Lin, Chenxi2023-09-192023-09-192023-05https://hdl.handle.net/11299/257108University of Minnesota Ph.D. dissertation. May 2023. Major: Bioproducts/Biosystems Science Engineering and Management. Advisor: Zhenong Jin. 1 computer file (PDF); xix, 159 pages.To feed a growing world population of 9.7 billion by 2050, doubling global food production is necessary. However, the environmental consequences of land clearing are uncertain. A better solution is to increase crop productivity by closely monitoring current croplands through mapping their distribution, extent, and crop types. Over the last 20 years, remote sensing has received much attention in cropland mapping, covering various algorithms, features, and scales. Simple machine learning algorithms, such as maximum likelihood estimation, have evolved into advanced deep learning algorithms. Despite improving classification performance still being important, recent studies emphasize solving practical problems in cropland mapping. Three frontiers were identified: cropland mapping in label-scarce scenarios, mapping of perennial tree crops, and developing new data fusion algorithms to improve spatial, temporal, and spectral resolution. Researchers face limited ground truth data not only in underdeveloped regions, but also in resource-abundant countries like the US. Mapping perennial tree crops is receiving increasing attention, while data fusion algorithms are still being developed. These frontiers reflect the changing landscape of current and future cropland mapping missions. The goal of this dissertation was to tackle obstacles in cropland mapping through three distinct studies. These studies aimed to address limitations in labels and mapping of large-scale tree crops. The first study focused on mapping various crop types in different parts of the world during the early season, when labels are not yet available. This was achieved through the development of a novel approach that transfers topology relationships between crops. The second and third studies both centered around mapping olive trees on a large scale. The second study examined fundamental questions in large-scale tree crop mapping, such as data selection and model generalizability, in pilot sites in Morocco. The third study involved mapping individual olive trees at a sub-national scale in northern Morocco. Overall, the findings of these studies offer new insights into how to approach and overcome these challenges in cropland mapping.enThesis or Dissertation