Skip to Main content Skip to Navigation

Une approche multiscalaire par télédétection pour la cartographie et la caractérisation des systèmes agricoles à l’échelle régionale

Abstract : : In a context of regional land-use planning, agricultural systems’ mapping - crop types and cropping practices – allows monitoring of what is being produced, where and how, and therefore represents a key element for regional assessment of the agricultural production and its environmental impact. The production of information on agricultural systems generally requires a lot of data and expertise. This information is thus very heterogeneous in quantity and quality in space and time, the availability and updates being extremely variable between countries and regions. Remote sensing, through its ability to retrieve synoptic spatial information on the state and dynamics of vegetation from satellite images, represents a valuable tool for agricultural monitoring. However, the conversion of images into regional-scale map products remains a field of research for many applications. This thesis presents original methodological developments in a semi-automatic multiscale approach based on the processing and analysis of optical satellite imagery for the mapping and characterization of agricultural systems at regional scale. The approach is composed of two main methods: (i) regional stratification into landscape units and classification of these units to produce a map of agricultural land-use systems; (ii) field-level segmentation and unsupervised classification of the segments by a “landscape-clustering” method to produce a cropping systems’ map. The methods were developed on a region of intensive agriculture, the Brazilian state of Tocantins, where the cultivated area, as well as the main agricultural land-use systems and cropping systems were successfully mapped from an annual NDVI-MODIS time series and a mosaic of Landsat images. The reproducibility of the approach was then assessed in Burkina Faso, where landscapes are shaped by the smallholder agriculture. Only the cultivated area could be mapped with satisfactory results, highlighting the limitations of these methods and the current offer in satellite imagery given the challenging specificities of this type of agriculture for remote sensing. The resulting maps were assessed with ground-truth data and agricultural statistics, and compared to other existing maps. The results of this thesis show the potential of the new method of regional stratification into landscape units which, based on NDVI time series and combined to the unsupervised “landscape-clustering” classification method, contributes to significantly improve discrimination of crop types and agricultural practices, and allows representing the agricultural systems at different levels of spatial organization. The originality of the developed methods lies mainly in the simplicity of their implementation. They are almost exclusively based on satellite data and require little “expert” intervention and external data, which gives them a high reproducibility potential. Thereupon, this thesis contributes, with these novel methods, to the development of generic tools to support large-scale monitoring of agriculture and to provide decision-support products for reasoned land-use planning
Document type :
Complete list of metadata

Cited literature [387 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Thursday, May 28, 2020 - 12:18:13 PM
Last modification on : Friday, August 5, 2022 - 2:38:11 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02639535, version 1


Beatriz Bellón de La Cruz. Une approche multiscalaire par télédétection pour la cartographie et la caractérisation des systèmes agricoles à l’échelle régionale. Traitement des images [eess.IV]. AgroParisTech, 2018. Français. ⟨NNT : 2018AGPT0002⟩. ⟨tel-02639535⟩



Record views


Files downloads