Agwilh Collet's picture
Land Cover with Sentinel 2 (CNES Internship 2018)

Among the objectives of the Recovery Observatory is the production of the updated land-use map of Haiti by exploiting optical satellite data as well as processing tools and open access data such as the iota² processing chain developed by the BIOspace Spatial Study Center (CESBIO) and OpenStreetMap data. In this context, the purpose of this internship was to identify the potential of iota² with limited baseline data. Iota², designed for the territory of metropolitan France, had never been tested elsewhere on a whole country.

Work done with the CESBIO IOTA2 - CNES chain

Imagery : Sentinel 2 (

Conclusion (Summary)

In this work, the iota² soil classification chain was executed with the 2017 time series of Sentinel-2 images on Haitian territory and reference data mainly from OpenStreetMap. The pre-treatment of the reference data was the major step in this stage given the large volume of databases and their quality in terms of cleanliness of the content. The main results showed that the performance of the chain is all the more satisfactory as relevant reference data are available. Iota² has been operational in Haiti with productions in line with baseline data. In this regard, the potential of OpenStreetMap data in the production of land use maps by large-scale classification remains debatable with respect to the heterogeneity of their availability in the world. The overall accuracy of the map produced was 96.4% and the Kappa index was 88.7%. Nevertheless, the visual evaluation of the result is not so satisfactory. Confusion between classes and overclassification, particularly of the Woody Vegetation class, highlighted the acute need for relevant baseline data from the Naked and Prairie classes.


Final delivery is available in the attached files :
Attached you will find the internship report, the final presentation of the internship and an example of classification.