# How can I interpret negative values of potential evapotranspiration?

If I extract Potential Evaporation (PET, W/m$^2$) from the National Centers for Environmental Prediction (NCEP), climate reanalysis data (downloadable as netCDF files here), there are some negative values, e.g. for summer in southern Manitoba (50N, 100W), I get the attached distribution of PET values.:

Is this a real physical process, such as condensation (dew formation), or is it an error (e.g. should I average over these values or set them to zero)?

The metadata (contained in the files) says that the valid range is -800, 5200; here is further description of the 'pevpr' variable (provided in the data header):

• long_name = "Monthly Mean of Potential Evaporation Rate"
• valid_range = -800,5200 ;
• units = "W/m^2" ;