Characterising maize and intercropped maize spectral signatures for cropping pattern classification

Intercropping - the planting of more than one crop in the same plot of land - is a prevalent agricultural management practice which can be used for risk reduction. Despite its widespread use, intercropping is not commonly reported in agricultural statistics, resulting to very limited spatially disaggregated information about its prevalence. Remote sensing-based approaches to detect and estimate the area of cropping patterns like intercropping require good understanding of the spectral response of (intercropped) crops at different crop growth phases. This study integrates field surveys, farmer interviews and temporal Sentinel-2 data from four crop growth phases and the post-harvest period of maize and intercropped maize (imaize). The goal is to identify the optimal crop growth phases, spectral regions and vegetation indices (VIs) that can accurately discriminate the two cropping patterns. We computed p-values for the spectral bands using Mann-Whitney U test and identified critical crop growth phases. Classification of maize and imaize cropping patterns was performed using random forest classifier. Our spectral analysis revealed effective discrimination between maize and imaize cropping patterns during the vegetative (in all spectral bands) and flowering-yield phases (in Blue, Green, Red, RE704, RE783, NIR833, NIR865). The most suitable VIs contained red-edge and near-infrared spectal bands. Utilizing spectral data and VIs from vegetative and flowering-yield phases, we achieved optimal discrimination during the vegetative phase (user's accuracy of 100 % and producer's accuracy of 100 %). However, accuracy decreased during the flowering yield phase (overall accuracy of 87 % for all spectral bands). The highest classification results using all spectral bands at the flowering yield phase resulted in 80 % producer's accuracy for maize and 100 % for imaize. This study illustrates the utility of temporal Sentinel-2 spectral data for identifying the critical crop growth phase, spectral regions and VIs for cropping patterns classification, particularly for intercropping.