Some ECMWF parameters are described as being valid at the specified time, the validity time. Sometimes they are referred to as "instantaneous" parameters, and may even have "instantaneous" in their name. Some forecast and most, if not all, analysed parameters fall into this "instantaneous" category.
However, although they are valid at the specified time, such parameters cannot represent variability on short time scales, so effectively they represent an average over time scales equivalent to length of the model time step because they do not represent variability on time scales shorter than this. Some forecast and most, if not all, analysed parameters fall into this "instantaneous" categorya time average. This arises because the values are a grid box average, without variability on spatial scales smaller than the spatial grid. In practice, the spatial grid determines the model time step , so the length of the effective time average is loosely related to the time step. When comparing with high frequency observations, the latter should be averaged so as to remove the variability that is not represented in the model.
Note, that in addition, such parameters may, or may not, be further averaged in time. For example, they hourly "instantaneous" values might be averaged over a month from hourly data, in which case they would be archived in a monthly mean "stream". This would be indicated in other metadata in the GRIB header of the data, as would details of the averaging frequency and period.
...