S2S rainfall forecast calibration in real-time for a dense network of hydropower catchments

The continuing improvement of seasonal rainfall outlooks means they are now skillful enough to predict inflows to hydropower schemes and help to anticipate operational decisions. Hydro Tasmania, Australia’s largest generator of hydropower, is working with CSIRO to develop a long-range inflows prediction system for its 6 hydroelectric schemes. The schemes complement each other and are operated as a single system. The inflows predictions will be generated with conceptual hydrological models calibrated to a dense network of 563 sub-catchments. In this study, we develop a real-time S2S rainfall forecast capability for this network with forecasts updated as frequently as daily. We seek to establish a forecast post-processing model that generates high-resolution, spatially correlated ensemble rainfall forecasts from coarse-resolution climate model forecasts. We compare lagged ensemble forecasts from the Bureau of Meteorology’s new ACCESS-S2 seasonal forecasting model with burst ensemble forecasts from ECMWF’s SEAS5 model, eliciting the value of simple versus complex post-processing methods for coarse-scale calibration of the ensemble climate forecast. A random selection from k nearest-neighbours provides a template for disaggregation of each ensemble member to the required spatial and temporal resolution. To understand the skill available to the real-time forecasting system, we evaluate the historical performance of the system from 1981-2018. Skilful forecasts can be obtained for the next month after which the ensemble time series ought to revert to an unbiased climatology forecast. The calibration method can be highly computationally efficient, allowing parameters to be re-estimated in the process of generating each new forecast, thereby updating parameters with the most up-to-date forecast and observation data available. Combined with an efficient hydrological model, reliable rainfall forecasts can add value to antecedent hydrological conditions to provide more skilful forecasts of inflows for the season ahead.