The problems were with the WATCH data set only, but because the WATCH+WFDEI data set also includes these data, the latter data set was also affected
Although the differences in total precipitation are below 10% in many areas, they reach 10-20% for large regions like Siberia and North-West America (and up to several tens of percent in low precipitation areas), and we consider these errors to be sufficiently severe to warrant re-running your simulations for the WATCH and WATCH+WFDEI data sets.
The figures below (click for hi-res) show multi-year (1958-2001) monthly means of absolute and relative differences between the old and the fixed WATCH data for total precipitation and snowfall (don't mind the year 2001 given in the figure headers, it refers to the last time step in the NetCDF files only).
Relative differences in Precipitation (rain + snowfall)
Absolute differences in Precipitation (rain + snowfall)
Relative differences in Snowfall
Absolute differences in Snowfall
Long wave radiation and humidity were also affected by a less dramatic truncation problem, resulting in monthly mean differences of less than 1%. All other variables have not been affected by the truncation error.
Furthermore, we have come across anomalies in the wind field for approximately 20 days in 1978 in the original WATCH data. We are in contact with the developers of the data set.
Additionally, pixels in the two easternmost columns (179.25E and 179.75W) of our data were shifted by one latitude to the south.