Predicting Winter Wheat Yield using Landsat 8-9 Based Vegetation Indices in Semi-Arid Regions

This study investigated the prediction of winter wheat yield in cultivation regions of Kumkale (Batakovası) Plain in Çanakkale Province, Türkiye, utilizing Landsat 8-9 imagery-based Vegetation Indices (VIs) along- side Machine Learning (ML) methodologies. The VIs dataset was created by calculating images collected during the 2022 and 2023 growth seasons. The resulting dataset was employed in a C4.5 Decision Tree (DT) algorithm to predict winter wheat yield. The findings indicated that winter wheat yield could be predicted in April for fields classified as ‘Low Yield, ’Medium Yield,’ and ‘High Yield’ utilizing all indices except for Enhanced Vegetation Index (EVI) and Soil Adjusted Vegetation Index (SAVI). Interestingly, High Yield’ fields could also be predicted in March using the EVI index and in February using the SAVI index. In the winter wheat yield estimation, NDVI with a performance rate of 97.5% was able to determine \"High Yield,\" \"Medium Yield,\" and \"Low Yield\" in April (heading-blooming), while the lowest performance was with EVI at 77.50%, determining \"High Yield” in April (heading-blooming), \"Medium Yield\" (tilling-jointing) in February, and \"Low Yield”. (tilling-jointing) in March. The study concluded that winter wheat yields can be predicted using VIs independently of climate data. Future research will concentrate on assessing yield predictions for additional crops by employing various ML algorithms alongside climate data and VIs derived from higher-resolution satellite imagery.