Tropical forests, which provide important ecosystem functions and services, are increasingly threatened by anthropogenic pressures. This has resulted in an urgent need to understand tree species diversity of those forests. Where knowledge of that diversity is largely from the botanical surveys and local ecological studies, data must inevitably be up-scaled from point observations to the landscape and regional level if a holistic perspective is required. This thesis explores aspects of the spatio-temporal heterogeneity of canopy reflectance patterns over the forests of French Guiana, in order to assess whether this information could help defining an ecologically consistent forest typology. To gain insight into both the spatial and temporal heterogeneity of French Guiana’s forests, instrumental artefacts affecting the satellite data first had to be addressed. Data used in this study represent the spectral response of forest canopies, and the way in which such data are captured makes them susceptible to the ‘bi-directional reflectance distribution function’ (BRDF). BRDF indicates that objects do not reflect light in equal proportions in all directions (isotropically). Thus, forest canopies will reflect light anisotropically depending on factors including canopy roughness, leaf optical properties and inclination, and the position of the sun relative to the sensor. The second chapter of this thesis examines how BRDF affects the canopy reflectance of forests in French Guiana, and how not correcting for BRDF affects spectral classifications of those forests. When monthly reflectance data corrected for the artefact are examined, these suggest seasonally-occurring changes in forest structure or spectral properties of French Guiana’s forests. The third chapter of this thesis thus examines temporal effects of BRDF, and used cross-regional comparisons and plot-level radiative transfer modelling to seek to understand the drivers of the monthly variation of the forests’ canopy reflectance. For the latter, the Discrete Anisotropic Radiative Transfer (DART) model was used along with aerial laser scanning (ALS) observations over different forest structures, indicating that the observed variation in reflectance (and derivatives known as vegetation indices) could not be explained by monthly variations in solar direction. At the regional scale, it was also demonstrated that forests in the Guiana Shield possess temporal variation distinct from forests in central Africa or northern Borneo, forests also lying just above the Equator. Had the observed temporal variation in vegetation indices been the result of BRDF, it would have been expected that the forests in the three zones would have similar patterns of variation, which they did not. Central African forests appear to have their greening synchronized with rainfall, whereas forests in the Guianas appear synchronized with the availability of solar radiation. Further analysis of the vegetation index time-series of observations also indicated that different types of forests in French Guiana possess distinct patterns of temporal variation, suggesting that tropical forest types can be discriminated on the basis of their respective “temporal signatures.” That was exploited in the fourth chapter of the thesis, which maps forests in French Guiana based on their combined spatio-temporal canopy reflectance patterns and by so doing presents a novel way of addressing forest typology, based on ecologically meaningful information. The thesis presented demonstrates that it is possible to adequately address remote sensing data artefacts to examine patterns of spatial and temporal variation in tropical forests. It has shown that phenological patterns of tropical rainforests can be deduced from remote sensing data, and that forest types can be mapped based on spatio-temporal canopy reflectance patterns. It is thus an important contribution to understand the ecology of tropical forests in French Guiana and to improve the toolbox of scientists dealing with the identification of spatio-temporal patterns observable in forests at the landscape level.

