search

noaa-ocs-modeling / EnsemblePerturbation

perturbation of coupled model input over a space of input variables
https://ensembleperturbation.readthedocs.io
Creative Commons Zero v1.0 Universal
7 stars 3 forks source link

NaNs in the surrogate model for Florence #133

Closed FariborzDaneshvar-NOAA closed 8 months ago

FariborzDaneshvar-NOAA commented 8 months ago

@WPringle The percentiles file percentiles.nc for Florence with 12-hr leadtime has values for 374,602 nodes out of which, almost half of them are NaNs (mostly land nodes, brown dots in the screenshot below). Is it OK to have NaNs? The number NaNs vary among different percentiles. For example, for quantile=50, there are 172,012 NaNs, but for quantile=90 this number drops to 166,912!

image

Path to the Percentile file on NHC_COLAB_2: /nhc/hurricanes/florence_2018_OFCL_12hr_korobov_30/setup/ensemble.dir/analyze/linear_k1_p1_n0.025/percentiles.nc

I'm checking ds_percentile.sel(source='surrogate',quantile=90).quantiles.values

WPringle commented 8 months ago

That makes sense. The code automatically sets it NaN once the water level is below the allowed minimum depth to indicate it is dry.

FariborzDaneshvar-NOAA commented 8 months ago

Thanks @WPringle. Wasn't sure if there is an issue with the surrogate model and NaNs are valid or not