Creative Commons License

Narin O., Bayik C., Sekertekin A., Madenoglu S., Pinar M., ABDİKAN S., ...More

8th International Conference on GeoInformation Advances, GeoAdvances 2024, İstanbul, Turkey, 11 - 12 January 2024, vol.48, pp.267-273 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 48
  • Doi Number: 10.5194/isprs-archives-xlviii-4-w9-2024-267-2024
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.267-273
  • Keywords: Crop Height, Linear Regression, Random Forest, Remote Sensing, Sentinel-1, Wheat
  • Yıldız Technical University Affiliated: Yes


Wheat is one of the primary crop productions for half of the global production. It is the most exported cereal, which reached up to 40% according to the records of the Food and Agriculture Organization of the United Nations (FAO) in 2020. The relationship between the crop parameters and remote sensing tools has become essential for the temporal monitoring and estimations of crop growth. In this study, we investigate the relationship between crop height and Synthetic Aperture Radar (SAR) backscatter. Plant height was collected twice, in the early (13 April 2023) and late stages (21 June 2023) of wheat, for a total of 70 samples. Sentinel-1 SAR data was also attempted to be synchronized with ground measurements. For this purpose, 8 images were obtained, 4 in both ascending (8 and 15 April, 16 and 26 June) and descending (9 and 14 April, 20 and 25 June) directions. The basic steps of image pre-processing, calibration, filtering, and topography correction, were applied to each image. By evaluating the correlation coefficients between plant height and images, data with low correlations were excluded and the first three data, dated 8 April and 19 June, were used for plant height estimation. For prediction purposes, Linear Regression (LR) and Random Forest (RF) methods were evaluated with three different training data sets. The highest correlation and minimum error value were achieved by VH polarization with random forest (r = 0.868, RMSE = 5.377 cm) in the early stage and LR (r = 0.616, RMSE = 14.451 cm) in the late stage. In general, higher training (80%) and lower test data (20%) produced better results.