Temporal Disaggregation of Seasonal Forecasting for Streamflow Simulation

Issuing seasonal outlooks of hydro-climatic variables (e.g., precipitation and temperature) has become the mission of such meteorological service agencies as the Climate Prediction Center, the European Centre for Medium-Range Weather Forecasts, and Taiwan's Central Weather Bureau. Most of their seasonal prediction products are available in the form of tercile (i.e., above-normal, near-normal, and below-normal) probabilities. Even though tercile probabilistic forecasts can possibly be translated into numerical values, without being disaggregated to daily (or finer) series, these monthly to seasonal forecasts cannot be used by a typical hydrological model to correctly simulate the rainfall-runoff and routing processes in a watershed. Popular temporal disaggregation schemes include historical analogues and weather generation tools. While the former resamples historical daily observations whose aggregated values bear the highest resemblance (e.g., shortest Euclidean distance) to the seasonal forecasts, the latter relies on the Markov Chain and Monte Carlo techniques to generate new data series preserving the same statistical characteristics as the observed data. This study aims to investigate the impacts of different temporal disaggregation schemes on seasonal streamflow forecasting. We developed a lumped, continuous hydrological model for a representative reservoir watershed in Taiwan for the above investigation. We found that none of the temporal disaggregation schemes can be effective without accounting for the spatial coherence of precipitation in a simultaneous fashion.