An aggregation of a datacube can be thought of as a grouped reduce. That means it consists of two steps:
- Grouping via a grouping variable, i.e. spatial geometries or temporal intervals
- Reducing these groups along the grouped dimension with a certain reducer function, e.g. calculating the mean pixel value per polygon or the maximum pixel values per month
While the layout of the reduced dimension is changed, other dimensions keep their resolution and geometry. But in contrast to pure reduce, the dimensions along which the reducer function is applied still exist after the operation.
Simplified aggregate(👪 👩👦 👨👩👦👦, countFamilyMembers) => [3️⃣, 2️⃣, 4️⃣]
A temporal aggregation (e.g. aggregate_temporal) is similar to the downsampling process, as it can be seen in the according image above. Intervals for grouping can either be set manually, or periods can be chosen: monthly, yearly, etc. All timesteps in an interval are then collapsed via a reducer function (mean, max, etc.) and assigned to the given new labels.
A spatial aggregation (e.g. aggregate_spatial) works in a similar manner. Polygons, lines and points can be selected for grouping. Their spatial dimension is then reduced by a given process and thus, a vector cube is returned. The vector cube then has dimensions containing features, attributes and time. In the graphic below, the grouping is only shown for the first timestep.
Learn how to use Aggregate operators with this interactive exercise: