ALADIM-VHR: Automatic LAndslide Detection and Inventory Mapping from multispectral Very-High Resolution images¶

Upload input data¶

The input images must be uploaded by the user. The image file names must contain the date and the time of the acquisition in the format YYYYMMDDTHHMMSS. For this tutorial we present the example of a couple of multispectral images and a couple of panchromatic images from Spot6 and Spot7 satellites. The first image was acquired before Hurricane Matthew on 14-04-2016 and the second after the event on 04-04-2017. Ideally, images acquired at the same season should be used to obtain similar radiometric signatures.

Upload your data:

Drag and drop your images in the fields of the service panel:

Create an archive for the ensemble of your input shapefiles (training_areas.shp, training_samples.shp and aoi.shp). The framework requires a flat .tar.gz format (i.e. the contents of the archive file must not include folders). A tutorial about the input dataset creation is available here: doc:tutorial <aladim_input_dataset_preparation>

Upload the archive:

Drop the archive in the field of the service panel “shape files uri”:

Set the processing parameters¶

There is a total of 11 processing parameters. When hovering over the parameter fields, you will see a short explanation for each of the parameters.

The figure below summarizes the parameter settings for this test.

• Image: VHR sensor: The types of VHR sensors currently supported are Pléiades, SPOT6/7, Planets (Dove) and Geo-Eye).

• Image: Fusion of multispectral and panchromatic images: If set True, a fusion between the multispectral and the panchromatic images is performed (using a meanshift pansharpening approach).

• Use OSM classes: Open Street Maps (OSM) classes used to exclude areas from the classification. The choice is defined by the user with the “OSM Tags” (see www.geofabrik.de/data/geofabrik-osm-gis-standard-0.7.pdf, pages 25-26).

• Use HOG (Histogram Of Gradient) attributes: If set to True the HOG attributes will be computed and integrated in the classification

• Spectral indice set used: If set to basic, the main spectral indices (17) will be computed and integrated in the classification. If set to advanced a large number (30) of spectral indices will be computed and integrated in the classification. The indices integrate with the advanced choice and not the basic one are dedicated to vegetation puropses. The advanced choice will increase the computation time.

• Use DEM derivated attributes: If set to True the DEM derivated attributes will be computed and integrated in the classification

• Use textural attributes: If set to True the textural attributes will be computed and integrated in the classification

• Use morphometric attributes: If set to True the morphometric attributes will be computed and integrated in the classification

• Image: Multispectral band order: Order of the bands for the multispectral image(s).

• Image: Homogeneity: Defines the homogeneity (low) or diversity (high) of the segment radiometry (see examples below).

• Image: Minimum segment size (pixel): Defines the minimal size of the segments (to be merged with their most similar neighbors or to be deleted if isolated).

  • A very low value corresponds to 10 pixels, e.g. 2.5 m2 for a 0.5m pixel resolution (Pleiades P+MS)
  • A low value corresponds to 100 pixels, e.g. 25 m2 for a 0.5m pixel resolution (Pleiades P+MS)
  • A high value corresponds to 1000 pixels, e.g. 250 m2 for a 0.5m pixel resolution (Pleiades P+MS)
  • A very high value corresponds to 5000 pixels, e.g. 1250 m2 for a 0.5m pixel resolution (Pleiades P+MS)

See examples of segment sizes and shapes for several parameter settings applied to Pleiades (P+MS) images.

Run the job¶

• You are good to go. Click on the button Run Job at the bottom of the right panel. Depending on the allocated resources the execution will require a few hours to terminate.

• Once the job has finished, click on the Show results button to get a list and a preview of the results (Job example).

• The image with the names landslide_inventory_map (i.e.the landslide inventory map which appears in the results) is binary classifications of shapefile with the name landslide_inventory_map_full (i.e. the whole map with probability) using the F-scores as a measure of model’s accuracy on the dataset. F-scores are used to evaluate binary classification systems, which classify examples into ‘positive, e.g. presence’ or ‘negative, e.g. absence’. The F-score is defined as the harmonic mean of the model’s precision and recall. Two scores are used: F-1 considers both precision and recall to be equally important in the model and F-2 considers recall to be twice as important as precision in the model.