You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This request originated from Erin Crandall.
It would be nice to have an option in zipsByProduct() and loadByProduct() where data was only downloaded/loaded if it was available from all the sites listed within a data range. E.g. Between Jan 2013 and Dec 2018, only download data if X product is available from all of HARV|BART|CLBJ.
Describe alternatives you've considered
This is possible to do post downloading but with data products that have large quantities of data one still has to deal with the large download size of data that isn't of interest. This also put additional data management burden on the data user.
The text was updated successfully, but these errors were encountered:
I'd actually misunderstood the original request which is about two different data products: Ability to query sites that have both of 2 different data products available. For example, I want data from all the sites that have fish data and stream geomorphology surveys done in an overlapping time range. I want to see all the sites that have both types of data available and download only those.
Thanks for the suggestion! Building this into the download functions directly would be a bit cumbersome, but I can definitely make some enhancements to the catalog-querying functions to produce a clean summary of availability, which would give you the info you'd need.
This request originated from Erin Crandall.
It would be nice to have an option in zipsByProduct() and loadByProduct() where data was only downloaded/loaded if it was available from all the sites listed within a data range. E.g. Between Jan 2013 and Dec 2018, only download data if X product is available from all of HARV|BART|CLBJ.
Describe alternatives you've considered
This is possible to do post downloading but with data products that have large quantities of data one still has to deal with the large download size of data that isn't of interest. This also put additional data management burden on the data user.
The text was updated successfully, but these errors were encountered: