From Land Cover to Land Use Systems Mapping: Detection and Characterization of Large Scale Agricultural Investments (LSAIs) from Satellite Imagery. Application to Senegal

I'm Yasmine Ngadi Scarpetta, a PhD student at UMR Espace-Dev (IRD) and UMR Tetis (CIRAD) in the Maison de la Télédétection (MTD) in Montpellier. My subject is the automatic detection of large-scale land acquisitions (also known as land grabs). As a bio-engineer in agronomy, I worked for several years in a research unit at the Université Libre de Bruxelles dedicated to the exploitation and analysis of satellite data. Following this experience, I specialized in remote sensing with a specialized MSc at Utrecht University, and came into contact with UMR Tetis, which offered me an internship on the subject. The interaction went so well, and the subject is so interesting, that I decided to undertake a thesis.
The research project I'm working on aims to explore the potential of satellite data to automatically detect and characterize large-scale agricultural land acquisitions (LGAs) at different scales. Direct and/or indirect indicators will be extracted and analyzed according to the different types of LSA being investigated. This work will initially be carried out on a national scale, in Senegal.
I'm interested in this subject because it's an important issue in my home country (Colombia). It's also an innovative, challenging subject with a strong social impact, combining my different areas of expertise (agronomy, remote sensing, data mining). In this first phase of research, the use of satellite image time series (SITS) to detect changes linked to the installation of ATGEs appears to be a promising technique.

• Starting date: 1st November 2020
• University: University of Montpellier
• PhD school: GAIA
• Scientific field: Geomatics– Remote sensing
• Thesis management: Anne-Elisabeth Laques (IRD, UMR Espace-Dev), Agnès Bégué (CIRAD, UMR Tetis)
• Thesis supervisor: Valentine Lebourgeois (CIRAD, UMR Tetis)
• Funding: Contrat doctoral Université de Montpellier
• #DigitAg : Cofunded thesis – Axe 5: Fouille de données, analyse de données, extraction de connaissances – Challenge 0 : sujet transversal et Challenge 6 : TIC et gestion du territoire agricole

Keywords: SITS, MODIS NDVI, land use and land cover change, BFASTm-L2, LSLA, unsupervised change detection, change metric

Abstract: Increasing demand for water, food and energy has led to dramatic competition for land, resulting in a global land rush in the form of large scale agricultural investments (LSAI). Due to their many potential negative impacts and the opacity surrounding them, accurate detection and characterisation of LSAIs in space and time is required. The increasing availability of dense satellite imagery time series (SITS), together with ever-improving change detection algorithms, is useful in this task. While SITS change detection algorithms are efficient at detecting abrupt and gradual changes phenological time series, there is still much room for improvement when it comes to detecting seasonal changes.
The primary objective of this research was to automatically detect, in an unsupervised manner, the implementation of LSAIs in Senegal based on remote sensing data. This work is structured around three interrelated papers. The first presents a fast and unsupervised approach (BFASTm-L2) developed to detect, in full MODIS NDVI SITS at the pixel level, the breakpoint associated with the largest pattern (i.e. mostly seasonal) change of the time series. Compared to other change detection algorithms (BFAST Lite, EDYN and BFAST monitor), BFASTm-L2 proved to be particularly sensitive to seasonal changes and efficient in highlighting LSAIs in Senegal. This supports the hypothesis that changes induced by land use systems such as LSAIs are very often of a seasonal type. The second paper sought to differentiate the contribution of LSAIs from the main drivers of change (climatic, natural and anthropogenic) at a national-scale, relying mainly on three time series-based change metrics calculated at the pixel level (magnitude of change, direction of change, dissimilarity), which, when combined into a unique composite map, provided insights into land dynamics. LSAIs were shown to have a specific ecoregional signature of change. Finally, the third paper aims to refine the detection of the deals by automatically locating potential hotspots of change related to LSAIs in two contrasting ecoregions of Senegal through the segmentation of a BFASTm-L2-based magnitude of change map combined with object-based K-means clustering. In this last study, key discriminative metrics (textural and structural) derived from higher resolution imagery (Landsat) were combined with the spectro-temporal ones coming from MODIS NDVI SITS to provide a generic characterization of LSAIs.
Through its specific focus on large-scale detection of LSAIs, this project contributed to the land change community by improving the understanding of land dynamics and the drivers of change behind the detected changes.

Jury compound:
Damien ARVOR, CNRS, Université de Rennes, LETG, Rapporteur
Nicolas DELBART, Université Paris Cité, LIED, Rapporteur
Liam WREN-LEWIS, INRAE, Paris School of Economics (PSE), Examinateur
Carmen GERVET, IRD, Université de Montpellier, Espace-DEV, Examinatrice
Agnès BEGUE, CIRAD Montpellier, TETIS, co- Directrice de Recherche
Valentine LEBOURGEOIS, CIRAD Montpellier, TETIS, Encadrante
Anne-Elisabeth LAQUES, IRD Madagascar, Espace-DEV, co-Directrice de Recherche

Read the thesis manuscript

Social network: ResearchGate

Contact : valentine.lebourgeois[AT]cirad.fr