Forecasting
Why Predicting Storms in the Mountains Is Difficult
One March 2021 blizzard, followed from Boulder over the Continental Divide, shows what makes mountain snow hard to forecast.
On Friday, March 12, 2021, the National Weather Service in Boulder told Nederland to expect two to four feet of snow. Nederland is a mountain town at 8,240 feet, up Boulder Canyon from the city. Boulder, 3,000 feet lower at the edge of the plains, was told 12 to 24 inches. Fraser, 16 miles west of Nederland on the far side of the Continental Divide, was told 5 to 12.
By Monday morning, Nederland’s volunteer observers had measured between 21 and 38 inches of snow. Boulder’s measured 14 to 21. Fraser got 7. Denver’s official total of 27.1 inches made it the city’s fourth-largest snowstorm since 1882 (NWS Boulder).
That was a good forecast. The snow landed inside, or just below, every range. But the ranges were wide: two to four feet is a factor of two, for towns less than 30 miles apart. This note follows the storm along one line on the map, from Boulder up to Nederland, over the Divide and down to Fraser, and uses what was measured along that line to show why the ranges had to be that wide. Five things decided how much snow each place got. Each is hard to predict anywhere, and harder in the mountains: which way the wind blew, how the models saw the terrain, where rain turned to snow, how much snow each inch of water made, and what the instruments could see.
Sat, 4:20 a.m. MST: 24–48 inches. “Heavy snow. Total snow accumulations of 2 to 4 feet possible with the heaviest totals in the northern Front Range Foothills. Locations west of the Eisenhower Tunnel will receive 8 to 16 inches. Winds could gust as high as 35 mph.”
The wind chose which side of the mountains got the snow
Most of Colorado’s winter snow comes from the west. Storms off the Pacific lose much of their moisture on the mountaintops and west-facing slopes (Colorado Climate Center).
This storm came the other way. Its center crossed the Four Corners and tracked east across southern Colorado. The air circling it blew toward the mountains from the east, 25 to 30 knots of nearly due-east wind carrying moisture from the Gulf of Mexico (NWS Boulder). When that air reached the Front Range, it had to climb. Rising air expands and cools, cooler air holds less water vapor, and the vapor condenses into cloud and falls as snow. This is upslope flow, and it is behind many of the Front Range’s biggest snowstorms.
Nederland’s observers measured 21 to 38 inches, Boulder’s 14 to 21. Fraser, 16 miles west over the Divide, got 7.
Switch the chart to water, and the storm’s shape is plain. Boulder and Nederland caught about the same amount, roughly 2 inches. Snow pillows near the Divide, at 9,700 to 10,300 feet, gained 1.7 to 2.1 inches. Then the storm ran out. Air crossing the Divide had already given up much of its water, and on the far side it sank and warmed, which dries air further. Tabernash, in the valley past Fraser, measured less than half an inch. A typical winter storm does the opposite: it soaks the west side and leaves the east slopes dry.
The exception is right at the crest. An observer near Winter Park, a few miles down the west slope, measured 17.5 inches. Snowflakes fall slowly, around a meter per second, so a strong wind can carry them miles past the ridge that lifted them (Houze, 2012).
All of this hinged on the storm’s track. On Friday afternoon, forecasters wrote that if the storm moved farther north, a dry slot would push across the area, bringing “much less snowfall.” They judged that unlikely, and it didn’t happen. But feet versus inches came down to where the center of the storm went, and which slopes that put in front of the wind.
The model’s mountains are smoother than the real ones
Forecasters rely on computer models to work out where the air will go. In 2021, the most detailed one the U.S. ran every hour was the High-Resolution Rapid Refresh, or HRRR. It divides the country into squares 3 kilometers (1.9 miles) on a side (NOAA). Each square has a single ground height, and the model lifts air over that simplified ground.
