Surface biophysical features fusion in remote sensing for improving land crop/cover classification accuracy

Area of Science:

• Remote Sensing
• Agricultural Science
• Geospatial Analysis

Background:

• Accurate land crop/cover maps are vital for global food security assessments.
• Existing classification methods can be limited by reliance on single data sources or features.

Purpose of the Study:

• To evaluate the impact of surface biophysical features on land crop/cover classification accuracy.
• To introduce and validate a novel fusion-based method for enhanced land crop/cover classification.

Main Methods:

• Utilized multi-temporal Sentinel 1 (VV, VH) and Sentinel 2 (NDVI, IBI, Wetness, Albedo, Brightness) imagery.
• Generated classification scenarios using individual and combined surface biophysical features.
• Employed Random Forest (RF) for classification and a voting strategy for decision-level fusion.

Main Results:

• Individual features yielded accuracies ranging from 57% (Brightness) to 68% (IBI).
• Feature-level fusion of biophysical features increased overall accuracy to 83%.
• Decision-level fusion of classification maps achieved a maximum overall accuracy of 89%.

Conclusions:

• Fusion of surface biophysical features significantly enhances land crop/cover classification accuracy.
• Combining multiple scenarios via a voting strategy further improves classification performance.
• The proposed fusion approach offers a more accurate method for land crop/cover mapping.