Based on field data simultaneous with Landsat overpasses from six different dates, we developed a robust linear model to predict subpixel fractions of water cover. The model was applied to a time series of 174 Landsat TM and ETM+ images to reconstruct the flooding regime of a system of small temporary ponds and to study their spatio-temporal changes in a 23-year period. We tried to differentiate natural fluctuations from trends in hydrologic variables (i.e., hydroperiod shortening) that may threaten the preservation of the system. Although medium-resolution remote sensing data have rarely been applied to the monitoring of small-sized wetlands, this study evidences its utility to understand the hydrology of temporary ponds at a local scale. We show that the temporary ponds in Doñana National Park constitute a large and heterogeneous system with high intra and inter-annual variability. We also evidence that the conservation value of this ecosystem is threatened by the observed tendency to shorter annual hydroperiods in recent years, probably due to aquifer exploitation. This system of temporary ponds deserves special attention for the high density and heterogeneity of natural ponds, not common in Europe. For this reason, management decisions to avoid its destruction or degradation are critical.