2026-09-20T20:53:53Zhttps://api.drum.lib.umd.edu/server/oai/request

oai:drum.lib.umd.edu:1903/354802026-07-01T07:50:39Zcom_1903_2242com_1903_8com_1903_2col_1903_2773col_1903_3

Song, Xiaopeng Li, Haijun Digital Repository at the University of Maryland University of Maryland (College Park, Md.) Geography 2026-07-01T05:47:13Z 2026 https://doi.org/10.13016/fchf-2hgi http://hdl.handle.net/1903/35480 The growing global population has intensified the need to increase agricultural production while minimizing environmental impacts for a sustainable future. Meeting these pressing challenges fundamentally depends on accurate, spatially explicit information on crop distribution. This dissertation advances national-scale crop type mapping in large countries by integrating remote sensing, sample-based field surveys, and machine learning to support both long-term and near-real-time agricultural monitoring. Three studies collectively contribute to methodological innovations in high-resolution crop mapping across diverse agricultural systems. First, I improved an existing crop mapping workflow that integrates field surveys with satellite-based classification, demonstrating its feasibility for generating the first openly available, national-scale 10-m maize and soybean maps for smallholder agriculture in China in 2019. Second, I further enhanced the workflow for industrial agriculture and produced annual 10-m maize and soybean maps across the Contiguous United States (CONUS) from 2019 to 2022. By comparing these maps to the widely used 30-m products, I quantified the advantages of higher-resolution crop mapping, showing that 10-m maps reduced 30-m mixed pixels by approximately 8% for maize and 9% for soybean across counties representing 99.9% of national cultivation. Finally, I evaluated the potential of progressive within-season crop mapping using Sentinel-2 time series and historical field data, and examined the earliest feasible date and phenological stage for accurate crop identification across the CONUS. Results show that, without current-year field labels, at least 50% of counties accounting for 79% of national cultivation could achieve 90% accurate identification of maize and soybean by July 29 and August 8, respectively. The identified spatial variability in the earliest reliable mapping timelines provides valuable guidance for region-specific map development to support timely crop monitoring and enhance national food security. Together, this dissertation contributes methodological advances for simultaneously obtaining unbiased crop area estimates and wall-to-wall crop maps in both smallholder and industrial agricultural contexts. It demonstrates the capability of remote sensing-based approaches to support spatially explicit, accurate, and timely crop monitoring from regional to national scales. en ADVANCING NATIONAL-SCALE HIGH-RESOLUTION CROP MAPPING USING REMOTE SENSING Dissertation Geography Agriculture Environmental science 10-m resolution Agricultural Monitoring Cropping Mapping Maize and Soybean Remote Sensing Satellite Observations embargo