This sounds similar to this issue: Base flow and Aggregate results are coming with NoData
In that case, nan
pixel values on the edges of the precip rasters were propagating through to the results. That might explain why buffering the inputs and filling in nodata values resolved the problem.
Also, we fixed the bug with nan
values being used in the math, that fix was released in version 3.14.3.
Which version of invest are you using @nicole.pommares ? Please always upload your logfile when posting about a model run, it can tell us the version, and other useful info. Thanks.