An evaluation of uncertainty in light measurement (PAR):

In practice, the ideal sensor characteristics will always be approximated or factually reproduced with certain tolerances.

These tolerances will manifest themselves when the actual illumination or radiation condition differs from the condition during calibration which is normaly the case.  Sensor outputs will show up different values when compared to each other.  Differences will be larger when sensors of different type and construction are compared.  For all sensors an absolute error will come up, hardly to predict, depending on the actual radiation condition.

This in fact is the uncertainty where all light measurements are dealing with.

One has to realize that values of 10 to 25% absolute error may occur besides long term effects.  Long term effects can be minimized by good treatment of the sensor(s) and regular inspection.  Inherent to their construction, 4PI-uniform spherical (mini)sensors are most fragile.  A flat cosine-response sensor can be constructed more rigid.

Some remarks on resolution:

Spatial resolution is the capability to distinguish irradiance variation with distance (for instance when measuring extinction as function of depth). lack of resolution may become critical, when artificial sources on relative small distance are used for let's say incubation purpose.

A flat sensor has a perfect resolution in one direction, a spherical sensor should be dimensionless to have perfect resolution.  The actual finite dimension causes a certain lack of resolution and inherently an over-estimation of a measured irradiance value in a certain position determined by the center of the spherical sensor element.

When the sensor is moved in the source (main) direction over a distance equal to half its spherical diameter, one get's an idea of the order of magnitude.  In fact sensors of different size cannot be compared in the same position.  For a point source there is a calculated over-estimation of about 10% with a 10mm sensor on 10cm distance.