Skip to Main content Skip to Navigation
New interface

Etude de la végétation à partir de nouveaux capteurs satellitaires radar

Abstract : In this thesis, we focus on how SAR images can be used to study vegetation. Vegetation lies at the core of human lives by providing both food and economic resources as well as participating in regulating climate. Traditionally, vegetation is classified into three categories: fields, flooded pastures, and forests. We follow this classification in our study. To tackle the first two, we chose to explore rice (in Camargue, France) since rice fields are initially flooded pastures and turn to fields when more mature. We illustrate the last category with forests in Madagascar.The aim of the first part is to provide a better understanding of the capabilities of Sentinel-1 radar images for agricultural land cover mapping through the use of deep learning techniques. We revealed that even with classical machine learning approaches (K nearest neighbors, random forest and support vector machines), good performance classification could be achieved with F-measure/Accuracy greater than 86% and Kappa coefficient better than 0.82. We found that the results of the two deep recurrent neural network (RNN)-based classifiers clearly outperformed the classical approaches.In the second part, the objective is to study the capabilities of multitemporal radar images for rice height and dry biomass retrievals using Sentinel-1 data. To do this, we train Sentinel-1 data against ground measurements with classical machine learning techniques (Multiple Linear Regression (MLR), Support Vector Regression (SVR) and Random Forest (RF)) to estimate rice height and dry biomass. The study is carried out on the same multitemporal Sentinel-1 dataset in the first part. The error of rice height estimation was 16% (7.9 cm), whereas the biomass was 18% (162 g¢m¡2) (both with Random Forest method). Such results indicate that the highly qualified Sentinel-1 radar data could be well exploited for rice biomass and height retrieval and they could be used for operational tasks.Finally, reducing carbon emissions from deforestation and degradation (REDD) requires detailed insight into how the forest biomass is measured and distributed. Studies so far haveestimated forest biomass stocks using rough assumptions and unreliable data. We aim to improve on previous approaches by using radar satellite ALOS PALSAR (25-m resolution) and optical Landsat-derived tree cover (30-m resolution) observations to estimate forest biomass stocks in Madagascar, for the years 2007-2010. The radar signal and in situ biomass were highly correlated (R2 = 0.71) and the root mean square error was 30% (for biomass ranging from 0 to 500 t/ha). Combining radar signal with optical tree cover data appears to be a promising approach for using by L-band SAR to map forest biomass (and hence carbon) over broad geographical scales.
Document type :
Complete list of metadata

Cited literature [287 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Monday, May 18, 2020 - 3:09:19 PM
Last modification on : Thursday, November 17, 2022 - 5:22:34 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02611643, version 2


Emile Ndikumana. Etude de la végétation à partir de nouveaux capteurs satellitaires radar. Traitement des images [eess.IV]. AgroParisTech, 2018. Français. ⟨NNT : 2018AGPT0010⟩. ⟨tel-02611643v2⟩



Record views


Files downloads