Updating Hydrologic Models for Improved Inflow Forecasts into Hydropower Reservoirs
MetadataShow full item record
The consumption of electricity is growing to record levels due to population growth and the global economic prosperity. The electricity generation projection for 2040 is that 25% of the global electricity generation will come from renewable resources of which 80% is attributed to hydro and wind power. Electricity generation during times of high demand is well recompensed by the energy markets since the emergence of electricity as a competitive commodity. This, along with maximizing the value of water resources is a good incentive for establishing efficient hydropower reservoir operation system. In this regard, hydrologic models enjoy wide application in forecasting inflows into hydropower reservoirs and hence play an important role in the operation and planning of hydropower systems. The inflow forecasts are yet far from being perfect due to the numerous uncertainty sources that confront hydrologic modelling. In this thesis, model updating procedures are developed and tested with the Elspot (a day-ahead market of the Nordic Pool) as a focus of attention. Furthermore, a skill to reduce the occurrence and accumulation of errors in the state of a system without modifying the structure and parameterization of the existing operational forecasting model was considered essential. Three updating procedures were developed and tested through application to the Krinsvatn catchment using the HBV model. Firstly a complementary modelling framework in which a data-driven model captures the structure the conceptual operational model may be missing was developed. Deterministic and probabilistic evaluations revealed that the procedure successfully improved the skill of an hourly forecasting system for a lead-time up to 17 hours. Secondly, a filter updating procedure that recursively updates the gain on the error forecasting scheme of the complementary modelling framework was developed. Application to the same catchment showed that the filter updating procedure extended the performance improvement of the complementary modelling framework to a lead-time of 20 hours. In order to address the limitations of these procedures that surfaced during applications to the study catchment, identifiability of the snow routine parameters and sensitivity of the HBV model to the temperature inputs were analysed using the GLUE and a pseudo-sensitivity assessment methods, respectively. The assessment revealed that the precipitation phase partitioning and snow melt/freeze processes were highly uncertain due to a combination of process parameterisation and input data quality. It further revealed the importance of model updating during the winter/spring season. On the basis of these findings a sequential temperature updating procedure was developed. The aim of this procedure was to correct the initial state of a forecasting system by correcting the temperature inputs. This procedure was able to identify the input time steps in the past and quantify the temperature corrections they needed. It resulted in a relative RMSE reduction of 21% for the spring season.