search

USGS-R / protoloads

Prototyping and exploring options for broad-scale load forecasting
0 stars 4 forks source link

Rerun models #67

Closed aappling-usgs closed 6 years ago

aappling-usgs commented 6 years ago

Models are done!

I'm rebuilding the figures now but think I'd better put those in a separate PR...preds_v_time seems to have acquired mismatched magnitudes for the NWIS vs NWM/loadmodel values, which will need some investigation. Preview of the other two looks good, though, I think:

fig_error_v_leadtime

fig_input_data

aappling-usgs commented 6 years ago

Only...are those magnitudes right? Here are our plots from 5/30 (error_v_leadtime) and 5/21 (input_data):

fig_error_v_leadtime ^ relative errors are approximately 100x higher in today's plots than they were in this one

fig_input_data 2 ^ discharge and flux are both 100x higher in today's plots than they were in this one

jzwart commented 6 years ago

cool! is there a unit conversion issue?

jzwart commented 6 years ago

ah I see you made an issue for units :)

aappling-usgs commented 6 years ago

Yes, I think the units issue is now captured in #68.

jzwart commented 6 years ago

do you want to add a divide by 100 in gather_forecasts to this PR @aappling-usgs ?

jordansread commented 6 years ago

this is so cool!

aappling-usgs commented 6 years ago

@jzwart - lemme do that as a second PR so it's really clear how to adjust it later when i have time to rerun the models. i'm ready to merge this PR if you are.

jzwart commented 6 years ago

sounds good, will merge