<p>The empirical attribution of rapid hydrologic change presents a unique data availability challenge in terms of establishing baseline prior conditions. On the one hand, one cannot go back in time to collect the necessary in situ data if it were not serendipitously collected when the change was taking place. On the other hand, modern satellite monitoring missions are often too recent to capture changes that are ancient enough to provide sufficient observations for adequate statistical inference. In that context, the four decades of continuous global high resolution monitoring enabled by the Landsat missions are an unrivaled source of information to study hydrologic change globally. However, extracting the relevant time series information in a systematic way across Landsat missions remains a monumental challenge. Cloud masking and inconsistent image quality often complicate the automatized interpretation of optical imagery.</p> <p>Focusing on the monitoring of lake water extents, we address this challenge by coupling supervised and unsupervised image classification techniques. Unsupervised classification is first used to detect water on unmasked (cloudless and high quality) pixels. Classification results are then compiled across images to estimate the inundation frequency of each pixel, hinging on the assumption that different pixels will be masked at different times. Inundation frequency is then leveraged to infer the inundation status of masked pixels on individual images through supervised classification. Applied to a representative set of global and rapidly changing lakes, the approach successfully captured water extent fluctuations obtained from in situ gauges (when applicable), or from other Landsat missions during overlapping time periods.</p>