New Documentation

 I finally got around to consolidating some of my more useful Python code and generated Sphinx documentation.

Please check out the following packages:

• GOES-2-go: 
• download GOES-16 or GOES-17 data from Amazon Web Services (ABI and GLM data).
• RGB recipes for ABI images (Natural Color, Air Mass, Water Vapor, etc.)
• SynopticPy:
• Retrieve mesonet data from the Synoptic API
• HRRR-B:
• Download HRRR data from the HRRR archives at the University of Utah's Pando archive, Google Cloud Platform, and Amazon Web Services.

Hot September

Temperatures in Utah, Colorado, and Wyoming have been hot considering we are in September now. Below shows that the 2-m air temperature in the HRRR analysis for yesterday afternoon is above the 95th percentile temperature by several degrees.

My weather station in Spanish Fork registered a 100 degree temperature on Labor Day during our backyard BBQ. It was hot. Below shows the September 1-3 temperatures since 2013. This has been a warmer start to September than we are used to. Over the next week we will begin to cool down as the upper-level ridge moves further east. Then those fall temperatures will start to creep in.

Cold June Day

A cold front rolled into town last night. It is much colder than usual for this time of year.

Below is the HRRR analysis for 12:00 noon (Mountain time) colored by departure from the 95th percentile. Blue is saying that temperatures are 6-10 degrees colder than the 5th percentile temperature.

My camp trip tonight will feel more like fall camping than summer camping.

Here is a time series of temperature at the University of Utah (WBB) and at Logan airport (KLGU) further north.

High Winds in Northern Utah

Easterly flow caused high winds to northern Utah on 23 May 2019 and with it, many power outages (KSL story).

This picture is from my grandpa's backyard in North Logan. Winds knocked down many trees and fences in the neighborhood...

At 0700 UTC (1:00 AM local time), winds gusts exceeded 60 MPH in Brigham City.

Maximum wind gusts exceeding 40 MPH for the previous 24 hours at 1800 UTC (12:00 PM local time) are shown below...

The HRRR forecasts compared with HRRR 3-yr climatology shows this was an exceptional wind event for this time of year and location. Below shows that the wind speeds in northern Utah exceeded the 95th percentile wind speed by over 10 m/s.

Area Forecast Discussion 
  National Weather Service Salt Lake City UT 
  413 AM MDT Thu May 23 2019 
   
  .SYNOPSIS... 
  The unsettled weather pattern with a trough over the western  
  United States will continue through Friday. There will be  
  somewhat of a hiatus Saturday into Sunday as the trough redevelops 
  a little farther west allowing a drier southwest flow aloft to 
  prevail. This new trough will move across Utah early next week. 
   
  && 
   
  .SHORT TERM (Through 00Z Monday)...Strong easterly winds have been 
  blowing overnight across the northern Wasatch Front and Cache 
  Valley. These winds are expected to remain strong through at least 
  early morning as the 700mb winds remain above 35 kts from about  
  Brigham City northward as well as across the Cache Valley. Farther 
  south the winds have not been as strong as the 700mb winds near  
  the Ogden area were forecast to only be 30 kts earlier this  
  evening and are forecast to be down to under 20 kts by mid  
  morning. May be able to drop the High Wind Warning highlight  
  sooner than later, especially for Davis and Weber counties. The  
  strength of these easterly winds downsloping along the northern  
  Wasatch Front and Cache Valley has prevented showers from  
  occurring overnight, but as these winds relax and the instability  
  of the cold pocket aloft moves in from the south later this  
  morning through this afternoon showers will become widespread.

Area Forecast Discussion 
  National Weather Service Salt Lake City UT 
  1033 AM MDT Thu May 23 2019 
   
  .SYNOPSIS... 
  The upper level low pressure over northwest Utah will maintain  
  unsettled weather across much of the region through early Friday. 
  The low pressure will exit the area and bring a break in  
  widespread precipitation Friday night through the weekend. The  
  next storm system will move across Utah early next week. 
   
  && 
   
  .DISCUSSION...The upper trough centered over the Great Basin is 
  showing multiple circulation centers this morning. These 
  circulations will consolidate into one center over northwest Utah 
  this afternoon, then lift northeast toward the northern Plains   
  later tonight through Friday. A residual upper trough will remain 
  behind the exiting upper low, with this feature merging with the 
  next upper low moving into the Pacific Northwest early in the 
  weekend.  
   
