fraction that bounces back off water droplets, ice, insects, dust, and the occasional flock of birds.

The strength of that return is converted to reflectivity in dBZ, and the software colors it green through red. What you see on your phone is an inference about droplet size and density, not a photograph of rainfall.

That inference happens at a specific altitude. The beam leaves the antenna at a fixed elevation angle — the lowest scan on the U.S. NEXRAD network is typically 0.5 degrees — and it climbs as it travels outward.

Keep in mind that 0.5 degrees is not flat. Over a long distance that shallow angle carries the beam far above the ground you are standing on.

Beam Overshoot: Why Distance From The Radar Changes Everything

Geometry does most of the damage here. At 25 miles from the radar site, the center of the lowest beam sits roughly 2,000 feet above ground; at 60 miles it is near 6,000 feet; at 120 miles it can exceed 15,000 feet.

Earth curvature compounds the climb. The ground falls away beneath a beam that is already angled upward, so the gap between what radar sees and what you feel widens with every mile.

This is why a drizzly coastal morning can show a blank radar while your windshield is beaded. Shallow stratiform rain from a marine layer often tops out below 5,000 feet, entirely underneath the beam if you live far from the nearest site.

The reverse failure also happens. Close to the radar, the beam can pass beneath a shower's core and underreport an intensity that is obvious from the window.

If you want to know whether this applies to you, find your nearest NEXRAD site and estimate your distance from it. Under 40 miles, the lowest scan is a reasonable proxy for what is falling; past 80 miles, treat radar as a story about the middle atmosphere rather than the sidewalk.

Virga: Precipitation That Exists And Never Lands

The opposite complaint is just as common. Radar is lit up with green, and nothing is falling.

That is usually virga — precipitation that leaves the cloud base and evaporates before reaching the surface. The radar is correct that water is in the air; it simply never completed the trip.

Virga is a dew point story. When the layer beneath the cloud is dry, falling droplets evaporate into it, and the deeper and drier that layer is, the higher the rain dies.

The rule of thumb worth memorizing: a surface dew point depression of more than 20 degrees Fahrenheit, paired with a high cloud base, makes virga likely. Our piece on what dew point tells you that humidity doesn't covers why that spread matters more than the relative humidity number your app displays.

Denver residents know this better than anyone. High-plains afternoons routinely produce dramatic gray streamers under a thunderstorm base with a completely dry street below, which is part of why dressing for Denver weather is less about precipitation totals and more about wind and temperature swing.

For photographers, virga is a gift rather than a nuisance. Those fibrous trailing streaks catch low-angle light beautifully, and they are among the most reliable subjects during the golden hour window on a high-based storm day.

The Timing Offset: A Forecast Hour Is Not Now

The third source of disagreement is the simplest and the most frequently missed. Your forecast is not describing this minute.

An hourly forecast row labeled 3 PM describes a window — usually the hour beginning at 3 PM, sometimes centered on it — and the underlying model output was generated earlier. A high-resolution model like the HRRR runs hourly and takes roughly 45 to 60 minutes to compute and distribute.

Add the app's own refresh interval, and the number on your screen at 2:58 PM may reflect an atmosphere the model last saw at 1 PM. That is not negligence; it is the cost of running physics on a supercomputer.

Radar, by contrast, is close to live. A NEXRAD volume scan completes in about four to six minutes depending on the scanning mode, so radar imagery is typically two to ten minutes old rather than two hours.

You are therefore comparing a near-live measurement against a projection made an hour or more ago about a period that has not finished. They can both be right.

Why The Models Disagree With Themselves Too

Even setting radar aside, a rain forecast is a probabilistic statement produced by imperfect physics. Models solve fluid equations on a grid, and the grid spacing determines what they can resolve.

The HRRR runs at roughly 3-kilometer spacing, which is fine enough to represent individual thunderstorm clusters but not individual cells. The GFS at 13 kilometers cannot resolve a summer pop-up shower at all — it parameterizes convection rather than simulating it.

This is why a 30 percent afternoon chance so often produces either a soaking or nothing at all. We unpack that in what a 40 percent chance of rain actually means, and the short version is that the number describes coverage and confidence together, not intensity.

Forecast disagreement is also measurable in advance. When ensemble members scatter widely on timing, the forecast is genuinely uncertain, and our piece on reading ensemble spread explains how to tell a confident forecast from a coin flip before you plan around it.

