Auditing observation bias in a multidecadal Landsat series of open-water decline in Peru's coastal wetlands, 1990-2025

Peru has not completed a national wetland inventory, and its coastal wetlands have mostly been assessed at individual sites over short windows. This study built a harmonised inventory of 72 polygons in 21 coastal wetland complexes from two national cartographic products, and derived a satellite extent series from Landsat 5, 7, 8 and 9 surface reflectance on Google Earth Engine, 1990-2025. Two classification designs, an annual median composite and an annual per-scene presence frequency, were discarded after an audit showed their vegetation classes remained correlated with the number of clear observations available, a structural feature of the growing Landsat archive rather than a real change on the ground. A design pooling scenes into six 6-year epochs was adopted instead, and only open water passed that audit, and is used as the primary metric. National open water extent inside a 500 m buffer around every complex declined from 9,916 ha in 1990-1995 to 8,617 ha in 2020-2025 (-13.1 percent, Mann-Kendall p = 0.017), with 9 of 19 complexes declining significantly after a Benjamini-Hochberg correction for multiple comparisons; the decline was directionally consistent across a sensitivity analysis of the open-water threshold. Granted agricultural water-use rights within 5, 10 and 20 km of each complex showed no significant association with the open-water trend (Spearman rho = -0.23 at 10 km, p = 0.35). Five of eleven coastal departments have no coastal wetland cartography in either product. Formal agricultural water rights, at the resolution available, do not by themselves explain the observed decline.