Use third party data for classification
Currently, it is possible to use several ancillary data sources as input to the chain:
- the
userFeat
sensor, which provides structured image data, for example spot images6 - exogenous raster data used to calculate
external_features
, but which does not necessarily have to be added directly to the classification primitives - parameters for
external_features
which allows to provide csv files containing information for instance.
However, one might need to provide other types of data, and not necessarily in raster format. This data could be of various resolutions.
For example, weather data provided in a netcdf format, with a spatial resolution of 1 degree. Coupled with a precipitation grid with a resolution of a hundred metres.
This kind of data can make sense in classification or primitive calculation. Handling this kind of data would currently require converting it to raster at the working resolution before starting the chain, implying writing intermediate products and therefore increasing the storage burden.
We would need to find a way to integrate this data more efficiently than at present, by resampling it at the last moment, and if possible in memory in order to concatenate it to the features stack.