National Land Cover Database (NLCD) [2024 INV#WO0000000107887]
Description
Characterization of the physical land cover is critical for managing the lands, waters, and resources of the United States. The National Land Cover Database (NLCD) is easily one of the most widely used datasets produced within the Department of Interior, with applications including natural resource management, hydrology, biodiversity and habitat, energy and minerals, natural disasters, agricultural sustainability, and more. The criticality of these data demand 1) frequent updates, and 2) a characterization of how our landscapes are changing across time and space. Our linkage of three deep learning models, including a generative transformer-based AI model, has been vital for not only improving our characterization of the landscape with NLCD, but for enabling us to produce land cover products faster and more efficiently. The application of AI to this problem has saved the government valuable resources.
Detailed example
Maps of land cover change over time for the United States.
AI / analytics pattern
Classical/Predictive Machine Learning: Models trained on data to make predictions or classifications based on identified patterns or relationships.
Automation level / stage
c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.
Expected benefit
Our revamping of the NLCD methodology with a series of 3 linked deep learning models has allowed us to improve the product, reduce latency of delivery, and to produce that product more cheaply.
Controls / human review
ATO: No; PIA: Not published
Data needed
The National Land Cover Database is built on a foundation of Landsat data, with a mapping interval of 1985 to present. Evaluation of performance of NLCD algorithms depends upon a rigorous accuracy assessment that uses a stratified sampling approach, high-resolution geospatial data, and manual interpretation from land cover experts. High-resolution data is visualized in TimeSync software, sourced from Google Earth and associated commercial high-resolution data such as Maxar and Planet.