Recent spikes in global food prices have emphasized how monitoring agriculture production at a national, regional and global scale is at the core of modern economic, geostrategic and humanitarian concerns. While crop growth models can be used to estimate potential yield based on the agro-meteorological growing conditions, their application at regional or continental scales for operational monitoring is hampered by uncertainties related to agricultural practices, nutrient availability, soil moisture content, pest management and landscape heterogeneity. Earth observation by satellite remote sensing has the potential to greatly reduce these uncertainties by providing synoptic information on crop status throughout the growing season and over large geographic extents. Despite remarkable advances in each field, there is still a major methodological gap between operational agricultural monitoring, crop growth modelling and remote sensing. This thesis contributes towards bridging this gap by establishing a framework to retrieve spatially and temporally consistent crop specific information from satellite imagery which can be coupled with crop growth models at regional scale. The spatial resolution requirements of the satellite imagery which are necessary for agricultural monitoring are first defined quantitatively based on the observed landscape. Green Area Index (GAI), a measure of the photosynthetic surface of the canopy, is then shown to be retrievable for a specific crop (wheat) all along its growing season from surface reflectance measurements by satellite using radiative transfer modelling. The method was then upscaled from field to regional level (using different instruments providing imagery with 20 m and 250 m pixels respectively) by controlling the adequacy between the observation footprint and the target fields. The extension over wider geographic areas (>10,000 km²) in highly fragmented agricultural landscapes for 6 growing seasons shows a high temporal and inter-annual consistency which outperforms currently available products. GAI time series are finally used to recalibrate a mechanistic crop growth model, thereby improving the estimation of yield. This large scale demonstration achieved in the framework of the GLOBAM project opens the path to take full advantage of the state-of-the-art in crop modelling coupled with the ever-growing availability of satellite remote sensing imagery for operational agriculture monitoring at a sub-continental scale.

English

Lucia Segura

Thesis Collection