The atmosphere is a chaotic system. Very small differences in its initial description can grow with time into different positions and intensities of weather systems.
Every new run adds newer observations from stations, satellites, aircraft, ships and radar. A change does not necessarily mean a mistake; it often means better information has arrived.
Uncertain initial conditions
We cannot measure every point in the atmosphere. Data assimilation combines observations and a previous forecast to construct the best possible initial state.
Uncertainty can be larger in data-sparse regions and later travel towards Greece with the atmospheric flow.
Model and resolution
Every model approximates clouds, rain, turbulence and land interactions. Different resolution or parameterisation produces different local detail.
Higher resolution may improve terrain but does not remove uncertainty or guarantee the exact location of a thunderstorm.
What an ensemble adds
An ensemble runs many variations with small changes in initial conditions and model formulation. Agreement increases confidence in the broad evolution; spread calls for a range of scenarios.
At longer lead times, place more weight on the trend and less on an exact hour or amount for one city.
These explanations and calculations are educational. For safety, occupational, health, marine or aviation decisions, follow the appropriate official services.