periodicity 'dekadal'?

[rabons]

Is your feature request related to a problem? Please describe.
I would like to calculate SPI values on dekadal periodicity, i.e. every ten days of a month (days 1-10 are dekad 1, days 11-20 dekad 2, days >= 21 are dekad 3 of each month). This is often used in agricultural drought assessments.

Describe the solution you'd like
Ideally, I would like to be able to run the spi command with --periodicity dekad, where the input netCDF file for precipitation is equally in dekadal time steps

Describe alternatives you've considered
I may have to consider using a different package altogether.

[monocongo]

Yes, this should be possible. We initially started with only monthly data but later added daily. So this would continue that work to add a new periodicity.

[rabons]

Thank you! Could you briefly point me to the files which would require modifying? I could attempt at doing it myself as a starter.

[monocongo]

Can anyone comment as to strategies we might use for testing our implementation of this new period type?

[monocongo]

Thanks @rabons

We need to add a new enumeration to the class compute.Periodicity, then make sure we can support 36 periods per year for each index type. Initially we can look for places where the check for monthly/daily is happening then add another case for dekadal datasets where it makes a difference.

[bennyistanto]

Can anyone comment as to strategies we might use for testing our implementation of this new period type?

+1 for this --periodicity dekad feature.

@rabons I just completed test using global dekad CHIRPS version 3.0, and calculate monthly accumulation rolling by dekad. Then calculate the SPI separately for dekad1, 2 and 3. Each of them produce 71 GB of file output for single scale (I tested for scale 6-month).

Concatenate all three together and transpose to time,lat,lon and save with compression using xarray, I got 57 GB for the final output (Global dekad SPI-6, 1981-2024, 0.05 spatial resolution).