An insight into long-term continuity in global land surface phenology: A comparative analysis of MODIS and VIIRS products

Abstract

The Visible Infrared Imaging Radiometer Suite (VIIRS) Global Land Surface Phenology (GLSP) product (VNP22Q2 C2) has been operationally produced since 2013, and is designed to provide annual measurements of the phenologies of vegetated land surfaces at the global scale, succeeding the Moderate Resolution Imaging Spectroradiometer (MODIS) Land Cover Dynamics (LCD) product (MCD12Q2 C61), which was first produced in 2001. Although separate validations have been conducted locally to ensure the reliability and accuracy of the detected phenometrics for each product, a comprehensive understanding of the differences between these two operational products is important for downstream applications. Therefore, this study conducted critical analyses and cross-comparisons of the land surface phenology (LSP) products to ensure long-term continuity in the global phenological dynamics. Specifically, we compared five LSP products at 500 m spatial resolution: two NASA operational products of MCD12Q2 C61 and VNP22Q2 C2, as well as three other LSP products that were generated by applying the VNP22Q2 C2 algorithm to different time series from NOAA-20 VIIRS only, both SNPP and NOAA-20 VIIRS, and MODIS on Aqua and Terra. First, we cross-validated the five products at 500 m using a high-quality reference LSP dataset at 30 m that was generated by fusing the Harmonized Landsat and Sentinel-2 (HLS) observations with near-surface PhenoCam time series across diverse ecosystems in North America, Europe, and Japan. Second, cross-comparisons were conducted between the five LSP products at 500 m across 12 golden tiles. Third, the long-term comparability and continuity between the MCD12Q2 C61 and the VNP22Q2 C2 products were assessed globally. These comprehensive evaluations demonstrated that the VNP22Q2 C2 product provides overall global continuity for the MCD12Q2 C61 record. Compared to independent reference data from the HLS-PhenoCam LSP product, the four LSP products derived from the VNP22Q2 C2 algorithm produced highly comparable results with a mean absolute difference (MAD) of ∼ 11 days and mean systematic bias (MSB) of ∼ 7 days. The cross-comparisons indicated a strong agreement among the five 500 m LSP products in the 12 selected golden tiles with MADs < 7 days; however, their agreement with the MCD12Q2 C61 was slightly lower with MADs ∼ 9–10 days. The global-scale evidence of continuity between the MCD12Q2 C61 and VNP22Q2 C2 products was an absolute difference of < 15 days, except in arid/semiarid, tropical, and high-latitude ecosystems. Finally, it is suggested that (1) the continuity from MODIS to VIIRS LSP products would be enhanced if the same phenological detection algorithm was applied to the MODIS data, and (2) the quality of the VIIRS GLSP product could be improved by integrating data from multiple VIIRS sensors.