Open Cloud Infrastructure for Geospatial Data Management and Analytics

Publikation: Bog/antologi/afhandling/rapportPh.d.-afhandlingForskning

The agricultural domain is becoming increasingly data-centric, as new technologies emerge to increase the yield while minimizing the environmental impact. These technologies have had a low adoption rate, partly due to proprietary software and formats. In this context, the recent advances in open source software, open data, and standardized formats have massive potential to accelerate the adoption of precision agriculture. This dissertation investigates the steps required for making the relevant technologies converge to a combined infrastructure, while keeping the cost to a minimum. Throughout the dissertation, emphasis will be on the infrastructure to be domain-agnostic, though most use cases will focus on the agricultural sector.
The first part covers the open cloud infrastructure, and the data platform deployed on it. This covers the issues arising with regard to standardization and scalability, and how the recent developments have overcome the previous obstacles when employing precision agriculture. It was found that the standardization carried out by the Open Geospatial Consortium (OGC) leverages many of the previous compatibility issues, and that open satellite data from the National Aeronautics and Space Administration (NASA) and the European Space Agency (ESA) significantly reduce the price of employing precision agriculture. The recent advances in cloud computing alleviates many of the previous scalability obstacles, and the increased Internet bandwidth combined with advances in web applications can decrease the technical obstacles for farmers using the system.
The second part investigates the recent advances in data analytics, with main focus on open satellite data. Due to the open data policies from governmental agencies, alongside with NASA and ESA, we have unprecedented amounts of data readily available. Several use cases demonstrate how valuable information can be extracted from these data. First, we investigate how the aforementioned infrastructure overcomes the obstacles for employing data on a single field level. Second, we investigate how regional statistics can be performed on thousands of fields, which is then used to categorize fields according to intra-field heterogeneity. The motivation is to localize the most heterogeneous fields, to identify fields where the benefit of employing precision agriculture are maximized. This was later commercialized by a partner in the Future Cropping project. Third, we employ deep learning for semantic segmentation, thereby designing a state of the art cloud classification algorithm for satellite imagery. Finally, the dissertation covers how the open data provide us with new possibilities, both within the agricultural domain, but also in a variety of other domains, exemplified by classifying coronal mass ejections.
StatusUdgivet - 2019

Se relationer på Aarhus Universitet Citationsformater

ID: 135444849