sjordan29
commented
10 months ago I ran each of the models provided by @jrutyna in a Jupyter notebook with a 1, 2, 5, 10, 20, and 30-second timestep. Results show we will need to stick with a 1-second timestep, as error increases from -4.2e-13% to 7.1% from the 1-second to 2-second time interval, and only increases from there:
@jrutyna & @aufdenkampe, just merged #54 to address #47, so I think we are ready to assess whether it'd be possible to have a longer time step than 1 second to decrease model run time.
@jrutyna - can you set up a few models with various timestep lengths (maybe ranging from 2-30 second timesteps depending on what you think is reasonable?). How long would the total model run need to be for you to feel comfortable assessing the impact? For example, could we do a 1-hour or 3-hour model run rather than a day or several day model run for this classification?
Let me know if you want to do the analysis between model runs or if you want me to take care of that. I can use your post-processing scripts if you want me to compare and I can present the results here, or you can take over. Whatever works best with your schedule this week!