Machine Learning-Based Classification for Crop-Type Mapping Using the Fusion of High-Resolution Satellite Imagery in...

Area of Science:

• Agricultural Science
• Remote Sensing
• Environmental Monitoring

Background:

• Accurate crop-type identification is crucial for agricultural land management and yield prediction.
• Semiarid regions present challenges due to crop heterogeneity and complex land cover.
• Existing methods struggle with the intricate agricultural landscape of Morocco.

Purpose of the Study:

• To evaluate the effectiveness of multisensor data fusion for crop-type identification.
• To assess machine learning classifiers for mapping crop types in a heterogeneous semiarid zone.
• To enhance land use and land cover mapping in Moroccan agricultural areas.

Main Methods:

• Fusion of high spatiotemporal resolution Sentinel-1 (C-band SAR) and Sentinel-2 (optical) satellite imagery.
• Application of three machine learning classifiers: Artificial Neural Network (ANN), Support Vector Machine (SVM), and Maximum Likelihood (ML).
• Utilized in situ observations from 2018 for the R3 perimeter of Haouz plain, Morocco.

Main Results:

• The combined use of Sentinel-1 and Sentinel-2 data significantly improved crop-type classification accuracy.
• Achieved an overall accuracy of 89% and a Kappa coefficient of 0.85 with data fusion.
• Multisensor classification outperformed classifications based on optical or SAR data alone.

Conclusions:

• Multisensor data fusion is a highly effective strategy for crop-type identification in complex agricultural environments.
• Machine learning classifiers, when applied to fused data, provide robust crop mapping solutions.
• This approach offers a valuable tool for agricultural land management and precision farming in semiarid regions.