Spatial aggregation fix

[dbkhout]

Although utils.dimension will be deprecated in pysteps-v2 (#223), I'm proposing a small update of its aggregate_fields_space function to fix a potential error.

The snippet below is currently used in both aggregate_fields_time and aggregate_fields_space to select the appropriate aggregation method based on the precipitation unit.

if unit == "mm/h":
    method = "mean"
elif unit == "mm":
    method = "sum"
else:
    raise ValueError(
        "can only aggregate units of 'mm/h' or 'mm' " + "not %s" % unit
    )

While this is indeed the correct approach for the temporal aggregation, it leads to unexpected results* for the spatial counterpart. Hence, I'm suggesting to update aggregate_fields_space to use 'mean' aggregation for both 'mm/h' and 'mm' units, as per this PR.

---

(*) illustration: corresponding example from pystep's test suite
Assuming grid cell dimensions in meter and 1 mm = 1 L/m², the total rainfall volume over the complete field is:

• 10 * 10 * 1 m² * 1 L/m² = 100 L before aggregation (10x10 grid of 1x1 cells with value 1 mm)
• 5 * 5 * 4 m² * 4 L/m² = 400 L after aggregation (5x5 grid of 2x2 cells with value 4 mm)

while this value is expected to remain unaffected by the aggregation.

[dnerini]

Good catch, @dbkhout ! Aggregating mm in space should use the mean instead of the sum, you're absolutely right.

Thanks for the bug fix and apologies for the very slow response.