Assessment of red-edge based vegetation indices for crop yield prediction at the field scale across large regions in Australia
Dhahi Al-Shammari, Brett Whelan, Chen Wang, R. G. V. Bramley, Thomas F. A. Bishop
Abstract
Vegetation indices have long been used to monitor vegetation using spectral information. The red-edge (RE) bands have gained attention for improved yield prediction capabilities over traditional red/near-infrared-based indices. This study introduces the triple red-edge index (TREI), a novel vegetation index that leverages the three RE bands provided by the Sentinel-2 satellite. It aims to enhance the accuracy of crop yield predictions. The TREI exploits changes in the transition slope between the red slope, influenced by photosynthesis , and the near-infrared (NIR) slope, affected by cell structure and leaf layers. It was evaluated against indices utilising none, one, two, or three RE bands for yield prediction efficacy. The study also incorporates a simple model combining weather and remote sensing data to predict crop yields, testing the approach across 168 canola and 123 wheat fields. The validation results demonstrated that the TREI significantly improves crop yield predictions by incorporating all three RE bands and effectively describing the RE region. The TREI yielded the highest concordance correlation coefficient (CCC) values for both canola (CCC = 0.89) and wheat (CCC = 0.85) crops, indicating their effectiveness in crop yield prediction. The study concludes that the TREI index outperforms existing vegetation indices in predicting crop yield due to using the Sentinel-2 three RE bands. The highest CCC values corresponded to using the TREI index in the crop yield prediction with a CCC = 0.89 (canola) and CCC = 0.85 (wheat) according to the validation results.