Please use this identifier to cite or link to this item: http://prr.hec.gov.pk/jspui/handle/123456789/21313
Title: Development of Model-Data Fusion System for Upper Indus Basin Stream-Flow
Authors: Hassan, Muhammad
Keywords: Engineering & Technology
Arti cial Neural Network (ANN), Snow Cover Area (SCA), Upper Indus Basin (UIB), Stream ow Estimation, Data Fusion.
Issue Date: 2022
Publisher: Capital University of Science & Technology, Islamabad
Abstract: Hydrological Processes are complex and highly nonlinear because of their depen- dency upon multiple climate and hydrological variables. Modeling the complexity of these processes is quite challenging because of the many factors that may hinder the e ciency of models in capturing the relationship between these variables and response of the catchment. These factors may include; complex terrain, contrast- ing regimes, limited meteorological network, noise present in the data and con ned economical resources. The Indus Basin is the main source of water for Pakistan. Almost 80% of water requirement of this basin is derived by the Upper Indus Basin (UIB). The UIB has many challenges: which include extreme complexity, varying hydro-meteo-cryospheric regimes, limited meteorological network, climate change pattern and the spread of UIB over political sensitive trans-boundary area. The contrasting regimes in di erent part of basin is the result that the response of the catchment is di cult to capture. Therefore, the trend discrepancies and model uncertainties in the UIB exist, which is re ected in the previous literature as well. The present research work is carried out by focusing; the larger part of the UIB, incorporating multi type/ source data, and applying data preprocessing techniques to minimize the uncertainties in the UIB stream ow measurement. The aim of the research work is to develop Arti cial Neural Network (ANN) based hydrological models that can e ciently estimate the stream ow in the Pakistani part of the UIB. A systematic approach is adopted to improve the di erent steps involved in the hydrological modeling process, which involves data improvement, data selection and data fusion. This ultimately leads to a development of model data-fusion system for the region that optimizes the performance of ANN based stream ow models. The research work is divided into three (03) parts with a main focus on improving ANN based stream ow estimation models for the UIB through; 1. Data preprocessing, 2. By incorporating satellite derived Snow Cover Area (SCA), and 3. Utilizing data fusion. Two-step data preprocessing is per- formed, which includes data transformation through Box-Cox transformation and input selection through Gamma Test. Satellite derived SCA is utilized in combi- nation with the on-ground ow observations to enhance the performance e ciency x of the stream ow estimation models in the region. The ANN models are also de- veloped using a variety of data combinations which are made either on the basis of type/nature of climate variable or through advanced input/feature selection methods. The results indicated; the models developed through data preprocessing performed well as compared to the models developed with original data-set, with more than 90% correlation coe cient in both training and testing phases. The ow de- pendency on satellite derived SCA of UIB region is clearly evidenced with the improved average values of Nash Sutcli e E ciency (NSE) = 99.5/97.5 (train- ing/testing), BIAS = -0.01/-6.6, Root Mean Squared Error (RMSE) = 251.4/532.3 and Variance (VAR) = 63218.0/286917.1 for the models developed using SCA in combination with the other on-ground observations, as compared to the NSE = 99.1/97.1 (training/testing), BIAS = 14.6/-26.1, RMSE = 327.6/531.4 and VAR = 106390.6/284363.4 for models developed using on-ground observations without SCA. The improvement in ANN based models through feature selection techniques including Genetic Algorithm (GA), Hill Climbing (HC) and Sequential Embed- ding (SE) has been observed with better values of statistical indices (NSE and R2 > 0:9), as compared to the models developed through manual selection of in- put variables. However, the models developed utilizing multiple climate variables like Precipitation, Discharge, Solar Radiation and SCA also performed well. Only one feature selection technique, which is Full Embedding (FE) does not provide good results with low values for R2, NSE and high corresponding values of other er- rors. Overall, the models developed through SE outperformed with R2 =93.7/91.4 (training/ testing) and NSE= 97/96. The outcomes of this research could be used to establish a comprehensive linkage between the changing climate variables and their impact on the response of the UIB. The ANN based data fusion models could be applied con dently for the stream ow estimation in the region and ultimately for better management of ood mitigation and reservoir operation at downstream of Tarbela. The research work recommends the use of multi type/ source data coupled with data preprocessing to capture the non-linearity and complexity of catchments which observe contrasting regimes.
Gov't Doc #: 27245
URI: http://prr.hec.gov.pk/jspui/handle/123456789/21313
Appears in Collections:PhD Thesis of All Public / Private Sector Universities / DAIs.

Files in This Item:
File Description SizeFormat 
Muhammad Hassan Civil Engineering 2022 cust isb.pdf 13.10.22.pdfPh.D Thesis5.8 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.