Landsat Collection 1 (C1) surface reflectance (SR) products have demonstrated effectiveness in monitoring inland and coastal aquatic environments, but now they are thoroughly replaced by the Collection 2 (C2) SR. However, whether the reflectance quality of C2 surpasses that of C1 in global water bodies remains unclear, hindering a comprehensive understanding of its uncertainty in aquatic environmental applications. To this point, using 12,355 global in-situ hyperspectral reflectance samples, we developed a spectral classification and scoring method that encompasses 30 spectral classes and evaluates spectral quality across visible-to-near-infrared domain. This method was then applied to assess the spectral classes and scores of C1 and C2 SR from Landsat 5 and 8 across global lakes, rivers, and coastal waters. Results showed that for Landsat 5, differences between C1 and C2 were negligible for all three water types, with spectral classes all nearly identical and average score differences <3.38 %. Substantial discrepancies were observed for Landsat 8, where C1 exhibits smaller spectral classes and higher scores than C2, which frequently exhibited negative reflectance values across visible-to-near- infrared. Notably, the mean scores of Landsat 8 C1 for lakes, rivers, and coastal waters are respectively 8.99 %, 12.50 %, and 48.26 % higher than those of C2. These disparities are hypothesized to result from different at mospheric correction processing in Landsat 8 C2 and the environmental characteristics of water bodies. Our findings contradict USGS'S expectations that C2 would improve data quality, and provide valuable insights for the Landsat SR product providers and the community using these products in aquatic applications.