  The canyon/downslope high wind event for areas north of SLC is  
  winding down late this morning. Decreasing easterly near 700 mb  
  winds and the retreat of the tight northeast surface gradient back 
  into western/central Wyoming signals the end of the strongest  
  winds, though gusty easterly winds will continue through around  
  mid-afternoon. Will keep the current High Wind Warning going  
  through the scheduled expiration at 19Z.

Summer Thunder

I woke up last night at 11:40 to strobing lights. A long line of thunderstorms stretched from Richfield to the Utah-Idaho border. This morning I measured 0.26 inches of rain in south Provo.

Radar 11:50 PM, Local Time

It was a forecast miss for the HRRR model, which didn't simulate this particular line of thunderstorms until the analysis hour. This is an important lesson on not using weather models as deterministic forecasts for thunderstorm prediction. This loop of HRRR forecasts shows all 18 forecasts and the analysis valid at 0600 UTC (local midnight). Beyond the first two hours, there is no warning of this particular line of storms.

Started a small fire on Y-Mountain: https://www.heraldextra.com/news/local/central/provo/lightning-sparks-small-brush-fire-on-y-mountain/article_ad4d2531-58a6-5b62-800e-0e38e3be22ca.html

Gusty winds were observed across the state. The Salt Lake International Airport and many other locations along the Wasatch Front recorded winds exceeding 50 MPH between 11 and Midnight last night.

In Spanish Fork, recorded winds gusted ranged between 17 and 30 MPH.

UKBKB (bench): Max gust 17 MPH at 11:35 PM

C9348 (bench): Max gust 18 MPH at 11:33 PM

UCC13 (river bottoms): Max gust 30 MPH at 11:15 PM

KSLC (Salt Lake City International Airport): Max gust 50.6 MPH at 11:20 PM

Spanish Fork HRRR terrain

If you live at the mouth of a canyon you know that it can get windy. One challenge in numerical weather prediction is the inability to resolve small scale features, such as terrain (well, we could run models at ultr-fine scale, but we don't have the computational capacity to do that). The HRRR model has a 3-km grid spacing, which is great, but still too coarse to simulate the actual depth of Spanish Fork Canyon. See the examples below:

Model terrain in the HRRR at Spanish Fork. You are looking south east towards Spansih Fork Canyon

Below is the actual terrain, from a 30 m Digital Elevation Model, is shown below. Again you are looking towards the southeast towards Spanish Fork Canyon.

Finally, a Google Earth image showing the area we are looking at...

HRRR "Hovmoller" Forecasts

Hovmoller diagrams have been used to show wave propagation over space and time. By changing the dimensions of these diagrams to a valid-time/forecast-time dimension, then these kind of plots can be useful to show an ensemble of model forecast results at a point.

For example, here is a "Hovmoller" diagram showing the HRRR forecasted temperature at the Salt Lake International Airport for two days. The observed air temperature is shown at the bottom and look almost like a bar code.

From this figure we can see how air temperature changes across the time period. Reading the figure from left to right and looking at all forecast hours (vertical columns), the temperature is generally cooler at night and warmer in the day with July 4th being forecasted as a hotter day than July 3rd.

A single model run can be read diagonally. For example, pick an analysis hour, then look at the 1-hour forecast by moving one box up and one box right. You can see that some model runs were cooler than other.

Next, we can look at how each model run differed from each other by reading a column from top to bottom, and see how the HRRR forecasts changed between successive model runs. This is useful for determining the likelihood for certain atmospheric conditions.

Now, in real-time operations you don't have all the HRRR data for the rest of the day. The Hovmoller plots instead look like this, with missing data because those HRRR models haven't run yet.

This figure shows forecasted wind speed at Antelope Island (station ID UFD09) for a 24 hour period on July 5th. The contours indicate stronger higher wind speeds forecasted within a 54x54 km box centered at UFD09. Again, the white is the missing information because those HRRR simulations have not run yet. 

You'll see that the latest HRRR run is showing strong winds for a period of time that were not previously forecasted. This may likely be a result of new data assimilated into the HRRR model leading to a greater chance of strong winds. In this case, this particular model run formed a thunderstorm in the area, causing a downdraft and stronger winds in the vicinity as shown in the Reflectivity Hovmoller...

