Using geographical information technologies in the context of food security: Parcel-based determination of agricultural products in Manisa sample

Geographic information systems and remote sensing technologies make a significant contribution to agricultural production, which is the most basic link of food security, and to monitoring and management of agricultural areas. Many satellite images are superior to each other with their spatial, spectral and temporal characteristics for detecting and classifying agricultural products and creating plant patterns. Multispectral and high spectral resolution satellite images such as Sentinel-2 may be insufficient temporally and spatially. It is envisaged that sharpening the images of different satellites within the framework of certain rules increases the sensitivity of the target study. This study examines the fusion of Sentinel-2 and Planet Scope satellite images, which are actively used in monitoring agricultural fields today, and the use of the sharpened image in product classification. It is aimed to detect and classify Tomato, Corn, Cotton, Grape and Olive products in Manisa province, within the borders of the Gediz Basin, which meets approximately 10% of Turkey's agricultural product needs. Four classification algorithms were used in the study: SVM, XGBoost, RO and KNN. Sentinel-2, Planet Scope and fusion images were classified separately using the same database, classification indexes and machine learning algorithms. Classification was made on a parcel basis and ÇKS records were used as terrestrial data. The fusion image gave better results than other images. SVM gave the highest accuracy results with 87.05% overall accuracy and 0.82 kappa value, and XgBoost algorithms with 85.09% overall accuracy and 0.80 kappa value, and these results are statistically significant.