DESIS Hyperspectral Satellite Data for Cropping Pattern Classification

Cropping patterns, including intercropping, are recognized as sustainable agricultural practices in many parts of Africa. However, there is a lack of data and information regarding their extent and specific locations. This study aims to understand the dynamics of these cropping patterns using hyperspectral satellite data. We examined the spectral reflectance of maize and intercropped maize (imaize) fields using DESIS hyperspectral satellite data during the flowering growth phase for discrimination and classification. We used mean reflectance spectra, coefficient of variation, and the Mann–Whitney U-test to characterize the cropping patterns and identify optimal spectral bands. Principal component analysis (PCA) was used to reduce the dimensionality of the hyperspectral data, followed by RF classification. Our findings reveal that the optimal spectral bands for classification fell within the 730–1000 nm range, with the spectral band at 849.8 nm being the most prominent. The RF algorithm performed well, achieving an overall accuracy of 80%. The maize class was distinguished with high precision (0.9) and recall (0.8), resulting in an F1-score of 0.9, indicating a robust ability to accurately identify and classify most of the maize class. In comparison, the imaize class exhibited lower precision (0.6) but a reasonable recall (0.7), leading to an F1-score of 0.6. These findings highlight the potential of using DESIS hyperspectral satellite data in conjunction with PCA and RF for the classification of maize cropping patterns. Additionally, our findings suggest avenues for further research using full waveform hyperspectral satellite data, such as EnMAP and PRISMA, to enhance our understanding of spectral dynamics among different cropping patterns.