"horizon" is a self-contained R package designed for analyzing the way in which dam and reservoir operators apply streamflow forecasts to guide water release decisions. For a given dam or reservoir, time series inputs (daily storage and release and/or inflow) are used to infer a"forecast release signature", which specifies the likely lead time of forecasts used throughout the water year. The model infers forecast use solely from the observed operational data; it does not require data describing actual forecasts applied in practice. The model is designed to work with a large number of reservoir datasets, and features multiple original functions for data pre-processing (gap infilling, conversion to water years, etc.), post-processing, and visualization.