Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Pays Bas → 2020 → Multi-temporal classification of Land Use & Land Cover of the Inner Niger Delta in Mali using Sentinel-2 satellite imagery in Google Earth Engine

University of Amsterdam (2020)

Multi-temporal classification of Land Use & Land Cover of the Inner Niger Delta in Mali using Sentinel-2 satellite imagery in Google Earth Engine

Juch SE

Titre : Multi-temporal classification of Land Use & Land Cover of the Inner Niger Delta in Mali using Sentinel-2 satellite imagery in Google Earth Engine

Auteur : Juch SE

Université de soutenance : University of Amsterdam

Grade : MSc Earth Sciences – Environmental Management 2020

Résumé
Worldwide, water and food security continue to be under major threat due to the consequences of anthropogenic climate change and growing populations. Such is also the case for the Inner Niger Delta (IND) of Mali. Here, the local population depends directly on agriculture, mostly rice cultivation, for their livelihood. They depend heavily on the yearly floods as their main water source. The farmers adjust their cultivation practices based on the timing of the floods and the flood levels to secure a successful harvest. The online early warning system and flood projection tool ‘OPIDIN’, provides such information. The locations of the agricultural fields, the location of residences and seasonal Land Use and Land Cover (LULC) should be known in detail to improve the OPIDIN projections. This research developed a methodology for temporal LULC classification using the cloud-based platform Google Earth Engine and Sentinel-2 imagery in order to do this. It consist of a 2-steps methodology consisting of a supervised LULC classification followed by a multi-temporal clustering method based on NDVI timeseries. The smile Random Forest proved most appropriate classifier to use despite performing poorly regarding urban areas, reaching an overall accuracy 80% and a kappa value of 0.74. Using a NDVI timeseries-based clustering improved the classification of the vegetation dataset, but not for the urban dataset. In order to use this methodology for the identification of urban areas, improvements are still required. However, this methodology could already improve the OPIDIN tool by separating the agricultural fields from the natural vegetation.

Présentation

Version intégrale

Page publiée le 16 avril 2021