Scholarlink

<p>This study aims to model sugarcane yield at field scale using time-series images of the Landsat satellite before the harvest season in the Shoeibieh region in the Khuzestan province of Iran. Sugarcane fields from 2004 to 2017 were considered as different samples. Accordingly, a dataset of 4767 of the observed sugarcane yield from an area of ​​about 20 thousand hectares was used. In addition to conventional vegetation indices, new indices were defined based on new spectral equations. The Landsat satellite images were utilized to derive the vegetation indices, which were then converted into a weekly time series. This resulted in 29 different time-series vegetation indices data for every field, covering 2004 to 2017. A total of 30 weekly vegetation index values were obtained from each time series. Based on these weekly vegetation indices, the sugarcane yield was predicted using an artificial neural network (ANN). The results revealed that all vegetation indices have a correlation coefficient (r) with sugarcane yield between 0.54 and 0.85. The Mid Infrared Non-Linear Index (M2NLI) vegetation index with <i>r</i> = 0.81, MAPE = 12.24%, and MAE = 8.89 t /ha has good power for sugarcane yield prediction. Also, the results revealed that the yield prediction using 29 vegetation indices has better accuracy with <i>r</i> = 0.85, MAPE = 9.92%, and MAE = 7.34 t /ha. For model evaluation, yield prediction was done for 505 sugarcane fields in 2017 with <i>r</i> = 0.76, MAPE = 10.52%, and MAE = 9.29 t /ha.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sugarcane Yield Estimation at Field Scale Using Time Series Data from LANDSAT 7

  • 期刊论文

  • 发表日期：

    2026年5月14日

查看全文
  • Mostafa Khosravirad,
  • Mahmoud Omid,
  • Fereydoun Sarmadian,
  • Soleiman Hosseinpour,
  • Hassan Ghasemi-Mobtaker

摘要

This study aims to model sugarcane yield at field scale using time-series images of the Landsat satellite before the harvest season in the Shoeibieh region in the Khuzestan province of Iran. Sugarcane fields from 2004 to 2017 were considered as different samples. Accordingly, a dataset of 4767 of the observed sugarcane yield from an area of ​​about 20 thousand hectares was used. In addition to conventional vegetation indices, new indices were defined based on new spectral equations. The Landsat satellite images were utilized to derive the vegetation indices, which were then converted into a weekly time series. This resulted in 29 different time-series vegetation indices data for every field, covering 2004 to 2017. A total of 30 weekly vegetation index values were obtained from each time series. Based on these weekly vegetation indices, the sugarcane yield was predicted using an artificial neural network (ANN). The results revealed that all vegetation indices have a correlation coefficient (r) with sugarcane yield between 0.54 and 0.85. The Mid Infrared Non-Linear Index (M2NLI) vegetation index with r = 0.81, MAPE = 12.24%, and MAE = 8.89 t /ha has good power for sugarcane yield prediction. Also, the results revealed that the yield prediction using 29 vegetation indices has better accuracy with r = 0.85, MAPE = 9.92%, and MAE = 7.34 t /ha. For model evaluation, yield prediction was done for 505 sugarcane fields in 2017 with r = 0.76, MAPE = 10.52%, and MAE = 9.29 t /ha.