Forecasting is using the best weather data available at the time (radar, satellite, historical) to derive max temp, min temp, humidity, dew, and wind. This ultimately is a forecast and is the best we can do at time of forecasting the ET value.
So this is interesting, the weather station you have chosen (MID_D2736) was reporting solar radiation data up until July 21st and then stopped. This is why you are seeing a large change from the data reported from the same station. We interpolate solar radiation when not received from a station (which is most of the time) and our interpolated value must be different than what the station was reporting. I can't verify the data validity of the chosen station. Might want to use an airport or another station that has more consistent data.
This is due to the delta in data we started receiving.
My thoughts on weather stations in general is that if you have one that is either under reporting or over reporting and modify (adjust) flex daily accordingly to how your yard is doing, everything will relatively work itself out.
Now, if you've modified your flex daily schedule, and your station starts to report very different data (or in this case missing data) then you will see large variations as you are witnessing.
I won't normally do this but exported your observed weather data to show the impact of the data change (absent solar radiation).
Which ET value is closer/better (with/without solar radiation)? Without doing a lot more analysis I can't say. If the solar radiation was bogus or not...
Observed weather data:
Hope this helps.