Along our line, the real Divide rises to 11,950 feet. The model’s crest is 11,560. Between Boulder and Nederland, the model smooths over Boulder Canyon: in places its ground is more than 900 feet above the real ground. Finer grids help. For the 2002 Winter Olympics in Utah, a model with 4-kilometer squares forecast precipitation more accurately than the same model at 12 kilometers (Hart, Steenburgh and Onton, 2005). But a 3-kilometer square is still wider than a canyon.
Along this line the Divide rises to 11,950 ft; the model’s crest is 11,560 ft. Between Boulder and Nederland, where it smooths over Boulder Canyon, the model’s ground is up to 940 ft above the real ground.
Switch the chart to precipitation to see the HRRR’s 48-hour forecast from 5 a.m. Saturday, as the storm arrived. It put the most water in the foothills, where it fell, but it forecast more than was measured everywhere:
- Nederland: 2.7 inches forecast, against a measured median of 1.9.
- Boulder: 2.4 forecast, 2.0 measured.
- West of the Divide: about 1 inch forecast, twice what Tabernash measured.
The model carried more water over the Divide than crossed it. On Sunday morning, forecasters in Boulder noted that the HRRR seemed to be “overdoing precipitation a bit.” Some of the gap is on the measuring side, because gauges miss snow in wind (more on that below). But the pattern was the model’s.
Rain in Boulder, snow in Nederland
The storm reached Boulder as rain. From 7 to 11 a.m. on Saturday, with the temperature at 1 to 2°C (34 to 35°F), the airport reported rain and drizzle. At noon the temperature slipped to freezing and the rain turned to snow. The snow pillows above Nederland, around 10,000 feet, read −4 to −7°C all morning.
The weather balloon launched from Denver at 5 a.m. shows why. The level that mattered was about 6,200 feet, where the wet-bulb temperature crossed 0°C. That was 900 feet above Boulder and 2,000 feet below Nederland.
At Sat, 5 a.m. MST the wet-bulb temperature crossed 0°C at 6,230 ft: above Boulder, below Nederland.
A falling snowflake doesn’t respond to air temperature alone. As it starts to melt, some of its water evaporates, and evaporation cools it. The wet-bulb temperature, what a thermometer reads with a wet cloth around it, captures that combined effect. In dry air it can sit several degrees below the air temperature, and snow survives well above freezing. Across 17.8 million observations from Northern Hemisphere weather stations, the temperature at which rain and snow were equally common ranged from 0.7°C in the most humid air to 4.5°C in the driest. Some of the highest values were in the Rocky Mountains, where the air is usually dry (Jennings et al., 2018).
This storm wasn’t dry. Its air came from the Gulf of Mexico, and on Saturday morning Boulder’s wet-bulb temperature was within 0.2°C of the air temperature. The switch to snow came almost exactly at freezing.
That leaves very little room for error. In the Denver sounding, temperature fell about 7°C for every kilometer of height. A forecast just 1°C too warm puts the rain–snow line about 150 meters, or 500 feet, too high. That is enough to move it past a town. Farther out on the plains, forecasters called the rain–snow line “a big question mark,” and along Interstate 25 they weren’t sure whether the snow would stick to warm roads.
The same water made very different snow
Snowfall depends on two things: how much water falls, and how much space each inch of it takes up as snow. The second is the snow-to-liquid ratio. The old rule of thumb is 10 to 1, ten inches of snow for every inch of water, but the ratio varies a great deal, even within a single storm (NWS).
CoCoRaHS volunteers measure fresh snow, then melt it to measure its water. On the plains, 70 observers’ snow came out at a median of 8.7 inches per inch of water. In the foothills above 7,700 feet, 19 observers’ came out at 14.6. With about 2 inches of water each, Boulder got about 15 inches of snow and Nederland about 30. The 10-to-1 rule would have given about 20 in Boulder and 19 in Nederland: five inches too many in one town and eleven too few in the other.
70 plains observers: median 8.7 to 1. 19 foothills observers: median 14.6 to 1.
The difference comes from temperature. Boulder’s snow fell through air near 0°C. Flakes close to melting are wet: they clump together and pack down under their own weight as they pile up. Nederland’s fell through colder air and stayed light. Forecasters expected this. On Saturday morning they wrote that ratios near Denver might be only 9 or 10 to 1, and higher in the colder foothills.
