CDO

I have 4 times 6-hourly precipitation for a day, and I want to calculate the daily precipitation total.

The 6-hourly precipitation files are:
ST4.2015120206.06h.grb2
ST4.2015120212.06h.grb2
ST4.2015120218.06h.grb2
ST4.2015120300.06h.grb2

I first put them into a single file:
grib_copy ST4.2015120206.06h.grb2 ST4.2015120212.06h.grb2 ST4.2015120218.06h.grb2 ST4.2015120300.06h.grb2 1.grb

Then I add 6-hourly precipitation to get daily precipitation:
cdo --timestat_date last timselsum,4 1.grb 2.grb

Using "grib_ls -p missingValuesPresent,numberOfDataPoints,numberOfMissingValues", I know that each 6-hourly precipitation file has missing values:
ST4.2015120206.06h.grb2: # of missing = 270159
ST4.2015120212.06h.grb2 : # of missing = 270159
ST4.2015120218.06h.grb2 : # of missing = 270159
ST4.2015120300.06h.grb2: # of missing = 270229

But when I check the number of missing values for the daily precipitation 2.grb, it has less missing values compared to each individual 6-hour file:
2.grb: # of missing = 270129

The only explanation is that, as long as a grid has at least 1 valid value out of 4 times, "timselsum" will treat this grid as NOT missing.

How to avoid this?