No composite reflectivity was forecasted in any other model run except for the most recent two. There was higher reflectivity in the vicinity as shown by the contours. For this case, there was little warning of a possible storm activity at Antelope Island (at the time of my writing, there has not been a convective system develop over Antelope Island).

A Hovmoller for Spanish Fork (station ID UKBKB) the HRRR forecasts show a greater potential for storm activity while more successive runs are indicating some convective activity for the rest of the afternoon and evening.

There has been some storm activity near Spanish Fork at this time, but mostly in the surrounding mountain area.

Another example of using the HRRR Hovmoller forecast fires to determine the potential for storm activity is from the Burro Fire...

A storm never developed directly over the fire at 21:00 UTC, but there was some other convective activity in the surrounding area.

What we learn: Forecasting for convective outflows in the HRRR model is difficult, especially when you want to pin-point when and where convection will occur. Convective outflows are hard to forecast even an hour or two before they occur, because the time scale of these clouds form are short--less than an hour. A probabilistic approach is most useful for forecasting these events. The ensemble forecast approach should consider an offset  in storm location and timing for such events. In the above Hovmoller for reflectivity for the Burro fire, I would know that convective activity is possible, more so in the vicinity, but not certain. As a fire manager, I would be need to be more aware of convective situations that could potential make firefighting more difficult.

Strong inversion and persistent cold air pool: January 2017

I'm looking at the inversion strength of the cold air pool in northern Utah between January 25 and February 5, 2017. This occurred during the Utah Winter Fine Particle Study (more about that study here: https://www.esrl.noaa.gov/csd/groups/csd7/measurements/2017uwfps/).

Inversions occur when the upper level air is warmer than the air below it. This makes it difficult for air to mix vertically and dilute pollutants from the surface. One way to determine the strength of an inversion is to compare the potential temperature at 700 mb (approximately the height of the mountain peaks) and the potential temperature at the surface. In the top figure in the graph below, I show potential temperature from the HRRR analyses at 700 mb (red) and the surface (blue). The 12-hr HRRR forecast is also shown in yellow. The observed potential temperature from the Salt Lake City radiosonde are plotted as black dots. The bottom figure is simply the difference between the upper level and surface level potential temperature, labeled "Surface Temperature Deficit."

For the most part, the HRRR analyses correspond well with the balloon observations, except the HRRR doesn't mix out as early as was observed on February 4th.

Below shows the potential temperature at the surface and 700 mb, as well as the differences from the HRRR model analyses for the course of the event. The right panel shows the difference between surface and 700 hPa potential temperature. Areas that turn dark red is where the inversion is strongest.

HRRR versus HRRR-Experimental: Vegetation Type and Water Area

Changes in land use between HRRRv2 (current operational version) and HRRRv3 (Experimental)

 There is no change in the elevation, but some lakes have been added. The Great Salt Lake has been modified in HRRRv3

HRRR: Top Soil and Surface Temperature

I'm looking at the surface temperature of the Great Salt Look and wondered if the top soil temperature or the surface temperature is what I should be looking at. Turns out, the top soil level and the lowest temperature level (surface temperature) are identical over water

Shown, difference between Soil Temperature (top soil level) and Surface Temperature:

As you can see, there is no difference between the surface temperature and top soil level temperature over the Great Salt Lake (thus, the white color over the lake area).

The same can not be said over mountainous areas, where there are the largest differences.

Operational RAP/HRRR update for wind gusts

As a note to myself:

AUTHOR:   Stan Benjamin
  DATE:     3-November-16 1459Z

A change was made in the wind gust diagnostic calculation for the operational
RAP and HRRR yesterday (2 Nov 2016) starting with the 14z run. This change
reverted to use of a theta-v profile PBL depth for the wind gust diagnostic
instead of using an alternative hybrid PBL-depth. Use of the hybrid PBL was
resulting in too-strong wind gust fields at nighttime. More information is
available in http://ruc.noaa.gov/rr/RAP_var_diagnosis.html#gust

HRRR Land Use

For reference, here is the land use in the HRRR model:

