DEVELOPMENT OF SYNTHETIC HYDROGRAPH DATABASE BASED ON CHARACTERISTICS OF WATERSHED

• Main Author: Faturochman Saefulloh, Dony
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70228
• Institution: Institut Teknologi Bandung

Most watersheds in Indonesia have problems with a lack of rainfall or discharge data, especially in ungauged watershed areas. This can affect the precision and accuracy of values in the design and planning of water resources infrastructure. Synthetic unit hydrographs were developed to estimate unmeasured watershed flows based on the characteristics of the watershed and the rainfall. The data set tested on the Synthetic Unit Hydrograph (SUH) generally comes from foreign watershed characteristics that are not by local watershed characteristics, so it is necessary to adjust the parameters of synthetic unit hydrographs by calibrating the hydrograph of observation units. This is shown in the calculation of SUH Snyder Alexeyev, ITB, SCS, and Limantara, which are recognized in the Indonesian National Standard (SNI) compared to hydrographs of units o in 32 watersheds in Indonesia that have a correlation value of R2 for peak discharges of about 0.38 to 0.84, and times to peak of about 0.05 to 0.16. Thus, a new SUH model is needed for unmeasured watersheds that match the characteristics in Indonesia without the parametric calibration required in its application. The relationship between rainfall and runoff in watersheds is highly linear and complex. Problems with non-linearity and lack of accuracy in synthetic unit hydrograph modeling in local watersheds can be addressed using an Artificial Neural Network (ANN) approach. The ANN method behaves as a black-box model and can extract the relationship between the input and output of a process without a process in physics. Learning and generalizing information from a suitable data group allows ANN models to solve complex, large-scale problems, and nonlinear modeling. Peak discharge (Qp) is the main parameter used in planning a water resources infrastructure. Time to Peak (Tp) and base time (Tb) are the parameters required in the early warning system to control the destructive power of natural resources. These three parameters will be used in the output of the triangular hydrograph model. This study aims to develop a triangular unit hydrograph using ANN models for various watersheds in Indonesia resulting from the process of training, validating, and testing models applied to watersheds in Indonesia based on hydrographs of observation units in 32 watersheds on the island of Java which have been verified by observational data of hourly data, both rain data and discharge data. The ANN model was developed based on input data on watershed characteristics in the form of watershed area (A), river length (L), watershed shape (F), flow coefficient (C), and river slope (S). This model will produce a triangular synthetic unit hydrograph output in the form of peak discharge (Qp), peak time (Tp), and base time (Tb). The backpropagation algorithm and the Explicit ANN methods are used to develop an ANN model. The most accurate synthetic unit hydrograph method used in Indonesia, according to SNI, is the Snyder-Alexyev method. In contrast, other SUH methods often used in Indonesia, such as Synder, SCS, Gama 1, ITB, and Limantara, are also listed in SNI with a comparison of their respective parameters. The regression optimization model was created based on the parameters of the SUH in SNI, and the addition of five characteristic watershed parameters (A, L, F, C, S) with several new equation models created in this research to determine Qp and Tp to compare which regression model is the most accurate with optimization using all watersheds in this study used for calibration, validation, and testing. The optimization of this regression model can prove each SUH model's accuracy in SNI and produce a new SUH equation model that can be applied directly without the need for calibration in the watershed concerned. This study proposes one ANN Backpropagation model, eight regression models for Qp, seven regression models for Tp, six ANN models for Qp, and two ANN models for Tp. This study shows that regression models that use more than two parameters characteristic of the watershed produce better results than those using one or two parameters as used in Snyder, SCS, and other common SUH contained in SNI. From the results of this study, several SUH models were produced to predict Qp and Tp, models to predict Qp, namely Regression Model Qp7, Explicit ANN Model Qp1, and ANN Backpropagation with R2 correlation values of the three models approaching one for model calibration data as well as validation and test data. The best performance models for predicting Tp with an R2 value close to one are the ANN Backpropagation Model and the Explicit ANN Model Tp 2. From the results above, the three SUH models can predict Qp accurately, and the two SUH models can predict the Tp value accurately. Training, validation, and simulation testing of ANN models in both explicit and backpropagation methods in 32 watersheds in Indonesia resulted in a corresponding line between observation and simulation data. This research shows that the SUH ANN model can be applied to watersheds in Indonesia without the parametric calibration required before its implementation.