HRRR Zarr Variable List

Notes & Errata

Float precision

The HRRR GRIB2 files contain data in 64-bit float format. The HRRR Zarr archive was originally created converting these to 16-bit floats, losing precision but not in a way that was relevant to many use cases for most variables. Over the years, certain variables such as surface pressure were discovered to require a higher precision and started to be generated as 32-bit floats.

Eventually the smoke variables (MASSDEN and COLMD) were updated to 64-bit floats and we decided to change the default from 16-bit to 32-bit for all other variables from the sfc files, leaving prs variables with 16-bits as the default. This change was implemented starting with the model run at 2024-08-19 16z. Then we started a job to redo the archive at this resolution back to June 1, 2024 so that summer 2024 has consistent datatypes.

As a result, most sfc variables will have their datatype change on 2024-06-02 00z. Some variables may have precision changes at other points in the history of the archive. Please check the .zmetadata file (or just have xarray load the dataset and display the datatype) when in doubt.

Smoke Variables

The smoke variables often have misleading official names. The notes section of the table explains what they actually are. Additionally, the near-surface smoke parameter (MASSDEN 8m_above_ground) was reported with units of μg/m3 until 21 December 2021, when it was then corrected to kg/m3 (Source: NWS Service Change Notice).

Missing data due to iris-grib #265

Sometimes you'll see a variable that has zarr array metadata generated for a given date/time, but is missing any actual data. In many cases this is caused by processing erroring out, most commonly due to an issue in an underlying library documented at iris-grib #265

For analysis files, this does not represent a loss of data, since the issue only affects grids where NCEP provided a constant value in the original HRRR GRIB2 files instead of providing a grid of data. So you can assume all the data is that constant value (zero for most but not all affected variables). For example, there may be no hail or snow in the CONUS at a particular time.

Unfortunately, meaningful forecast data is missing in some instances because if one forecast hour is set to a constant, the entire forecast zarr will be missing. As a result, you must go to the GRIB2 files to get the data for the hours where it does exist. Solar radiation is always affected by this bug at night and thus all forecasts for these variables are missing in the Zarr archive.

General Variable Info

The variables in HRRR Zarr are ultimately copied directly from the HRRR GRIB2 files. The metadata and documenetation of those files may help answer questions about the corresponding variables in Zarr.

For more info see NCEP Product Inventory:
https://www.nco.ncep.noaa.gov/pmb/products/hrrr/hrrr.t00z.wrfsfcf00.grib2.shtml
https://www.nco.ncep.noaa.gov/pmb/products/hrrr/hrrr.t00z.wrfsfcf02.grib2.shtml


Table of Variables

Click on a specific column header to sort the table by that specific cloumn

Structure of Bucket