Categories are based on the MODIS 21 Category data set as indicated below:
1 Evergreen Needleleaf Forest
2 Evergreen Broadleaf Forest
3 Deciduous Needleleaf Forest
4 Deciduous Broadleaf Forest
5 Mixed Forests
6 Closed Shrublands
7 Open Shrublands
8 Woody Savannas
9 Savannas
10 Grasslands
11 Permanent Wetlands
12 Croplands
13 Urban and Built-Up
14 Cropland/Natural Vegetation Mosaic
15 Snow and Ice
16 Barren or Sparsely Vegetated
17 Water
18 Wooded Tundra
19 Mixed Tundra
20 Barren Tundra
21 Lake

Can find other MODIS landuse products here: http://webmap.ornl.gov/ogcdown/dataset.jsp?ds_id=10004&startPos=0&maxRecords=100&orderBy=category_name&bAscend=false

RAPv3 and HRRRv2 Release Notes

Upgrade to the RAP and HRRR model

http://www.nws.noaa.gov/om/notification/tin16-26rap_hrrrraaa.htm

The above link explains that the RAP and HRRR weather models used operationally by the National Weather Service were upgraded to RAPv3 and HRRRv2.

Effective on or about Tuesday, August 23, 2016, beginning with

the 1200 Coordinated Universal Time (UTC) run, the National

Centers for Environmental Prediction (NCEP) will implement

Version 3 of the Rapid Refresh (RAP) and Version 2 of the High-

Resolution Rapid Refresh (HRRR) systems.

Major Changes:

A major change to the RAP will be an expanded computational

domain which will now include Hawaii. This expansion will

facilitate future NCEP plans for ensemble systems and, in time,

improve the initialization of Short Range Ensemble Forecast

(SREF) members that use the RAP for initial conditions.

Analysis Changes:

Both the RAP and HRRR will use an updated version of the

Gridpoint Statistical Interpolation (GSI) analysis code.

Refinements are made to the GSI to improve the assimilation of

surface observations, soil moisture adjustment, and three-

dimensional cloud and precipitation hydrometeors. In addition,

the HRRR will start using the ensemble/hybrid data assimilation;

it is already used in the RAP, but the weighting of the ensemble-

based component in the RAP will increase from 0.50 to 0.75. In

addition, while the RAP already cycles land-surface states, this

cycling is being introduced into the HRRR. In HRRR Version 1, all

runs are independent.

Other analysis changes include:

-Assimilating radial wind and mesonet data

-Applying PBL-based pseudo-innovations for 2-meter temperatures

(already used for 2-meter dew points)

-Changing the cloud-hydrometeor assimilation to avoid METAR-based

cloud building when satellite data shows clear skies at all times

of day (currently used just in daytime)

-Introducing direct use of 2-meter temperature and dew point

model diagnostics in the GSI.

Specific to the HRRR, the application of radar reflectivity data

in the GSI to direct specification of 3-dimensional hydrometeors

is increased to apply to a broader range of weather conditions,

including warm-season events with reflectivity up to 28 dBZ.

Changes to Model:

- The RAP and HRRR will both begin using WRF version 3.6.1; both

will continue to use the ARW core.

- The MYNN planetary boundary layer scheme is being updated to

include the effects of subgrid-scale clouds. The mixing length

formulation in the boundary layer scheme and thermal roughness in

the surface layer are being changed.

- The 9-level RUC land-surface model is being updated to add a

mosaic approach for fractional snow cover, improve the fluxes

from snow cover, and modify the wilting point for cropland use.

- Major updates are being made to the Thompson microphysics

scheme, including making it aerosol-aware with use of an ice-

friendly and water-friendly aerosol field.

- Shortwave and longwave radiation have been changed to use the

RRTMG (RRTM global) scheme that includes the effects of aerosols

and boundary layer subgrid-scale clouds.

- The WRF-ARW diagnostics for 2-meter temperature and dew point

are being improved. 

- The convective scheme in the RAP is changed from the Grell 3-D

scheme to the scale-aware Grell-Freitas scheme. The HRRR, at 3 km

horizontal resolution, explicitly resolves convection and does

not use a convective scheme.

Many of these changes to the data assimilation, land-surface

model, boundary layer scheme, microphysics, radiation, and (in

the RAP only) convective scheme are designed to mitigate the low-

level warm, dry bias in the RAP and HRRR, most notable during

afternoons in the warm season. Significant reduction of these

biases has been evident in extensive testing.

