Interpreting GFS products

[WenMeng-NOAA]

GFS user's questions:

We are trying to understand the exact meaning of the various variables included in the GRIB files provided for the Global Forecast System (GFS). We have found the page https://www.nco.ncep.noaa.gov/pmb/products/gfs/ where we can find the complete list of GRIB messages contained in the full data offering of “GFS - Global longitude-latitude grid - 0.25 degree resolution” (https://www.nco.ncep.noaa.gov/pmb/products/gfs/gfs.t00z.pgrb2.0p25.f003.shtml). However, this succinct document still leaves us with some questions. If there is some further documentation on how to interpret this data, I fear we have failed to find it ourselves, and I apologize for potentially wasting your time.

We currently have the following questions regarding the data:

In the documentation page listed above, there are 5 different values for the column ‘Forecast Valid’, but we are unsure how to correctly interpret the data based on these values. As an example, the variable “DSWRF” (Downward Short-Wave Radiation Flux, meaning roughly the amount of sunlight hitting the ground at a certain location) is listed as ‘0-3 hour ave’, which we had taken to mean that over the course of three hours, the value given is the average since the start of that 3-hour interval. After that, the value would be reset, and a new average is accumulated. This would mean that the value at t=3 provides the 3-hour-average since t=0, and the value at t=4 provides 1-hour-average since t=3. But when plotting the data with this interpretation, we found it to make strange jumps. But when we changed our interpretation to an average over a 6-hour interval instead, the result was a smooth curve like one would expect.

[WenMeng-NOAA]

Answers from @HuiyaChuang-NOAA

1. The GFS uses a 6 hour bucket which means most time-averaged and accumulated variables are reset to zeros
   every 6 hours. Thus, F04 is actually accumulated from F00 to F04 while F07 is accumulated from F06 to F07.
   As you already found out, when you plot downward shortwave radiative flux, using a 6 hour bucket makes more sense.
   I will suggest to NCO Dataflow (copied on this email) to add the bucket information for each model on their web page.

2. The time average and instantaneous categorical rain/snow/sleet use similar algorithm except that the former use

time averaged precipitation rate while the latter use instantaneous precipitation rate in determining whether to

even compute categorical precipitation type for a given grid point.

3. We started to provide continuous bucket precipitation a few years ago in addition to 6 hour bucket precipitation.

This continuous bucket never reset to zero so is always accumulated from F00

SInce GFS uses a 6 hour bucket, you will see two identical records until F06. If you check F07, you will see that

one has time label F06-F07 while the other has time label F00-F07.