The radar looked over the storm
Radar is how forecasters watch a storm as it happens, and how they check it against their forecast. The radar that covers Boulder sits on the plains east of Denver. Its beam travels in a straight line while the Earth curves away beneath it, so the farther it goes, the higher above the ground it is.
Aim it toward Nederland and on over the Divide, and the foothills start clipping the bottom of its lowest beam, tilted half a degree up, 12 miles before town. That beam passes about 1,200 feet over Nederland and runs into the mountains 3 miles later. The next beam up, at 0.9°, passes 3,100 feet over Nederland and only just clears the Divide, which is 12,600 feet high on that bearing, losing its lower half to the mountains. By the time it reaches the Fraser valley it is nearly 5,000 feet up. Snow falling from cloud below that is invisible to it.
The national MRMS mosaic combines every radar in range and turns what they see into a precipitation estimate. Along our line, that radar-only estimate came up short everywhere:
- Boulder: 0.8 inches, where gauges caught 2.0.
- Nederland: 0.9, where gauges caught 1.9.
- West of the Divide: nothing at all. Observers in Tabernash measured half an inch of water and 7 inches of snow.
Turning radar echoes from snow into water is hard anywhere. Here, behind the mountains, the radar missed the snow entirely.
The 0.5° beam is 1,170 ft above Nederland, with terrain already clipping its lower edge, and is blocked before Tabernash.
Gauges have their own blind spot. Wind blows snowflakes around a gauge’s opening instead of into it. In an international experiment at eight sites, gauges with a single wind shield caught 24% less snow than reference instruments, and unshielded gauges caught 34% less (Kochendorfer et al., 2017). SNOTEL sites measure the same snow twice, with a gauge and a snow pillow side by side. During this storm the two disagreed by as much as 0.4 inches, and the gauge read lower at six of ten sites. The measurements used to judge a mountain forecast carry errors of tens of percent, most of all in wind and on ridges.
Why the ranges were wide
Taken together, these are the reasons the forecast for Nederland was two to four feet rather than three:
- The water: how much fell depended on the wind’s direction and on models lifting air over mountains they can only draw roughly.
- The phase: whether that water fell as rain or snow depended on a degree or two.
- The depth: how deep the snow got depended on a ratio that nearly doubled between Boulder and Nederland.
- The check: the instruments for watching the storm missed part of it.
A mountain forecast has to carry all of that. Nederland’s forecast on March 12, 2021 was two to four feet, and about 30 inches fell. The range was the honest part of that forecast.
Sources and data
Every chart is drawn from measurements and archived forecasts. Download the data behind them (JSON).
- The storm: NWS Boulder, March 13–14, 2021 Northeast Colorado Blizzard. Winter storm watches, warnings and forecast discussions from the Iowa Environmental Mesonet text archive. Quotations are from NWS Boulder forecast discussions issued 12–14 March 2021.
- Snow and water on the ground: CoCoRaHS daily reports; NRCS SNOTEL hourly data; NWS local storm reports and ASOS/AWOS observations from the Iowa Environmental Mesonet. Storm totals run from 7 a.m. MST 13 March to 7 a.m. MDT 16 March.
- The air above: Denver radiosonde profiles from the IEM RAOB archive. Wet-bulb temperatures computed with Stull (2011).
- The model: HRRR run of 12 UTC 13 March 2021, from NOAA’s archive on AWS.
- Radar: MRMS radar-only 72-hour precipitation, from the IEM MRMS archive. Beam heights computed for the Denver radar (KFTG) with standard 4/3-earth refraction.
- Terrain: Copernicus GLO-90 heights, via the Open-Meteo elevation API.
- Science: Colorado Climate Center; Houze (2012); Hart, Steenburgh and Onton (2005); Jennings et al. (2018); NWS on snow ratios; Kochendorfer et al. (2017).
The header is an AI-generated illustration, not a photograph of this storm.