I am trying to use the GLDAS 2.1 3-hourly precipitation datasets available at:

The dataset is documented here: https://hydro1.gesdisc.eosdis.nasa.gov/data/GLDAS/README_GLDAS2.pdf

The data is available as NetCDF4 files, one file per time-step, for example:


where 3H means 3-hour time step, A20180208 means the date 8th February 2018, 0600 means the time 6:00 UTC, 021 means the product version. The NetCDF file contains a variable Rainf_f_tavg (Total precipitation rate) in units kg m-2 s-1.

What is not clear to me from the documentation is how to interpret the precipitation rate:

  1. Is this the average precipitation rate between 6:00 UTC and 9:00 UTC?
  2. or is this the average precipitation rate between 3:00 UTC and 6:00 UTC?
  3. or is this the averate precipitation rate between 4:30 UTC and 7:30 UTC?

This information is critical for me if I want to correctly calculate accumulated precipitation for example how much rain+snow in mm has fallen between 7th February 2018 6:00 UTC and 8th February 2018 6:00 UTC.

2 Answers 2


The metadata says the following:

"The GPCP 1-degree Daily (1DD) dataset is used and disaggregated to 3-hourly interval"

From that, and since the timestamps in the filenames run from *0000* to *2100*, I would conclude that this time is the starting point from where to count 3 hours forwards. For your example, this would mean *0600* should probably be the time from *0600* to *0900*.


The answer to your question is that neither 1-3 of your proposed answers is correct, exactly. From the manual, we read:

"file name for 3-hourly 0.25 degree GLDAS-2.1 Noah data at 03:00Z on 1 January 2000 is “GLDAS_NOAH025_3H.A20000101.0300.021.nc4.”

This means that the 3-hourly data are instantaneous, and only formally represent the exact moment in time that is 0300 hours. It is up to you, the user, to decide how to interpolate between times. In that case, 1), 2), and 3) that you propose are valid, but none is intended or implied by the dataset itself. It would be slightly easier coding to assume 0300 represents 3:00-6:00, but you might prefer the 1:30-4:30 window as it centers the datapoint in the time window.

Note that due to the coarseness (in space and time) of the underlying data, you should take the values you get (however you choose to interpolate) with a grain of salt. If your choice of interpolation vastly alters your ultimate results, you might try to find a different dataset (MSWEP is one possible alternative, although it is also heavily interpolated in many regions). MSWEP also offers 0.1 degree products: https://data.princetonclimate.com/

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.