Video abstract for the HESS Highlight Paper "Learning Evaporative Fraction with Memory" (egusphere-2025-4082, Hydrology and Earth System Sciences). An explainable LSTM model trained on 90 eddy-covariance sites predicts evaporative fraction from routine meteorology alone, and its learned memory reveals how ecosystems integrate past climate: grasslands forget within days, while deep-rooted forests draw on conditions from months earlier. Memory timescales correlate with observed rooting depth across biomes.