SPI index error: FitError: Optimization converged to parameters that are outside the range allowed by the distribution.

[PraewKPT]

Setup Information

INSTALLED VERSIONS

python: 3.11.7
boltons: installed
bottleneck: 1.3.7
cf_xarray: 0.8.7
cftime: 1.6.2
click: 8.1.6
clisops: None
dask: 2022.02.0
flox: None
jsonpickle: 3.0.2
lmoments3: 1.0.6
numba: 0.58.1
pandas: 2.1.4
pint: 0.23
scipy: 1.9.3
sklearn: 1.3.2
statsmodels: 0.14.0
xarray: 2023.11.0
xclim: 0.47.0
Anaconda-based environment: no

Context

How can I fix this problem?
"FitError: Optimization converged to parameters that are outside the range allowed by the distribution."
I don't know where is the problem from?

Code of Conduct

• I agree to follow this project's Code of Conduct

[aulemahal]

Hi @PraewKPT,

This error does not ring a bell to me, but may be my colleague who coded SPI knows about it. In any case, would you be able to provide a minimal example that raises the error ? What data are you using, is it shareable ? Could you show us what code you have and which parameters you use ?

Thanks.

[coxipi]

Hi! It seems like a scipy error. I'll wait for details. It could be a bug on our side, but I suspect it could also be the data that is hard to fit? We'll see.

[PraewKPT]

Thank you for your help. I am new to SPI.

• I downloaded the data .nc file from this link >> https://data.ceda.ac.uk/badc/cru/data/cru_ts/cru_ts_4.03/data/pre
• file name "cru_ts4.03.1901.2018.pre.dat.nc" , file size 2.7 GB

[coxipi]

Hi @PraewKPT, thanks for sharing more info.

• Could you also show the python code you used to produce this? A bigger error message for context would be good too.
• I'm assuming you won't get the error if you set method="APP", in SPI, e.g. xclim.atmos.standardized_precipitation_index(pr,...,method="APP",...) can you confirm this?
• Can you isolate a small spatial region / a single pixel where the problem occurs? e.g. try your computation with:

pr = pr.isel(lat=slice(100,120), lon=slice(100,120)) # or something else that makes sense in your context

and see if you still have the error. Maybe your problem only happens on a few pixels and it would be hard to find it, so if you're not successful after trying a couple small regions, don't lose too much time on this. But if it works, it would be nice if you can simply share a small .nc file here.

Cheers!

[PraewKPT]

Hi @coxipi , thank you for your advice.
I haven't found an error when I set method = "APP".

[coxipi]

The method "APP" is an approximation. In the case of the gamma distribution, it's based on an approximation of the maximum likelihood formulation of the gamma parameters, see eq.22 here