Classification of Maize and Intercropped Maize Using Sentinel-2 Satellite Imagery

Accurate crop type maps from satellite data are crucial for evaluating crop growth and yields. However, in many parts of Africa, especially where intercropping is common, this information is lacking. This study introduces a mapping method using freely available Sentinel-2 (S2) multi-temporal imagery to outline maize and intercropped maize in Busia, Kenya. It specifically focuses on identifying cropping patterns, including intercropping, and examines how dynamic management practices, diverse crop varieties, and varying climates affect the sensor's response. Random Forest classification was utilized for three scenarios: (i) using all spectral bands, (ii) significant spectral bands, and (iii) red-edge vegetation indices (VIs). The best results were achieved using all spectral bands, with 100% user accuracy (UA) and 80% producer accuracy (PA) for maize, while imaize showed 69% user accuracy and 100% producer accuracy. The overall accuracy (OA) was 86%, with a Kc value of 0.71. In contrast, the utilization of VIs resulted in the lowest accuracy. This research contributes to the limited remote sensing studies on intercropping and the oversight of intercropping as a distinct class in crop mapping. It further demonstrates the applicability of Sentinel-2 for mapping cropping patterns, particularly those involving intercropping.