The Other Radar Artifacts Worth Knowing

Beam geometry produces several other illusions, and recognizing them prevents most misreadings. These are the ones you will encounter most often:

• Ground clutter. Buildings, terrain, and wind turbines return energy near the radar site, producing a persistent speckled ring that does not move with the wind. If the echo has sat in the same place for an hour, it is not weather.
• The bright band. Snow melting into rain produces a layer of wet-coated ice that reflects far more strongly than either pure state. It shows as a ring of enhanced returns at the freezing level and exaggerates apparent intensity.
• Anomalous propagation. A strong temperature inversion bends the beam downward into the ground, painting phantom returns across a clear night. This is most common on calm, cool mornings with a shallow surface inversion.
• Biological scatter. Insect and bird returns produce a low-reflectivity bloom that expands outward from a radar at dusk, especially in summer. Dual-polarization data distinguishes it from rain, but most consumer apps do not show you that field.
• Range folding. Echoes beyond the radar's unambiguous range can be displayed at the wrong distance, appearing as purple or blank wedges on some products.

All of these produce the same user experience: a mismatch between the picture and the window. Each of them has a specific physical cause, and none of them means the forecast is broken.

How To Read The Two Together

The productive move is to stop treating radar and forecast as competing claims and start treating them as different instruments. Radar answers a question about now; the forecast answers a question about later.

Here is the sequence we use when the two disagree:

• Check your distance from the radar site. Beyond roughly 80 miles, an empty low-level scan tells you very little about drizzle or shallow rain.
• Check the dew point spread. A wide depression under a high cloud base means the green on screen is likely virga and will not reach you until the sub-cloud layer moistens.
• Check the forecast hour label. Confirm whether the row you are reading covers the hour ahead or the hour you are in, and note how recently the app refreshed.
• Check the trend, not the frame. A single radar image is nearly useless; a loop of the last hour tells you direction, speed, and whether echoes are building or collapsing.
• Check what you are actually deciding. If the decision is whether to walk the dog in ten minutes, radar wins. If it is whether to pack a shell for a 6 PM dinner, the forecast wins.

That last step is the one that resolves the argument. The instruments are not competing for authority; they are answering questions on different timescales, and the right one depends on which question you have.

Virga Is Not A Failure Mode For A Photographer

Worth a brief detour, because the same conditions that frustrate a commuter reward a camera. Virga under a high base at low sun angle produces fibrous vertical texture with strong backlighting and visible halation at the edges.

The dry sub-cloud layer also keeps atmospheric haze low, which preserves contrast at distance. You get the drama of precipitation without the flat gray light that an actual rain shaft imposes.

The practical setup is a long focal length to compress the streamers against the cloud base, and metering for the sky rather than the foreground. It is one of the few situations where an empty rain gauge and a full radar screen are both good news.

What This Means For How We Write Briefs

We built Vesper around the belief that the numbers are the easy part and the translation is the product. A raw radar tile and a precipitation probability are both real data, and both of them leave the reader to do the interpretation alone.

So our briefs say what the disagreement means rather than displaying both and shrugging. If the model has rain at four and the sub-cloud layer is bone dry, the honest sentence is that the first hour of it will not reach the ground.

If you want the longer version of that argument, how we write a daily brief lays out the editorial pass every entry goes through. And if you want to know how much to trust any given day's call, reading forecast confidence is the companion piece.

Remember that neither instrument was designed to answer your question directly. Radar was built to find severe storms for warning meteorologists, and numerical models were built to simulate the atmosphere, not to tell you whether to carry an umbrella.

The gap between those purposes and your actual decision is where the complaint lives. Closing it is a writing problem more than a data problem, which is the whole premise of what we do here.

Common Questions

Why does my app say it's raining when the radar looks clear?

Most often because the nearest radar's lowest beam is passing above shallow rain. Beyond about 80 miles from a radar site, the beam sits over 8,000 feet up, so drizzle and low stratiform rain fall entirely beneath what the scan can detect.

Can radar see rain that isn't reaching the ground?

Yes, and that is exactly what virga is. Radar detects droplets aloft, and when the air beneath the cloud is dry enough, those droplets evaporate mid-fall. The echo is genuine precipitation, it just never completes the trip to the surface.

How old is the radar image on my phone?

A NEXRAD volume scan takes about four to six minutes to complete, and delivery adds a little more. In practice the frame you are looking at is typically two to ten minutes old, which is far fresher than the forecast row beside it.

Which should I trust when the radar and forecast disagree?

It depends on your timescale. For the next fifteen to thirty minutes, trust a radar loop and its motion vector. For anything beyond about an hour, trust the forecast, since radar cannot tell you about precipitation that has not formed yet.

What are those echoes that never move on the radar?

That is ground clutter — returns from buildings, terrain, or wind turbines near the radar site. It appears as a stationary speckled ring and is a strong tell, because real precipitation echoes drift with the wind while clutter stays put.

Why is a pop-up summer shower so hard to forecast?

Coarse global models at 13-kilometer grid spacing cannot resolve individual convective cells and approximate them instead. Even 3-kilometer models represent storm clusters rather than single cells, so placement errors of ten miles are routine.