See review attached

The paper presents a thorough cross comparison of multiple precipitation sensors deployed at the ARM Southern Great Plains site over the last 30 years alongside a methodology for, description and application of a new data product which combines those sensors. The paper is well written and easy to understand. My review focuses mainly on the first cross comparison section as the new value added product, while not ground breaking in its methodology, is a sensible addition, and is well described with a fitting application chosen to use as an example.

In addition to the comments within the attached pdf I have the following more general comments which are easier to present here:

1. The authors do not give reference at any point to the WMO intercomparison studies (both field and laboratory) of rain gauges (see citations in the pdf). I believe these papers may provide the authors with extra context for their findings and can be incorporated well into the paper. In particular, they should provide more information around the choice of the PWD as the reference sensor, and the potential implications for bias this may introduce.

2. The random forest application add little value to the paper. Only 2 of the models have an R2 above 0.5 and in both those cases the RF is dominated by the event mean and max rate. I'd suggest removing it, especially as its findings don't impact the resulting development of the new combined product.

3. The methodology chosen for the combined methodology does not take into account the findings from the previous section beyond the removal of the two gauges that deviate most strongly. I believe if the WMO intercomparison results are considered and the findings in section 2 there is potential future scope for an update which weights the instruments based on previously published accuracy findings and their accuracy within each event (based on duration and total accumulation). For instance the TBR and Pluvio would seem to be the more accurate choices based on the WMO results, and the mean could be weighted to account for this in events which suit their use. I'd suggest the authors consider this as a future extension/improvement rather than discounting the current methodology which provides a sensible first step.