Improving satellite data interoperability: Large-scale data integration and agricultural applications in the US Midwest

Accurate ecosystem monitoring is more needed than ever. Satellite remote sensing (RS) offers unique insights into this field of research. Scalability is perhaps the most irreplaceable advantage of satellite RS for earth system models. However, accompanied with the surge of satellite data availability is the increasing burden of large-scale data manipulation and preparation. The motivation of this study is to make satellite data more usable for earth system modelers. This work features an interdisciplinary approach that harnesses the powers of supercomputing, machine learning, as well as agroecosystem domain sciences. Specifically, this study aims at improving the interoperability of satellite data—meaning the ability of multiple data sources to provide a single analysis-ready product. This thesis is comprised of three projects. The first work introduces a novel approach to improve satellite cloud detection for vegetative landscapes; the second work presents a complete pipeline to derive reliable surface reflectance data from the PlanetScope CubeSat constellation; the third study introduces a method of real-time crop stress monitoring enabled by predicting the Green Chlorophyll Vegetation Index (GCVI) with a neural network. The algorithms and methods proposed in this thesis contribute to the scalability and interoperability of satellite data. The thesis also demonstrates that the improved scalability and interoperability of satellite data benefit the modeling of the earth, particularly agroecosystems