Assessing cropland abandonment in rural conflict areas using satellite imagery in humanitarian contexts

Details

Supervisors

Defourny, Pierre

Faculty

Faculté des bioingénieurs

Degree label

Master [120] en sciences agronomiques et industries du vivant, à finalité spécialisée: Geo-Information Science and Earth Observation for Environmental Modelling and Management

Abstract

In the struggles of armed conflict, challenges often inhibit food related humanitarian assessments to embattled regions. One viable solution to this hurdle is the application of satellite imagery. This Master's thesis investigates the interplay of conflict and population displacement on cropland abandonment in two areas of interest of the Borno province, Nigeria through a remotely sensed analysis. The study leverages advancements in remote sensing technology, utilizing high-resolution satellite imagery from Sentinel-2 and Planetscope sensors to monitor cropland abandonment from 2018 to 2022, during and after the Chad Basin campaign. The research identifies temporal patterns related to phenology, conflict, and population displacement and relates these patterns to remote sensing data. A discernible north-south rainfall gradient is noted, determining October as the optimal month for cropland classification. A conflict pattern is established with the intensification of the conflict following the Chad basin campaign (2018 – 2020) correlated with population displacement and return. The random forest algorithm for classification of 'cropland' and 'non-cropland' areas reveals variable patterns across areas of interest. The algorithm, initially based on the year 2018, has been effectively ported to subsequent years, thus validating a 'portable classification model'. Baga, subjected to intense conflict, demonstrated cropland decrease during the high intensity conflict period, followed by recovery as conflict subsided, confirming the hypothesized correlation. In contrast, Dikwa, under government control and a center of displaced population camps, displayed consistent cropland patterns, demonstrating relative stability. Change detection results elucidate shifts from croplands to non-croplands during conflict intensification and subsequent recovery. Overall, despite the few discrepancies, the sensors demonstrated high accuracy in data portability and classification, underscoring the utility of remote sensing in monitoring spatio-temporal cropland abandonment dynamics in hard-to-reach conflict areas. In summary, this research provides valuable insights into cropland abandonment in conflict zones and proposes a rapid, yet nuanced, remote sensing methodology to assist humanitarian response.