I’ve found that checking data on weather underground more often than once every 15mins does little to add useful information for calculating watering schedule modifications. Even a single check just prior to a scheduled watering event is often enough, given returned data is valid and useful.
If weather data is used as a rain sensor of sorts, then simply factoring in forecast, and current co editions to that single check, will reduce accidental watering during rain events to a negligible level.
My suggestion would then be the definition of acceptable defaults for the correction of odd data loss and on demand requests for data used to adjust schedules, or display current/ forecast information. I highly doubt a computer of any sort cares what the weather was like 15mins ago if you yourself have no interest in using the information.
Outside of the meteorological fields high resolution historical weather information is geewiz at best in most cases. It’s good practice to constantly refine programs toward the average, and then handle the outliers.