25 Sep 2025

From Analog to Digital: A Systematic Workflow for Converting Published Landform Maps to Georeferenced Datasets

Abstract: Reference datasets for geomorphological analysis often require the integration of multiple data sources, including legacy maps and published figures that exist only as scanned images or hard copies. This report documents a systematic five-step workflow for converting landform information from these analog sources into georeferenced point datasets suitable for digital analysis. The methodology encompasses acquiring and evaluating imagery, georeferencing using ground control points, manually digitizing landform polygons, converting to centroid points using a systematic grid-based approach, and assigning attributes with quality control measures. In a case study on East Asia, we demonstrate the workflow’s practical application by processing 15 published sources to generate over 2 million labeled landform points representing approximately 1,015 km² of land across China and Mongolia. The dataset encompasses seven landform classes commonly found in arid environments: active washes, alluvial fans, bedrock, pediments, playas, sand dunes, and sand sheets. Quality assessments using analyst confidence ratings revealed reliable classification performance for most landform types. This workflow provides researchers with an efficient approach to leveraging existing published landform data, thus expanding the spatial coverage and temporal depth of reference datasets that are available for geomorphological analysis and machine learning applications.

16 Aug 2021

Reproducibility Assessment and Uncertainty Quantification in Subjective Dust Source Mapping

Abstract: Accurate dust-source characterizations are critical for effectively modeling dust storms. A previous study developed an approach to manually map dust plume-head point sources in a geographic information system (GIS) framework using Moderate Resolution Imaging Spectroradiometer (MODIS) imagery processed through dust-enhancement algorithms. With this technique, the location of a dust source is digitized and recorded if an analyst observes an unobscured plume head in the imagery. Because airborne dust must be sufficiently elevated for overland dust-enhancement algorithms to work, this technique may include up to 10 km in digitized dust-source location error due to downwind advection. However, the potential for error in this method due to analyst subjectivity has never been formally quantified. In this study, we evaluate a version of the methodology adapted to better enable reproducibility assessments amongst multiple analysts to determine the role of analyst subjectivity on recorded dust source location error. Four analysts individually mapped dust plumes in Southwest Asia and Northwest Africa using five years of MODIS imagery collected from 15 May to 31 August. A plume-source location is considered reproducible if the maximum distance between the analyst point-source markers for a single plume is ≤10 km. Results suggest analyst marker placement is reproducible; however, additional analyst subjectivity-induced error (7 km determined in this study) should be considered to fully characterize locational uncertainty. Additionally, most of the identified plume heads (> 90%) were not marked by all participating analysts, which indicates dust source maps generated using this technique may differ substantially between users.