Identifying the optimal phenological period for discriminating fruit tree crops using multi-temporal Sentinel-2 data and Google Earth Engine in a heterogeneous subtropical horticultural region

Abstract

Accurate, timely monitoring of the spatial distribution of fruit tree crops is crucial for management and yield forecasting. Classification of fruit-tree crops in subtropical agriculture using single-date images is hindered by data unavailability caused by cloud cover. This research analysed growth conditions of fruit tree crops and co-existing land cover types in Levubu, South Africa, using multi-temporal Sentinel-2 Multispectral data and random forest (RF). Twelve-month MSI images were evaluated for mapping fruit-tree crops and the surrounding land cover types across the growing seasons. The RF classification algorithm used to distinguish the crop types was robust, with an overall accuracy of 84.89% and a kappa coefficient of 0.83%. Analysis of variance was used to assess if there was statistical significance in overall accuracies among the S2 monthly composites. The results showed distinct spectral differences between fruit trees and co-existing land cover types during different months. In April, there were differences observed during the harvesting and senescence of mango and macadamia nut crops. In May, there were differences observed during the senescence of macadamia nut, mango, and guava crops. In June and July, there were distinct spectral differences during the peak flowering of avocado, macadamia nut, and mango crops, as well as the fruiting stage of banana crops. The shortwave infrared bands significantly improved the discrimination of fruit trees, followed by the red-edge bands. This mapping approach can assist in characterising the spatial distribution of fruit trees and co-existing land cover types at the smallholder level. The results provide evidence-based information that can assist farm managers and horticulturists in making informed decision-making and management strategies for effective agricultural management and sustainability of local horticultural systems.