Output Changes:

- The HRRR directory structure will be migrated out of

"nonoperational" on the NCEP ftp and http servers. Data will also

be available on both the primary and secondary servers.

OLD:

http://nomads.ncep.noaa.gov/pub/data/nccf/nonoperational/com/hrrr/

ftp://ftp.ncep.noaa.gov/pub/data/nccf/nonoperational/com/hrrr/

http://www.ftp.ncep.noaa.gov/data/nccf/nonoperational/com/hrrr/

Rain and Hail in Spanish Fork

It's raining and hailing in Spanish Fork...

Time: 18:50 UTC

You can see the cooler air accompanied by the storm from the temperature and dew point time series...Temperature goes down, and wet bulb (humidity) goes up. When the temperature and dew point temperature are the same that is about all the cooling that can be done by evaporation...a natural air conditioner.

There is some good cloud development shown up in the satellite images...

Time: 18:56 UTC
weather.cod.edu

One interesting aspect of this storm is how the HRRR model handles it. The latest analysis hour has the reflective in the right spot, but the HRRR model gets rid of the storm by the first forecast hour. The storm has lasted into that hour, and so this is a case that the HRRR model is not useful for a short term forecast of this storm.

Time: 17:00 UTC

 Now look at the one hour forecast...the hail storm has disappeared. Yep, not really a useful short-term forecast.

Time: One hour forecast valid at 18:00 UTC

HRRR under-predicting temperature the last few days

HRRR Temperature Verification at KSLC and UKBKB between July 31st, and August 3rd, 2016.

The HRRR model is running an average of 3.4 degrees too cold over the last few days. Top shows the temperature observed at the SLC airport (red) and the HRRR analysis. Next is the HRRR temperature bias (HRRR-Observed)

On the other hand, the HRRR analysis for Spanish Fork (UKBKB) is much better, and is on average 1.1 degrees too warm...

 Bias:

HRRR Model Verification for Spanish Fork, Utah

I've grabbed the HRRR analysis forecasts, plucked out the 2-meter temperature and wind values at the location of my weather station (UKBKB), and compared that with the observations from February 2016 to July 27 2016.
(black line is the HRRR analysis, and color line is the observed value)

The HRRR's Temperature analysis does pretty well, except for periods when we are in a cold pool in mid-February.

The wind analyses are all over the place and nearly always over estimated wind speed. This could be a siting issue. My station sits in a residential area between three houses where is could be sheltered from high winds. Also, I don't think models can quite nail down the winds yet. Definitely an area of research applicable to wind farms like the one in Spanish Fork Canyon.

Would be nice to see how the HRRR forecasts verify, but I just started saving forecasts fields starting yesterday. Maybe over the next few months we'll be able to see how forecasts do.

HRRR Geography Data Issues

The High Resolution Rapid Refresh model has a few issues around the Great Salt Lake. One we have noticed is that the modeled lake is too large! This causes increased dew point temperatures around the lake's location than are actually observed. Increased dew point temperatures lead to increased CAPE values, or the potential for thunderstorm development.

On August 10, 2015 at 23:00z the HRRR modeled dew point temperature was at least six degrees higher at the surface than observed during the 00z sounding. This produced nearly 2000 Joules of CAPE in the HRRR when there was actually about 200 Joules of CAPE. This means that the HRRR was saying that thunderstorm development was very favorable when in reality there was marginal potential energy available to cause deep convection. This issue will be looked at further. We intend to modify the lake area to produce more accurate simulations around the Great Salt Lake.

Skew-T diagram of HRRR analysis at 23z on August 8th (thick red) and actual weather balloon observation (thin blue). Surface specific humidity values were less than 7 g/kg, but the HRRR model thinks there are over 10 g/kg. Imagine lifting a parcel with dew point temperatures of 5C rather than 13C, you would get much less CAPE. 

The outer line representing the Great Salt Lake is the lake/land mask used by the HRRR. The inner line is a more accurate representation of the's current size. The coloring represents the specific humidity. Specific humidity is much higher following closer to the large lake boarder. The white dot represents the location of the KSLC sounding site where weather balloons are launched. You can see that this location is in the pink (> 12 g/kg) when in reality the humidity at that location is much less.