Uncovering optimal vegetation indices for estimating wetland plant species diversity

Abstract

Prior research on vegetation indices (VIs) to estimate species diversity in forest and grassland ecosystems has shown limitations when applied to wetland ecosystems due to their complex structure. Consequently, the predictive capacity of various VIs for wetland plant species diversity and their susceptibility to image noise remain largely unknown. To address these gaps, we utilized high-resolution multispectral images from Unmanned Aerial Vehicles (UAV) and field survey data in marsh wetlands. Various VIs and species diversity indices were computed, and univariate and multiple linear regression models were employed to assess predictive ability of the mean, standard deviation, and coefficient of variation of VIs for wetland plant species diversity. Results revealed that MTCI and NDREI exhibited the highest predictive ability for plant species diversity both before and after masking image noise. While most VIs generally improved in predictive ability after masking image noise, their susceptibility to it varied. MTCI, NDREI, SAVI, MSAVI, and MSAVI2 were less affected by image noise, with minimal changes in predictive ability before and after masking image noise. Conversely, CTVI, NDVI, and other VIs showed high susceptibility to image noise, with significant improvement in predictive ability after masking. ANOVA results indicated that integrating the mean and standard deviation of VIs into models significantly enhanced estimates of species diversity, highlighting their complementarity in predicting species diversity. These findings provide valuable insights into the predictive ability of different VIs for wetland plant species diversity, guiding selection of optimal VIs for predicting plant species diversity at broader scales in the future.

Keywords

• Vegetation indices
• Plant diversity
• UAV
• Image noise
• Wetlands