What lead time would minimize data assimilation artifacts (if any) in 2D variables?

[snowidget]

Hi there! I am not a meteorologist in any practical sense -- so please correct me if I am mistaken or incomplete anywhere.

I am conducting some analyses that uses the total precipitation, percent of frozen precipitation, surface pressure, and 2-meter air temperature, dewpoint, and specific humidity variables from 1-hr HRRR forecasts. I am aware that some hydrometeor and temperature observations are assimilated, and am curious to what extent these would "echo" in the short range compared to if no assimilation were done and these variables were calculated/diagnosed from WRF alone. I imagine DA artifacts in these 2D fields could be most present at early lead times, and progressively be replaced by "pure physics" at longer lead times as the model does more integrations. But, it also seems likely to me that even longer lead times could still be sensitive to DA if we try to string together a time series for a given lead time.

It would be great to have technical guidance on: (1) Is there a minimum/recommended lead time that would more or less neutralize DA artifacts in those 2D fields during storms? (2) If there is, would stringing together a time series of a given lead time make physical sense? (e.g., for fxx=12, could two maps of valid times 1300 and 1400 be drastically different from each other [and physically "senseless" in a time series], depending on what gets assimilated at 0100 and 0200?)

Happy to give more context if necessary. Thanks so much

[EricJames-NOAA]

@snowidget sorry for the late response! It's a very good question. The answer depends to some extent on the forecast variable under consideration. There are several components of the data assimilation, including the 15-min radar data assimilation (Weygandt et al. 2022), the non-variational cloud analysis (Benjamin et al), and the hybrid 3DEnvar meteorological analysis.

To answer your questions:

(1) In general, we don't expect DA artifacts to last beyond 1 h or so.

(2) Using a time series of a given lead time could work for certain applications, such as looking at long-term "climatologies" or average diurnal cycles. However, it may not work as well for case studies for individual storm events, since the evolution of precipitation / reflectivity may be non-physical. While the impact of data assimilation should be fairly small beyond 6 h or 12 h into the forecast, there is still the impact of different initial and lateral boundary conditions.