Performance Evaluation of UAVSAR and Simulated NISAR Data for Crop/Noncrop Classification Over Stoneville, MS

Area of Science:

• Remote Sensing
• Agricultural Monitoring
• Earth Observation

Background:

• Synthetic Aperture Radar (SAR) data offers high resolution for agricultural change detection.
• NASA's airborne Uninhabited Aerial Vehicle SAR (UAVSAR) platform collects data suitable for evaluating satellite algorithms.
• The upcoming NASA-ISRO SAR (NISAR) mission aims for global-scale cropland mapping.

Purpose of the Study:

• To evaluate the science algorithm for the NISAR Cropland Area product.
• To assess the algorithm's performance using UAVSAR and simulated NISAR data (mode 129A).
• To determine optimal coefficient of variation thresholds for crop/noncrop classification at various spatial resolutions.

Main Methods:

• Utilized UAVSAR and simulated NISAR mode 129A data for analysis.
• Employed the coefficient of variation (CV) method for crop/noncrop classification.
• Evaluated classification accuracy using overall accuracy, J-statistic, and Cohen's Kappa at 10, 30, and 100 m spatial resolutions.

Main Results:

• Most tested resolutions exceeded NISAR's mission accuracy requirement of 80%.
• UAVSAR 10 m data achieved the highest accuracy (85% overall accuracy, 0.62 J-statistic, 0.60 Kappa).
• The 129A product achieved 81% accuracy at 30 m and 80% at 100 m, but only 77% at 10 m. A literature-recommended CV threshold of 0.5 resulted in suboptimal 65% accuracy.

Conclusions:

• The NISAR Cropland Area product algorithm demonstrates strong potential for accurate global mapping.
• Optimal coefficient of variation thresholds for classification are dependent on spatial resolution, decreasing as resolution coarsens.
• Further refinement of CV thresholds is necessary for maximizing classification accuracy across different spatial scales.