Improved observation of the global water cycle with satellite remote sensing and neural network modeling ; Amélioration de l'observation du cycle de l'eau à l'échelle globale grâce à la télédétection par satellite et à la modélisation par réseaux de neurones

Satellite remote sensing is commonly used to observe the hydrologic cycle at spatial scales ranging from river basins to the globe. Yet, it remains difficult to obtain a balanced water budget using remote sensing data, which highlights the errors and uncertainties in earth observation (EO) data. This research aimed to improve estimates of precipitation, evapotranspiration, runoff, and total water storage change at the global scale using a combination of analytical methods (optimal interpolation, OI) and statistical modeling methods including neural networks (NN). Models were trained on a set of 1,358~river basins and validated them on an independent set of 340~basins and in-situ observations of precipitation, evapotranspiration, and river discharge. The models are extended to make pixel-scale predictions in 0.5° grid cells for near-global coverage. Calibrated datasets result in lower water budget residuals in validation basins: the mean and standard deviation of the imbalance is 11±44 mm/mo when calculated with uncorrected EO data and 0.03±24 mm/mo after calibration by the NN models. The results allow us to make more accurate estimates of missing water cycle components, for example to estimate evapotranspiration in un-instrumented areas, or to predict discharge in ungaged basins. The results can also indicate to data producers where their products seem incoherent with other datasets and where enhanced calibration could lead to improvements. Finally, this research demonstrates the use of neural networks and machine learning for the integration of satellite data and for the study of the water cycle. ; La télédétection par satellite est couramment utilisée pour observer le cycle hydrologique à des échelles spatiales allant des bassins fluviaux au globe terrestre. Pourtant, il reste difficile d'obtenir un bilan hydrique équilibré en utilisant des données de télédétection, ce qui met en évidence les erreurs et les incertitudes des données d'observation de la Terre. Cette recherche visait à améliorer les estimations ...