what to do when dates have differing number of rows?

site nwis_04136900 has 24 obs per day for 5/6 of its record, 96 obs per day for the last 1/6. bayesian models with split_days=FALSE break here where MLE models and no-pooling bayesian models worked fine. what to do? could

1. refuse to model the site
2. manually set start-end dates to only include those dates with the maximum obs per day
3. manually set start-end dates and run multiple models, one per resolution. then recombine to create gpp_estBest, etc.
4. subset to the least common denominator (for nwis_04136900 that's 24 obs/day) and model all the days at that resolution