Comparison of vegetation indices measured with proximal reflectance sensing to assess leaf N content and estimate crop yield in vegetable crops

Accurate assessment of leaf nitrogen (N) content and crop yield is essential for optimizing N fertilization strategies and improving agricultural productivity. Proximal reflectance sensing provides a non-destructive method for monitoring crop N status using vegetation indices derived from different spectral bands. However, models calibrated for specific cultivars may have limited applicability to other cultivars within the same species, highlighting the need for across-cultivar models for a species. This study aimed to evaluate the performance of several vegetation indices focused on red (NDVI, RVI), green (GNDVI, GVI) and red edge (RENDVI, CI, CCCI, RENDVI/NDVI ratio, and MTCI) bands, obtained by proximal reflectance sensing, for assessing leaf N content and estimating crop yield in vegetable crops. Data were collected from three cultivars of sweet pepper and four cultivars of muskmelon, across four phenological stages: vegetative, flowering, early fruit growth, and harvest stages. The results demonstrated that the performance of the vegetation indices to estimate leaf N content was relatively low during the vegetative and flowering stages, with a significant improvement in model accuracy during the early fruit growth and harvest stages. Notably, NDVI and GNDVI exhibited superior estimation performance during the mid-development stages, as shown by high R2 values and low RMSE values. Therefore, phenological stage-specific calibration models are important for improving estimation accuracy. Similarly, integrating data from multiple cultivars may have slightly reduced model performance compared to studies focused on a single cultivar. However, these comprehensive studies including several cultivars of each species are recommended to improve the performance of species-specific estimation models when using proximal sensing in vegetable crops. In conclusion, our study demonstrated that the use of NDVI and GNDVI provided the most accurate assessment of leaf N and estimation of crop yield of the various vegetation indices examined. This has contributed to the development of across-cultivar models for different phenological stages for each species and provided a foundation for versatile, cultivar-independent models that will contribute to optimal crop N fertilization.