Dans le contexte des changements climatiques, l'objectif du travail effectué est de caractériser l'hétérogénéité du continent africain afin de mieux comprendre et quantifier les processus de surface agissant sur les flux hydriques. Ce travail s'inscrit dans le cadre de la mise à jour de la base de données ECOCLIMAP-I constituée d'une carte d'occupation des sols et des cartes de paramètres biophysiques. Pour cela, on s'appuie sur des données de télédétection acquises par les capteurs de dernière génération MODIS et SPOT/VEGETATION entre 2000 et 2007. Dans un premier temps, deux techniques de classification ont été développées afin de cartographier les différents écosystèmes. L'une, supervisée, a été conduite dans le contexte du programme AMMA afin de discriminer les écosystèmes sur la région ouest-africaine en combinant l'information complémentaire contenue dans les cartes d'occupation du sol GLC2000 et ECOCLIMAP-I par analyse supervisée de l'indice foliaire (LAI) MODIS. L'autre, non supervisée et hybride, utilise les principes de regroupement hiérarchique et dynamique de manière automatique en combinant l'usage du classificateur k-NN et celui de la transformée de Fourier Discrète sur la base des données d'indice de végétation normalisé (NDVI) SPOT/VEGETATION pour identifier les écosystèmes africains. Dans un deuxième temps, des méthodes d'estimation des paramètres biophysiques tels que l'albédo, la fraction de végétation, l'indice foliaire ont été développées et/ou appliquées sur le continent. Une approche statistique permet de déterminer la contribution du sol nu et de la végétation à la constitution de l'albédo de surface comme tel que cela est requis dans les modèles de surface. La méthode a d'abord été appliquée sur la région ouest-africaine et sa robustesse a été prouvée lors de son application à l'intégralité du continent africain. Ces conditions de surface ont ensuite été implémentées dans le modèle de surface ISBA pour reproduire les processus de surface. La sensibilité d'ISBA au forçage physiographique a été étudiée en analysant deux simulations avec le même forçage atmosphérique sur la région ouest-africaine : l'une en utilisant la classification ECOCLIMAP-I et l'autre en utilisant la nouvelle paramétrisation de la surface développée sur la région ouest-africaine. Les flux de chaleur latente et sensible sont principalement pilotés par la fraction de végétation. Le modèle ISBA peut être utilisé pour prédire l'impact d'un changement d'occupation du sol et par conséquent des actions anthropiques sur le bilan hydrique.

Tropical forests, which provide important ecosystem functions and services, are increasingly threatened by anthropogenic pressures. This has resulted in an urgent need to understand tree species diversity of those forests. Where knowledge of that diversity is largely from the botanical surveys and local ecological studies, data must inevitably be up-scaled from point observations to the landscape and regional level if a holistic perspective is required. This thesis explores aspects of the spatio-temporal heterogeneity of canopy reflectance patterns over the forests of French Guiana, in order to assess whether this information could help defining an ecologically consistent forest typology. To gain insight into both the spatial and temporal heterogeneity of French Guiana’s forests, instrumental artefacts affecting the satellite data first had to be addressed. Data used in this study represent the spectral response of forest canopies, and the way in which such data are captured makes them susceptible to the ‘bi-directional reflectance distribution function’ (BRDF). BRDF indicates that objects do not reflect light in equal proportions in all directions (isotropically). Thus, forest canopies will reflect light anisotropically depending on factors including canopy roughness, leaf optical properties and inclination, and the position of the sun relative to the sensor. The second chapter of this thesis examines how BRDF affects the canopy reflectance of forests in French Guiana, and how not correcting for BRDF affects spectral classifications of those forests. When monthly reflectance data corrected for the artefact are examined, these suggest seasonally-occurring changes in forest structure or spectral properties of French Guiana’s forests. The third chapter of this thesis thus examines temporal effects of BRDF, and used cross-regional comparisons and plot-level radiative transfer modelling to seek to understand the drivers of the monthly variation of the forests’ canopy reflectance. For the latter, the Discrete Anisotropic Radiative Transfer (DART) model was used along with aerial laser scanning (ALS) observations over different forest structures, indicating that the observed variation in reflectance (and derivatives known as vegetation indices) could not be explained by monthly variations in solar direction. At the regional scale, it was also demonstrated that forests in the Guiana Shield possess temporal variation distinct from forests in central Africa or northern Borneo, forests also lying just above the Equator. Had the observed temporal variation in vegetation indices been the result of BRDF, it would have been expected that the forests in the three zones would have similar patterns of variation, which they did not. Central African forests appear to have their greening synchronized with rainfall, whereas forests in the Guianas appear synchronized with the availability of solar radiation. Further analysis of the vegetation index time-series of observations also indicated that different types of forests in French Guiana possess distinct patterns of temporal variation, suggesting that tropical forest types can be discriminated on the basis of their respective “temporal signatures.” That was exploited in the fourth chapter of the thesis, which maps forests in French Guiana based on their combined spatio-temporal canopy reflectance patterns and by so doing presents a novel way of addressing forest typology, based on ecologically meaningful information. The thesis presented demonstrates that it is possible to adequately address remote sensing data artefacts to examine patterns of spatial and temporal variation in tropical forests. It has shown that phenological patterns of tropical rainforests can be deduced from remote sensing data, and that forest types can be mapped based on spatio-temporal canopy reflectance patterns. It is thus an important contribution to understand the ecology of tropical forests in French Guiana and to improve the toolbox of scientists dealing with the identification of spatio-temporal patterns observable in forests at the landscape level.