Abstract In this study, we have analyzed systematically the oxygen isotopic compositions of South China Sea (SCS) and Luzon Strait (LS) seawater so that a comprehensive interpretation of their temporary and spatial variability can be delineated. The oxygen isotopic compositions of the samples collected in the two areas were determined using the Epstein-Mayeda technique, and the overall precision of the d18OSMOW measurements is +0.1 ?. Our data suggest that LS seawater is a mixture of SCS and Kuroshio waters. The precipitation (1.5 mm) on SCS in April 1998 (during ORI517 cruise) is much less than that (169.5 mm) in April 1999 (during ORI546 cruise). As LS and SCS waters are concerned, the profiles of d18OSMOW generally share the same trend as those of salinity. The d18OSMOW values of the SCS surface water decrease toward the south, while those of the LS surface water decrease toward the west. These suggest that the effect of the Kuroshio water decreases are it moves from northeast toward southwest after its intrusion through LS. The salinity of Kuroshio water reaches the maximum at 34.92 with the corresponding d18OSMOW value of 0.42 ?, whereas the salinity of the representative SCS surface water is 33.34 and the corresponding d18OSMOW is -0.25 ?. As these two data were selected as end members, the estimated proportion of the intruding Kuroshio water in the composition of LS water could be high up to 80 % at 121.5o E. The plot of d18OSMOW versus salinity for cruises OR517 and 546 shows a difference between the slopes of the two regression lines, indicating the effect of variability in precipitation and the depth of the mixed layer. The average d18OSMOW value for surface water decreases toward the west as the water moves from Western Philippine Sea (WPS) to LS and SCS. The average d18OSMOW values for the surface, subsurface, intermediate, and deep waters are listed in the corresponding order in parentheses preceded by the studied area as follows: WPS (0.29 ?, 0.25 ?, -0.06 ?, -0.10 ?); LS (0.15 ?, 0.17 ?, -0.02 ?, -0.08 ?); SCS (-0.03 ?, 0.18 ?, -0.05 ?, -0.08 ?).

This study utilized, for the first time, the d18Osw as a tracer to investigate the seasonal variations of circulation in the Taiwan Strait (TS), which is the predominant sea passage with an average depth of 60 m connecting the East China Sea (ECS) and the South China Sea (SCS). The result shows that the circulation system in TS is mainly influenced by the inter-mixing among the China Coastal Water (CCW), the SCS water (SCSW), and the Kuroshio Water (KW). In spring, the KW dominates in TS, whereas the CCW is still observed in northwest TS. During the summer, SCSW replaces the KW and becomes the major water type in the TS, yet the KW is found to be restricted in the southwest part and the bottom of the TS. Due to the larger discharge from rivers (mainly the Yangtz River), the CCW has a more extensive distribution in the TS in summer than other seasons. In fall and winter, the CCW occupies the northern part of TS due to the stronger northeastern monsoon which limits the intrusion of the KW through the Luzon Strait to the northern TS. The two distinct water types inevitably form a front in the central TS. The hydrographic variations at Penghu Channel (PHC) were also explored in this study. The d18Osw indicates that the perennial intrusion of the KW into the PHC is varying throughout different seasons. This intrusion is found strongest in fall and winter. In summer, the upper layer of PHC is occupied chiefly by SCSW, while the KW remains at the bottom layer in PHC. By including an additional inflow of 0.5Sv from TS to ECS, this study further reconstructed a box model of the ECS, which was previously furnished by Lin(1999). The new estimates suggest that ~0.38*104 km3/year of the Kuroshio surface water (0-50m) and ~1.54*104 km3/year of the upwelled Kuroshio subsurface water (50-150m) are transported to the ECS, while ~3.83*104 km3/year of the ECS water are exported to the western Pacific Ocean.

