Lake level forecasting for hydro power operation in the Clutha catchment
Citation
Export citationWaters, A. (2019). Lake level forecasting for hydro power operation in the Clutha catchment (Thesis, Master of Science (MSc)). The University of Waikato, Hamilton, New Zealand. Retrieved from https://hdl.handle.net/10289/12622
Permanent Research Commons link: https://hdl.handle.net/10289/12622
Abstract
The Clutha River is New Zealand’s largest river, located within New Zealand’s largest catchment. The river runs through two hydro-power stations, the Clyde Dam and downstream the Roxburgh Dam, together providing around 10% of New Zealand’s electrical energy. Over 80% of the river flow at Clyde Dam is sourced from the outflow of three large lakes further up in the catchment, Lake Wanaka, Lake Wakatipu and Lake Hāwea. To efficiently operate the hydro-scheme the operator, Contact Energy, requires hourly forecast of the inflows into the head-pond lake of the Clyde Dam, Lake Dunstan. The current model is subject to some degree of predictive error and it has been identified that improving the accuracy of the forecast levels of Lakes Wanaka and Wakatipu may reduce the error of the inflow model.
This project develops an empirical model for estimating the lake level changes, with a lead-time from 1 hour out to 48 hours, in Lakes Wanaka and Wakatipu, using only a few independent variables. The empirical model was developed, calibrated, and first evaluated using recorded rainfall and temperature data (hindcast validation). The model was then revaluated using forecast rainfall and temperature data to gain an understanding of how the model will perform during operation (forecast validation).
The empirical model has two main components, the first component estimates the lake level recession rates as a function of current lake level. The second component estimates the lake rise resulting from recent rainfall in the catchment, the model uses an inverse Gaussian distribution to represent the lake level rise hydrograph form resulting from recent effective rainfall in the catchment. Average catchment temperature is used as a proxy for estimating the proportion of rainfall contributing to lake level rise.
During the hindcast validation the model performed reasonably well, closely estimating the time distribution of positive lake level changes and lake level recessions. For Lakes Wanaka and Wakatipu, the Nash-Sutcliffe efficiency index values for each season, at lead-time of 48 hours, had minimal change from the calibration index values, apart from in spring where the values notably dropped.
To perform the forecast validation, the forecast rainfall was first correlated to recorded rainfall, for which the model was calibrated. The use of weather-based forecasts of rainfall and temperature introduced a significant amount of error. While most of the time the forecast rainfall accurately predicted the presence of rainfall, the forecast rainfall depth was subject to a large error. This transferred through to the model output, with the model largely accurately predicting if the lake level will rise or fall but lost accuracy predicting the magnitude of lake level rises.
For the forecast validation the Nash-Sutcliffe efficiency index values for each season, at a lead-time of 48 hours, all dropped, apart from winter, when compared to the hindcast calibration fits. Again, spring performed the worst for both lakes. The model tended to underpredict most lake levels rises, with the extent of underprediction increasing with increasing predicted lake level rise magnitudes. The quality of the forecast validation of the model is subject to a level of uncertainty itself. This is because the time period of the validation had low rainfall, with relatively few lake level rise events.
The model was shown to be limited by the assumption being made by the model that rain gauge rainfall is representative of the spatially average catchment rainfall. This is most often not the case and leads to errors in the model outputs. It may be possible to improve the predictive ability of the model by sourcing rainfall data more representative of the spatially average catchment rainfall. The model will also likely be improved by performing a recalibration using forecast data, this is highly recommended once more data becomes available.
While not perfect, as with any hydrological model, the model has shown to be able to reasonably forecast lake level changes in an operational situation, when considering the uncertainties associated with forecast rainfall and rain gauge data.
Date
2019Type
Degree Name
Supervisors
Publisher
The University of Waikato
Rights
All items in Research Commons are provided for private study and research purposes and are protected by copyright with all rights reserved unless otherwise indicated.
Collections
- Masters Degree Theses [2435]