Prediction of Stream Flow in Ungauged Basins - a Comprehensive Framework
Abstract
It is well established that critical information on stream-flow is essential in reducing uncertainties in planning and design of various water resource projects. Lack of data, at the desired spatial and temporal resolution, poses an enormous challenge in developing meaningful prediction models. Powerful techniques like Artificial Neural Network (ANN) modeling provide reasonably accurate prediction models, however development of such models require substantial amount of past data. Currently, empirical equations developed across the span of several hundred years are used on a regionalized basis. These equations are usually very simple, allowing for easy application, however not very accurate. This limited accuracy can be attributed to the use of noisy data and inclusion of only limited stream-flow variables. This study is an attempt to process noisy data and incorporate catchment variables to improve the accuracy of existing relationships whilst maintaining their simplicity. This study presents a comprehensive framework starting from data-processing to data-analysis that enables the development of regionalized empirical equations. A case-study has been presented for the sub-basins in "Dakshina Kannada" (Coastal Karnataka, India). Firstly, the data has first been processed to remove any outliers and estimate missing values, by replacing missing values with the average values of the neighboring entries for discrete data-sets or by using Least Square principles (LS) for continuously distributed date. Secondly, the existing models have been improved based on the processed dataset obtained through Exploratory Data Analysis (EDA). Further, utilizing Principal Component Analysis (PCA) other important parameters have been identified. All these parameters have then been included to arrive at an "improved regionalized relationship". Finally, the improved regionalized relationships have been evaluated for their performance based on the Correlation Coefficient and Standard Error apart from graphical analysis based on the scatter and time-series plots. Tools such as Geographic Information System (GIS), EDA, Least Square (LS) and simultaneous regression method have been used in arriving at the improved regionalized models. The power of EDA and GIS techniques has clearly been brought out through the development of regional regression models. Further, it was concluded that runoff processes are extremely sensitive to rainfall (as they are insignificantly affected by change in the PET values.) Parameters percent runoff and AWC also don't seem to influence runoff. The framework developed provides necessary platform to determine more accurate regionalized models, with deep insights into the watershed characteristics and becomes an essential component in Prediction in Ungauged Basins (PUBs) application.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2012
- Bibcode:
- 2012AGUFM.H13A1308G
- Keywords:
-
- 1804 HYDROLOGY / Catchment;
- 1816 HYDROLOGY / Estimation and forecasting;
- 1847 HYDROLOGY / Modeling