It’s a highly vulnerable spot, an urbanized strip along the bottom of a deep valley through which the Patapsco River flows. This place, historic Ellicott City, Md., has seen plenty of serious floods: 1868, 1923, 1952. More recently, the remnants of Hurricane Agnes (1972) left an extreme high-water mark, measured in many feet. The Great Mid-Atlantic Flood of June 2006, once again drowned parts of the town.

And now, last Saturday, what the National Weather Service is calling an “off-the-charts” thousand-year rainfall event (in terms of recurrence interval) created a harrowing drama involving 120 swift-water rescues, dozens of ruined cars, demolished businesses and two fatalities.

The meteorology of Saturday evening’s event was complex, involving multiple scales upon which moisture and energy became focused, laserlike, over Ellicott City. The result was epic amounts of rain in an unthinkably short amount of time.

Why this was difficult to predict

Like so many flash floods, the exact time and location of this one was difficult to pinpoint. Around 3 p.m., Capital Weather Gang meteorologists warned of “heavy downpours” and that “flash flooding … will be the main concern this evening.”

This is how a flash flood occurs and what you need to do to stay safe. (Claritza Jimenez/The Washington Post)

By 5 p.m., meteorologists at the National Weather Service cautioned that a large area extending from Northern Virginia northeast through Pennsylvania and Connecticut was under the gun for heavy rainfall.

The air mass had destabilized, humidity was approaching historically high levels, and air currents were beginning to converge across this region — in a manner that would literally squeeze moisture out of the atmosphere.

But the best any meteorologist could do was portray the threat region in broad strokes. The reason: Summertime flash floods almost always issue forth from highly localized convective storms, dropping vast amounts of water on small locations. Flash flood warning essentially becomes a “nowcasting” exercise: Once those storms have formed, you try to stay ahead of where they will track and how long they will persist. There is almost always never any lead time.

The meteorology behind Saturday night’s flood

The highly localized, convective nature of Saturday’s flood is underscored in the following image, which depicts radar estimated rainfall in and around Ellicott City.

Compare this five-inch estimate of rainfall, with the analysis from the National Weather Service’s Greg Carbin, which showed up to 8.2 inches of had fallen in just three hours, from 6 to 9 p.m. As is often the case, the radar will indicate significantly less rain than gauges, due to many factors. These include distance from the radar, width of the radar beam, the time it takes to scan a storm repeatedly, and the theoretical framework describing how scattered microwave energy translates into actual drops on the ground.

The convective system that dumped on Ellicott City left behind a swath of heavy rain extending from Parr’s Ridge (Damascus) in upper Montgomery County to the western suburbs of Baltimore (shown below).

To get rainfall so extreme, there must be abundant moisture. Indeed this was the case, as the next figure shows. This is a map depicting total precipitable water, which vertically integrates the total mass of water vapor from top of the troposphere to the surface — expressing the result as an equivalent depth of rainwater. Total precipitable water values were in the 2 to 2.2-inch-range across central Maryland, thanks to southerly flow pumping low-level, humid air off the Atlantic.

Total precipitable water tells only part of the story. Values this large raise the prospect of flash flooding, but much more water vapor can be made available to a convective storm when the airflow converges.

When air streams converge, humid air is fed into a storm complex from a large, surrounding area. This explains why the rain totals can exceed total precipitable water values by a factor of two to three, or more.

The figure below is an analysis at 5 p.m. that shows how air streams (heavy red, dashed arrows) at the 5,000-foot level were converging over Maryland. The air flow converged thanks to an area of low pressure over western Pennsylvania and high pressure off Long Island.

But there’s more to this figure that’s important.

First, note the thin red lines, which show contours of convective available potential energy (CAPE) — a measure of the buoyant energy feeding convective updrafts. The greater this energy, the more water vapor is lofted into the clouds and processed as rain. A tongue of significant CAPE (1500 J/kg) was feeding northward into central Maryland, in the region of convergent airflow — sustaining vigorous thunderstorms.

Second, I have annotated the position of a warm front on this diagram (red dashed line, oriented along the Mason-Dixon Line). This frontal zone was pushing north across the region during the day. With slightly cooler air to the north of the boundary, the convergent air flow was also rising along this sloped thermal boundary, from south to north. This helped cool the air to saturation. The combination of converging and up-gliding air focused very intense ascent over central Maryland.

Third, you’ll note the words “back-building convection” within the flash flood threat region (green, scalloped lines). This refers to the tendency for larger convective clusters to remain stationary (or nearly so) for hours. This happens when individual convective cells form repeatedly over the same locations. Winds carry off the older cells while new cells pop up to replace them. The larger complex of cells stays put, leading to steadily accumulating rain.

My review of the radar loop from the event revealed the backbuilding process over Ellicott City, for part of the time — as a giant convective cluster congealed and moved very, very slowly toward the east.

Earlier in the evening, before the main back-building complex took shape, Ellicott City was over-swept by three separate convective cells, moving rapidly from the south. The repeated passage of discrete storm cells is called training. Each of these cells dumped a quick a half-inch to one inch of rain, before the main, back-building cluster congealed.

Prospects for better prediction?

Of all the hydrometeorological hazards in the United States, flash floods are the No. 1 killer. When compared to tornadoes, derechos and hailstorms, it does seem somewhat ironic: Humble rain, often gentle, is life-sustaining, nourishing and thus benign 99 percent of the time. But what we take for granted sometimes quickly turns deadly.

To understand flash flooding, you need to examine the behavior of the smallest convective storm cells: Are they training? Are they developing into a larger, nearly-stationary complex? Computer models, even the highest resolution simulations, cannot yet resolve these details. Until they do, unfortunately, most flash flood warnings will only be issued once the heavy rain is already underway.