Development Of A Reservoir Inflow Forecasting Model For An Ungauged Catchment
A user-friendly single-event distributed reservoir inflow forecasting model for the ungauged Batu Dam Catchment is presented. The Batu Dam Catchment located in the Gombak District, Selangor Darul Ehsan, approximately 20 km north of Kuala Lumpur, is a 50.7 km2 tropical forested rural catchment. The m...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English English |
Published: |
2005
|
Subjects: | |
Online Access: | http://psasir.upm.edu.my/id/eprint/5132/1/FK_2005_31.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | A user-friendly single-event distributed reservoir inflow forecasting model for the ungauged Batu Dam Catchment is presented. The Batu Dam Catchment located in the Gombak District, Selangor Darul Ehsan, approximately 20 km north of Kuala Lumpur, is a 50.7 km2 tropical forested rural catchment. The model consists of five sub-models, namely the physical data input sub-model, the rainfall data input and excess rainfall computation sub-model, the rainfall runoff simulation sub-model, the baseflow volume computation sub-model and the reservoir water level increment simulation sub-model. The whole formulation of model was set up using the MapBasic and MapInfo Geographical Information System package. The catchment was delimitated based on the finite element concept. The rainfall losses in the catchment were assumed to be consistent throughout an event and uniform over the entire catchment. The catchment losses rate concept developed was assumed to be dependent on catchment antecedent soil moisture condition (catchment wetness index) and weighted average rainfall intensity. A catchment wetness index was formulated empirically based on the net total rainfall volume retained in the catchment cumulated from a five-day period prior to the simulated event following the 5-day Antecedent Precipitation Index (API5) approach. This catchment losses rate works in conjunction with the areal reduction factor to compute excess rainfall. With excess rainfall as input, the rainfall runoff simulation sub-model was developed based on the one dimensional Saint-Venant equations with kinematic wave approximation and solved using the finite element standard Galerkin’s residual method, and incorporating Manning’s equation. The spurious oscillatory behaviour of the simulated direct runoff hydrographs when approximated by the standard Galerkin’s residual method can be suppressed by using a one minute time increment based on investigations taking into consideration the Courant condition. An empirical equation for computating baseflow volume for reservoir water level increment simulation was developed based on the five previous day approach similar to that in the API5. The reservoir water level increment sub-model is used to simulate the reservoir water level increment, by considering all the other inflows and outflows of the reservoir.
Historical rainfall events from 1989 to 2001 were used for model parameter calibration and model verification purposes. One hundred and forty cases selected were divided into thirteen groups according to their weighted average rainfall intensities. Cases from each group were then further sub-divided randomly into two separate sets in order to form two sets of cases. One set was used for the calibration of the unknown parameter, catchment losses rate. The Catchment Losses Rate-Catchment Wetness Index-Weighted Average Rainfall Intensity (LWRI) curves were proposed. Seven LWRI curves were finalized and selected, and were programmed into the model for model verification and forecasting purposes. The accuracy of Manning’s coefficients used in model parameter calibration was confirmed by extending the 24-hour simulation period of the selected calibration cases to 48 hours. The 0.400 and 0.040 Manning’s coefficients for overland and channels were confirmed to be accurate. This was supported with statistical tests on the simulated increment and the respective measured increment, where a very strong 0.9799 correlation coefficient from the correlation analysis, a relatively small mean absolute error that does not exceed 1.47 cm at 95% level of confidence from the single mean t-test, and not enough evidence to support that the means and the variances of simulated increments and measured increments are different through the paired t-test and the F-distribution variance ratio test respectively.
The other set of cases was used for LWRI curves verification and model verification purposes. The LWRI curves were found to be accurate in determining catchment losses rates. The model was verified to be able to simulate the reservoir water level increment accurately. This was supported by the results of the statistical tests carried out on the simulated and the respective measured increments. A very strong 0.9799 correlation coefficient from the correlation analysis, a relatively small mean of absolute error not exceeding 2.20 cm at 95% level of confidence from the single mean t-test, and not enough evidence to support the means and the variances between the simulated increment and the measured increment are different from the paired t-test and the F-distribution variance ratio test.
The model was evaluated by comparing it with the rational method. Results of statistical tests show the model performing much better than the rational method. The respective correlation coefficient and mean of absolute error for the rational method were found to be 0.8602 and does not exceed 12.58 cm at 95% level of confidence, respectively, while the paired t-test shows that there is not enough evidence to support that the simulated increment and the measured increment are the same. The computed Theil’s coefficients for the model and the rational method, which are 0.062 and 0.266 respectively, also show that the model is more reliable compared to the rational method.
From the sensitivity analyses, the impact of changing Manning’s Coefficient of overland on the simulated direct runoff hydrograph, as well as the reservoir water level increment, is higher than the impacts of changing Manning’s Coefficient of the channels. The study reveals that more caution and effort should be emphasized in deciding Manning’s coefficient of overland than that of channels. The results also show that the impact decreases with increasing rainfall intensity. The impact of catchment wetness index on the catchment losses rate and the corresponding reservoir water level increment was found can be moderately high, but is case dependent. |
---|