Thumbnail Image

Forecasting Lake Waikaremoana Water Availability for Hydro Power

Lake Waikaremoana is a high-altitude, large-volume lake located within the rugged terrain of the Urewera National Park. At the lake outflow a rapid elevation change of nearly 450 metres in 8 kilometres facilitates the lake’s use as the upper intake portal for the Waikaremoana Hydro Power Scheme. At the time of this study, Genesis Energy operated the Waikaremoana Power Scheme in response to a water availability model based on daily lake level differencing from which daily generation capacity is predicted, allowing strategic bidding into the electricity market. However, when river flows are low this model is subject to error, as small changes in lake level sometimes cannot be determined accurately beyond background noise on daily timescales. This project develops a method of estimating both current day and day-ahead water availability of Lake Waikaremoana, independent of lake levels using simple hydrological models, thereby improving operational efficiency of the Waikaremoana Power Scheme. The forecasting is developed specifically for the lower inflow conditions when the lake level differencing approach is most error prone. It has long been recognised that a significant volume of Lake Waikaremoana water leaks through the ancient landslide dam which created the lake. Previous to this study, it was considered that an inaccurate estimation of this leakage rate combined with evaporative losses might contribute to the error within the existing water availability model. A modified catchment water balance and simple regression approach was applied to Lake Waikaremoana to estimate the lake water loss not accounted for by recorded outflows. Estimating this unrecorded loss translates to estimating the intercept of a linear regression relation, where the assumption is made that there is a linear relationship between the discharge of the Aniwaniwa Stream and the net lake water balance (excluding known outflows) under low inflow conditions. On the basis of the confidence intervals about the intercept, the balance term (constant background lake inflow minus leakage and evaporative loss) is estimated to within the range of 2.89 and -1.17 m³s-¹ suggesting that the unknown portion of leakage and evaporative losses are not significant contributors to model error. A useful consequence of the regression was that regression coefficients could be used as a means of upscaling to give net lake storage change for low-flow conditions. This enabled day-ahead water availability forecasts to be acquired from Aniwaniwa Stream discharge day-ahead forecasts. Two forecasting methodologies are developed to forecast the Aniwaniwa Stream discharge: a finite mixture rainfall-runoff model, and a multiple linear regression method. The rainfall-runoff model is formulated initially as a many-parameter model which is then subjected to a lasso-based model simplification concurrent with model calibration. The simplified model forecasts next-day inflows by using a weighted linear combination of hydrograph forms which best match the previous observed discharges in the calibration set where the various weights are linear functions of recent rainfalls. An auto-recalibrating version of the rainfall-runoff model was also developed where model simplification and calibration is carried out for each forecast, with the greatest fitting weights most likely on the most recent discharges to allow for changing catchment conditions. The rainfall-runoff model was calibrated under a range of lasso-based parameter elimination pressures to determine the number of parameters which gave the best validation fit as quantified by the Nash-Sutcliffe fit. The highest validation fit using the original rainfall-runoff model was 50.7%. Using the auto-recalibrating rainfall-runoff model a slightly better maximum validation fit of (52.3%) occurred at an elimination pressure giving 14 final parameters from an initial 300. However, a validation fit which is not much lower (46.8%) is achieved at a higher elimination pressure yielding just 6 final parameters, demonstrating a trade-off between model simplification and validation fit. As expected, the rainfall-runoff model was more successful at predicting low to medium flows because forecasting focus was on the lower flows. Higher discharges were consistently under-predicted. Validation fits of the rainfall-runoff model could probably be improved by increasing the range of possible hydrograph forms available for selection at the expense of model simplicity. The multiple regression technique was applied to forecast ‘next-day’ Aniwaniwa inflows in a simpler way, in this case using just current daily rainfall and discharges as independent variables. The discharge forecasts derived from both techniques are then scaled using the regression equation mentioned earlier to give net storage change estimates into Lake Waikaremoana for low to medium inflows. The regression approach was the more successful for overall day-ahead Aniwaniwa flow forecasts. The final prediction for current day storage change is: ΔS = 0.399(AQ) + 0.59 [1] Where AQ is the observed daily total discharge of the Aniwaniwa Stream. Day-ahead Aniwaniwa Stream forecasts can be approximated by equation [2] then scaled to storage change using equation [1] AQ(Next-day) = 0.095(Train) + 0.558(TQ) + 0.611 [2] Where Train is current day total rainfall and TQ is current day total discharge. This single equation gave higher calibration fits than separate regressions based on season. Using only current rainfall and current discharge as independent variables, the Nash-Sutcliffe validation fits were as high as 66%. The linear regression approach gives the most useful inflow estimates to Lake Waikaremoana for the current day, based on upscaling the Aniwaniwa Stream discharges for low to medium flows. Estimating the day-ahead lake inflows is then equated to estimating day-ahead Aniwaniwa discharges for conditions outside of high flows. For this day-ahead forecasting the regression technique proved better than the rainfall-runoff models. It is thus recommended that the multiple regression technique is applied at the Waikaremoana Power Scheme.
Type of thesis
Hopkirk, C. L. (2011). Forecasting Lake Waikaremoana Water Availability for Hydro Power (Thesis, Master of Science (MSc)). University of Waikato, Hamilton, New Zealand. Retrieved from https://hdl.handle.net/10289/5297
University of Waikato
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.