Differences in Phenological Estimation From Multi-Vegetation Indices Across the Yellow River Basin

Satellite-derived vegetation indices (VIs) are widely used to monitor land surface phenology (LSP). Plant Phenology Index (PPI) and kernel NDVI (kNDVI) have emerged as promising tools for improving accuracy of LSP estimation. However, differential phenological performance of these two indices relative to traditional ones remains unclear. Here, we evaluated consistency of PPI, kNDVI, and EVI derived from MODIS against solar-induced chlorophyll fluorescence (SIF) and flux tower GPP data, and investigated spatiotemporal changes in retrieving four phenological metrics across the Yellow River Basin (YRB). Results indicated that EVI exhibited a stronger overall correlation with SIF, whereas PPI achieved the best performance for GPP estimation (R ² = 0.76, RMSE = 1.31 g C m^-2 day^-1), particularly in snow-affected alpine regions. PPI also showed more balanced and stable late-season errors. Pronounced temporal discrepancies were detected in downturn date (DD) and recession date (RD) among three VIs. PPI demonstrated closer alignment with SIF and GPP during the autumn decline and consistently identified both the onset and termination of autumn senescence approximately 25-50 days earlier than kNDVI and EVI. These variations were particularly evident in DD, where kNDVI and EVI indicated delays (0.247 days year^-1 and 0.038 days year^-1, respectively), while PPI showed an advance (-0.120 days year^-1). In addition, widespread advances were observed in upturn date (UD) and stabilization date (SD), primarily in the transitional zones between forest and grassland sub-regions, along with significant delays in the RD across the west-central region of basin. Furthermore, 70% of the YRB exhibited an advancing trend in the DD of PPI, which contrasted distinctly with the trends observed for kNDVI and EVI. These findings suggest improved performance of PPI in detecting autumn senescence and provide valuable insights into the potential of various VIs for LSP retrievals.