Meet with Debasmita - Githubissues

genophenoenvo / neon-datasets

Repository for code and derived data from NEON data products

MIT License

0 stars 4 forks source link

Meet with Debasmita #76

Closed jessicaguo closed 2 years ago

jessicaguo commented 2 years ago

To continue from 11/18 discussion about gcc_sd

[x] Confirm issues with pulling weather forecasts
[x] How to model error around the the mean?
[x] Meet with Arun and Debasmita about the EFI forecast
[x] Provide Debasmita with code for 30 separate weather forecasts

jessicaguo commented 2 years ago

Debugged the noaa_stack() function, which requires a date if doing consecutive pulls. Otherwise, it will stack all of the forecasted dates.

Also showed the native temporal SD in the gcc_90 data, which arises due to random processes. Therefore, adding weather or averaging across ensemble runs still won't improve it. Hopefully, the temporal SD in a moving window from the past can be used to adjust the across-ensemble SD and improve performance with CRPS.

jessicaguo commented 2 years ago

Met with Debasmita and Arun to decide how to obtain the CI for the gcc/rcc predictions. Empirical methods considered, then looked at the weather forecasts and decided to use each of 30 forecasts separately. Arun also proposed a weighting for the precipitation, as it appears that gcc variability depends in part on how bright or cloudy the day was