Thunderstorms can develop suddenly, transforming calm skies into scenes of heavy rain, fierce winds and dangerous lightning. And while the United States is no stranger to this weather phenomena, predicting when and where a thunderstorm will strike remains one of the most difficult challenges in meteorology, despite major advances in forecasting technology.

Unlike larger weather systems, such as hurricanes or winter storms, which can be tracked over hundreds of miles and several days, thunderstorms tend to be hyperlocal and short-lived. Bill Bunting, deputy director at the federal Storm Prediction Center, said their unpredictability stems from their complexity and size.

Forecasters often know up to a week in advance if conditions are likely to be favorable for thunderstorms, but their ability to pinpoint exactly where and when those thunderstorms kick off is much more limited. In Texas, they knew July 4 could feature severe thunderstorms, but it wasn’t until just a few hours before heavy rain began to inundate the area around the Guadalupe River that their exact effects became clear.

“Thunderstorms are among the smallest and shortest-lived forms of hazardous weather in the U.S.,” Bunting said. “They’re typically 5 to 15 miles in diameter and last 30 minutes to a few hours.”

Thunderstorms can occur nationwide, and at any time, “day or night, throughout the entire year,” the National Weather Service says, but they are most common in the late afternoon and evening during the warm months.

To understand the difficulty of forecasting thunderstorms, it helps to understand the different types, as well as how they form.

There are three key ingredients needed for the development of a thunderstorm:

Atmospheric moisture, which becomes the fuel for the storm to develop and intensify.

An unstable atmosphere, which allows the storm cloud to rise high enough to produce lightning and precipitation.

A source of “lift” to force the air upward, which starts the process of thunderstorm development. Sources of lift can include weather systems like cold or warm fronts, or geographical features.

Once those conditions align, thunderstorms can rapidly intensify, especially if warm, moist air continues to feed upward motion in the atmosphere.

There are also three main types of thunderstorms:

Single-cell storms contain one updraft and one downdraft — currents of air that rise and fall within the storm. The downdrafts are typically associated with rain and, at times, hail. These storms usually last around 30 to 45 minutes.

Multicell storms are clusters or groups of thunderstorms. They can increase the risk of strong wind gusts, large hail and heavy rainfall, and can persist for a few hours.

Supercell storms consist of an updraft that rotates. These occur within a narrower range of atmospheric conditions that are conducive to the formation of large hail and occasionally tornadoes and very heavy rainfall.

Sometimes, multicell or supercell storms develop and repeatedly move over the same area, much like a train’s movement confined to its track. The phenomenon is known as “training thunderstorms,” which was observed during the catastrophic floods in Kerr County, Texas, last weekend.

Today’s meteorologists use a sophisticated suite of tools to monitor for thunderstorms: Doppler radar, satellite imagery, surface and upper-air observations and computer simulations known as numerical weather models. But no single tool tells the full story.

“Forecasters combine these bits of information and use their own experiences and judgment in order to produce accurate thunderstorm forecasts,” Bunting said.

Still, these methods are a far cry from the 1950s, when forecasters relied mostly on hand-plotted weather maps and used “empirical rules of thumb.” Back then, thunderstorm forecasts were made by manually extending current conditions forward in time.

“Thunderstorm forecasting has benefited greatly from improvements in numerical weather prediction that began in the 1960s and ’70s,” Bunting said. “As these models have become more accurate and capable of resolving smaller details in the atmosphere, meteorologists have used this information to issue more accurate forecasts.”

But even with modern computers, forecasting a thunderstorm is often based on probability. Most forecasts beyond a few days highlight general risk areas, rather than pinpointing exact timing or locations. This is how last week unfolded in Texas.

“Thunderstorm forecasts, in a general sense, can be made as far as seven to 10 days in advance,” Bunting said. “However, forecasts beyond a few days typically don’t contain detailed information on intensity or the timing of storms at a particular location.”