You are looking the documentation of a development version. The release version is available at master.

Playing with features in iota2

Write outputs


It is assumed that examples in tutorial were properly run.

The first steps of iota2 correspond to the definition of data cubes: all temporal acquisitions are stacked together. For instance, with 10 temporal acquisitions and 4 spectral bands, an hypercube of 40 spectro-temporal features is used for the processing. If needed, additionnal steps are required to handle irregular time sampling and clouds/shadows, such as describe in this article: this is a common situation when processing several Sentinel-2 tiles. Such steps will basically project all the data onto a same and regular sampletime (whose step is define by the variable Sentinel-2 temporal_resolution). It is termed Gapfilling.

Because it can leads to a very large set of files, the default behavior of iota2 is to not write the temporal cubes used for the processing. Everythings are computed in memory (i.e., in the RAM), and only the necessary (or light) files are written to the hard drive. It also improves the processing time since writting data is time comsumming.

Off course, it is possible to tell iota2 to write outputs, for instance for further analysis. This feature is controled by the variable write_outputs . To do so, the variable must be set to True in the configuration file:

# Your parameters

With respect to the standard ouput, for each tile, the data cube before and after the gapfilling, and the complete cube of feature (spectral bands + spectral indices) after gapfilling are written. Using the same data than in classification tutorial, the folder /XXXX//features/T31TCJ/tmp/ contains now 3 additional files after a proper run of iota2.

└── T31TCJ
    ├── CloudThreshold_0.dbf
    ├── CloudThreshold_0.prj
    ├── CloudThreshold_0.shp
    ├── CloudThreshold_0.shx
    ├── nbView.tif
    └── tmp
        ├── MaskCommunSL.dbf
        ├── MaskCommunSL.prj
        ├── MaskCommunSL.shp
        ├── MaskCommunSL.shx
        ├── MaskCommunSL.tif
        ├── Sentinel2_T31TCJ_input_dates.txt
        ├── Sentinel2_T31TCJ_interpolation_dates.txt
        ├── Sentinel2_T31TCJ_MASKS.tif
        ├── Sentinel2_T31TCJ_reference.tif
        ├── Sentinel2_T31TCJ_TSG.tif
        ├── Sentinel2_T31TCJ_TS.tif
        └── T31TCJ_Features.tif

Select a subset of features

Default iota2’ behavior is to use the spectral bands and compute few spectral indices. It is possible to select a few of them for the processing. For instance, with Sentinel-2 to use only the spectral bands at 10m/pixel the variable keep_bands can be changed to

# Your parameters
keep_bands:["B2", "B3", "B4", "B8"]

Also, the variable extract_bands must be set to true:


If you don’t want the spectral indices, the variable features can be set to an empty list:

# Your parameters
features : []

Note that it is possible to include additionnal spectral indices using the variable additional_features. Again, it is possible to write such feature to the hard drive by setting write_outputs to True.

Use precomputed features

Once the features have been written on the hard drive, it possible to use them in iota2 and saving pre-processing time. It also possible to include additionnal precomputed features such as elevation or any feature that match the spatial resolution and footprint of the data.


It is up to you to check spatial coherence (resolution and footprint) of the provided data.

To tell iota2 to use your feature, your first have to specify the path to the data with the variable user_feat_path.

# Your parameters
user_feat_path: "/path/to/the/features/"

Then you have to explain to iota2 which features use and how to combine them with the other feature, this is done with the section userFeat. Assuming you want to use the datacube written in the first section (/Sentinel2_T31TCJ_TSG.tif/) you have to make a folder corresponding to the tile T31TCJ (one folder per tile, as for Sentinel-2 original data sets) and copy the data:

             └── T31TCJ
                 └── Sentinel2_T31TCJ_TSG.tif


You can use symbolic link

Then, in the configuration file you should set the section as


arbo:"/*" tells iota2 to scan all the folders in the directory, and patterns:"TSG" tells iota2 to process files with the pattern “TSG” in their names. It is possible to provide a list of pattern, e.g., “TSG,MNT,NDVI”.

see also the iota2 tiler tutorial which help users to tile their data to feed iota2.