Water Change

A probability-based sample analysis is the recommended good practice approach to estimating land cover and land use extent and change. The global surface water dynamics map enables a higher sampling efficiency through stratification at the sub-national, national, and global scales. The map should not be used as the only source data for national and international reporting due to unknown map uncertainty that may vary in space and between time intervals.

Ice cover, lava, dark vegetation cover, high turbidity or very shallow water, the combination of unmasked shadow or haze with one of these covers, and mixed pixels are the most common sources of error in mapped water and land in the Landsat imagery.

The dynamics map overestimates the extent of dynamic water presence, labeling some of the reference areas of permanent water or land as places with dynamic water presence. Much of this overestimation is within the above land covers or in the mixed pixels on the edge of water bodies. Conversely, the areas mapped as permanent water or land can be used with high confidence, committing only 1.2 (±0.7)% and 0.1 (±0.0)%, respectively.

Individual classes representing dynamic water presence should be used with care, having lower accuracies. However, most errors are either mixed pixels or real dynamics mislabeled as another type of dynamic water presence rather than as permanent land or water.