The ERA5 runoff is a grid cell average, If we are considering a grid cell which is designated as a land cell ( i.e. where the value of the land-sea mask (LSM) is >0.5), then the runoff is the flux of water (per unit area) going out of the entire grid . 

Hence, the total runoff from a grid point is:

runoff x grid_box_area

Please note that on the 0.25x0.25 grid, there will be some interpolation errors, because this is not the native grid of the ERA5 data.

Over glaciers (eg Antarctica and Greenland), the runoff is not very accurate because there is no glacier mass balance model being used at the moment.

Note, the runoff parameter (code 205) is an accumulated parameter, so it is the amount of runoff during the accumulating period, which for hourly ERA5 (HRES) is in the preceding hour (for monthly means of daily means, where its per day). Please see the ERA5 data documentation for more details about mean rates and accumulations:

https://confluence.ecmwf.int/display/CKB/ERA5%3A+data+documentation#ERA5:datadocumentation-Meanratesandaccumulations

Thanks,

Kevin