Özet:

Determining cropping patterns is crucial for quantifying irrigation water requirements at a catchment scale. For this reason, new and innovative technologies such as remote sensing (RS) and artificial neural networks (ANNs) are robust tools for generating the spatiotemporal variation of crops. In line with this, this study aims to classify each crop type using the ANN algorithm and calculate crop evapotranspiration (ETc). This study was conducted in the Akarsu Irrigation District (9495 ha) in the Lower Seyhan Plain in southeastern Turkey in the 2021 hydrological year. Crop types were classified using the ANN algorithm in the Environment for Visualizing Images (ENVI) program based on combined data from Sentinel-2 images with a 10-m resolution and ground truth data collected during the winter and summer seasons. The image analysis results demonstrated that bare soil and citrus made up 3666 ha and 3742 ha respectively in the winter season, while first crop corn (1586 ha) and citrus (4121 ha) were preponderant in summer. The confusion matrix of the ANN algorithm showed high agreement (wheat 89.76%, onion 91.67%; citrus 97.67% in winter and 98.98% in summer; 100% for lettuce, potato, sesame-2, palm, and watermelon) and medium agreement (fruit 58.33% in winter, 42.86% in summer) with ground truth data in growing seasons. Furthermore, the agreement was more than 80% for the first and second crops (cotton, soybean, peanut, and corn) in the summer season. Annual reference evapotranspiration and ETc were around 1308 mm and 890 mm, respectively. The ETc values for wheat, citrus, first-crop corn, and second-crop soybean were found to be consistent with previous studies of direct evapotranspiration methods conducted in the Cukurova region. Overall, RS and ANNs can be used to classify crop types accurately in the growing season. This study builds upon and expands the application of RS and ANNs in large-scale irrigation schemes.

Anahtar Kelimeler: