Toward a Scalable Agriculture Classification Model: Comparison of Remote Sensing Crop Classification Methods in the Red River Valley, Minnesota

Authors: Jesse Bakker*, University of Minnesota
Topics: Remote Sensing, Geographic Information Science and Systems
Keywords: remote sensing, land use classification, GIS, agriculture, crop classification, python, object-based image analysis
Session Type: Paper
Day: 4/4/2019
Start / End Time: 1:10 PM / 2:50 PM
Room: Roosevelt 5, Marriott, Exhibition Level
Presentation File: No File Uploaded

Remote sensing is a common tool in agriculture for crop classification and monitoring. For the most accurate results, the methods employed in classification are often fine-tuned to the agricultural conditions within the local geographic context of interest. This fine-tuning, however, can make it difficult to implement the same models in other locations to the same degree of accuracy. While these locally-tuned examples provide valuable insight into developing crop classification systems, a classification model that is applicable at larger spatial scales (e.g. national, global) requires a different approach. When scaling a model to such large spatial extents, other performance factors take on greater significance, including computational performance (the time it takes to run the model), robustness (the ability to identify a wide range of crop types), and accuracy (crop type identification and spatial bounds of fields). This paper develops metrics for measuring computational performance, robustness, and accuracy and tests them against multiple agricultural classification methods, including unsupervised, supervised, and object-based image analysis. It uses the Red River Valley in northwest Minnesota as a study area, a primarily agricultural region whose crop diversity provides a good test environment for the three performance measures. This paper uses DigitalGlobe and Sentinel-2 imagery taken during the 2017 growing season and the contemporaneous USDA Cropland Data Layer as a crop classification reference. Comparison of the multiple classification methods reveals some benefits as well as trade-offs in these three metrics for potential use in a scalable, geographically broad classification model.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login