Akademik Veri Yönetim Sistemi
Potato Research, cilt.69, sa.1, 2026 (SCI-Expanded, Scopus)
Accurate mapping of crop types is essential for agricultural monitoring, resource management, and food security. This study evaluates the performance of two ensemble machine learning algorithms—random forest (RF) and gradient tree boosting (GTB)—for classifying potato fields using multispectral satellite imagery from Landsat 8 and Sentinel-2 in the Konya Plain, Türkiye. The methodology involved generating median composite images from the 2020 growing season (June–August), followed by feature extraction from training samples collected via ground truth, satellite, and synthetically generated data. Model performances were assessed using overall accuracy, kappa coefficient, F-score, and user’s and producer’s accuracy metrics. Five-fold cross-validation was employed to evaluate model generalizability. A sensitivity analysis was conducted on the number of trees, and 100 was selected for model training. Results indicate that the Sentinel-2 random forest model achieved the highest average overall accuracy (0.94) and kappa coefficient (0.86) across folds, demonstrating robust and consistent classification. Feature importance analysis showed that red-edge and SWIR bands were the most effective for model performance. Sentinel-2 imagery combined with random forest provided a reliable and efficient approach for potato classification in arid agricultural regions. These findings support the integration of remote sensing and machine learning for operational agricultural monitoring and suggest avenues for future research involving temporal analysis and multi-sensor data fusion. This approach can be extended to other crop types and regions, enhancing the use of satellite-based crop mapping in precision agriculture. It enables policymakers and agricultural agencies to implement timely, data-driven strategies for agricultural planning and food security.