Publications des agents du Cirad

Cirad

Toward an integrated approach to address LSLA processes. [ID: 768]

Lebourgeois V., Jahel C., Bourgoin J., Interdonato R., Augusseau X.. 2019. s.l., 1 p.. Open Science Meeting of the Global Land Programme OSM2019, 2019-04-21/2019-04-24, Bern (Suisse).

Large Scale Land Acquisitions (LSLAs) by private companies or states have seen a sudden increase in recent years, mainly due to the increase in demand for biofuel (i.e., caused by the increase in oil prices) and the increase in food demand (i.e., caused by the increase in world population and changes in dietary habits). These highly controversial phenomena raise many questions about production models, people's rights, resource governance, and are often at the root of conflicts with local populations. Even though global scale LSLA-related initiatives exist (i.e., GRAIN and LAND MATRIX), their data is often based on sources which may be incomplete or strongly biased (press articles, government data, individual contributions, scientific publications). For the above reasons, we here propose an approach that aims at detecting and characterizing LSLAs by exploiting multi-source satellite images. The idea is to use a multi-scale approach, i.e., using satellite images at medium, high and very high spatial resolution. After defining the spatio-temporal criteria for the discrimination of agro-industries, the main steps of the proposed approach are: (i) detection of potential LSLAs at national scale using MSR MODIS time series available since 2000; (ii) confirmation of the presence at local scale of an LSLA with landscape metrics from HSR imagery (i.e., Sentinel-2, Landsat-8); (iii) detailed characterization of identified agro-industries based on all previously cited satellite data completed with VHSR data (SPOT 6/7 or Planet). The process can be completed/integrated by an impact analysis on a test site of the implementation of an agro-industry on the territory. While all the steps may be performed by using classic Remote Sensing techniques, our perspective is also to test the effectiveness of advanced machine learning techniques (e.g., deep learning architectures) which can be trained on existing LSLA data to build models able to detect and characterize new LSLAs.

Documents associés

Communication de congrès

Agents Cirad, auteurs de cette publication :