L'acquisition de paramètres marins à partir des données spatiales de la couleur de l'eau nécessite l'élimination de la contribution de l'atmosphère au signal mesuré par le capteur. En effet, la majorité du rayonnement solaire mesuré par les instruments optiques dans les longueurs d'ondes qui intéressent la couleur de l'eau provient de la diffusion par les molécules de l'air et les aérosols atmosphériques. L'élimination de la contribution de l'atmosphère est appelée correction atmosphérique (CA). Pour les eaux claires, les méthodes de CA supposent une réflectance marine nulle dans le proche infra-rouge (PIR). Ceci permet d'estimer la réflectane de l'atmosphère et de l'extrapoler vers les bandes du visible, et donc de déterminer le signal marin qui contient les informations sur les propriétés optiques des eaux marines. Cette hypothèse n'est cependant pas vérifiée pour les eaux turbides, qui représentent la quasi totalité des eaux côtières. Par conséquent, de nombreux algorithmes de CA ont été développés pour les eaux côtières incluant des hypothèses alternatives. L'objectif de ce travail de thèse est de valider et d'améliorer ces méthodes de CA pour les images MODIS Aqua. Pour cela, diverses approches de CA développées pour les eaux cotières ont été comparées et validées : (1) l'algorithme standard de la NASA, (2) le "NIR Similary spectrum algorithm" qui inclut des hypothèes d'homogénéité spatiale des réflectances marines et atmosphériques, (3) l'algorithme qui utilise les bandes dans l'infrarouge moyen pour la CA dans les eaux très turbides, et (4) un algorithme utilisant un réseau de neurones artificiels. L'exercice de validation à partir de données in situ, et en fonction des types d'eaux, a permis d'identifier différentes pistes d'amélioration pour l'estimation du signal marin. L'un d'entre elles comprend l'utilisation de relations spectrales pour forcer les modèles de réflectances marines utilisés par les algorithmes CA pour estimer le signal marin dans le PIR. Des modifications ont été apportées aux modèles de réflectances marines de l'algorithme standard de la NASA et du "NIR Similarity spectrum algorithm". Chacun des modèles a été forcé avec des relations spectrales préalablement validées grâce à des données globales. Une étude de sensibilité et une validation de ces algorithmes modifiés à partir de données MODIS-Aqua dans la Manhe Orientale/Mer du Nord et la Guyane Française ont démontré que les modifications suggérées amélioraient les estimations du signal marin dans les eaux côtières optiquement complexes. / To acquire marine parameters from remote sensing ocean color data, the sensor-measured signal needs to be corrected for the atmospheric contribution. Indeed, the solar radiation reflected by air molecules and atmospheric aerosols is significant in the sensor bands of interest for ocean color applications. The removal of the atmospheric contribution is called the atmospheric correction (AC). In open ocean waters, the AC relies on the assumption that the water is totally absorbent in the near infrared (NIR) part of the spectral region, allowing to retrieve the atmospheric contribution and to extrapolate it to the visible spectral range, and thus to determine the marine signal that contains the information on the optical properties of seawaters. However, this assumption is not valid in highly productive and turbid coastal waters. Hence, AC approaches for coastal waters need to rely on alternative assumptions. This Ph. D. thesis has as main objective to validate and improve these AC methods developed for contrasted coastal waters, with a focus on MODIS Aqua images. First, a validation and comparison of existing AC methods, relying on diverse assumptions and methods, is performed. Therefore, four commonly used AC methods are selected, (1) the standard NIR AC approach of NASA, (2) the NIR similarity spectrum AC approach including assumptions of spatial homogeneity in the water and aerosol reflectance, (3) the switching algorithm using the short wave infrared bands for AC in highly turbid waters, (4) an Artificial Neural Network algorithm. With the help of a validation exercise based on in situ data and as a function of the water type, several areas of improvement are delineated, including the use of spectral relationships to constrain NIR-modelling schemes. Modified NIR-modelling schemes are suggested for the standard NASA and NIR similarity spectrum AC methods. Both are forced with globally valid spectral relationships. Sensitivity studies and validation exercises, using MODIS-Aqua images in the Eastern English Channel/North Sea and French Guiana waters, are conducted showing that the suggested modified NIR-modelling schemes improve the estimations of the marine signal in contrasted coastal waters.