An Interface for Data Curation and Mapping of Irrigated Areas Using Active Learning

作者：Kin Wai Ng, Jack Marquez, Gabriel Laboy, Asfaw Kebede, Holly A. Michael, Kyle Frankel Davis, Michela Taufer · 年份：2025 · DOI：10.22541/essoar.176538448.89301468/v1 · 研究领域：Water resources management and optimization、Smart Agriculture and AI、Environmental Monitoring and Data Management

High-resolution irrigation mapping is critical for understanding agricultural water use, supporting food security, and informing sustainable land and water management policies. However, existing global irrigation maps are often too coarse, infrequently updated, and reliant on labor-intensive ground-truth data collection. While machine learning models have shown promise in classifying irrigated areas, their effectiveness is often limited by the scarcity of high-quality labeled data, especially in small-scale regions that are not typically captured by existing surveys or global datasets. We introduce an interactive labeling interface that integrates active learning with human feedback to accelerate the creation of high-quality irrigation datasets. Leveraging openly available environmental datasets from agencies such as the United States Geological Survey (USGS) and NASA, the interface enables scientists to define initial labeling criteria such as threshold values for variables known to correlate with irrigation (e.g., vegetation indices or land surface temperature) to generate a preliminary set of labeled data. After training an initial model on this data, the interface then presents scientists with a small subset of uncertain or high-impact predictions, which can be reviewed and relabeled using domain expertise. The updated labels are incorporated into the training dataset to iteratively retrain the machine learning model. This targeted feedback loop reduces manual annotation ...