Infilling Missing Rainfall Data in the Lake Tana Basin

Adimitew, Zemenu

BDU IR Home
→
Bahir Dar Institute of Technology (BiT)
→
Faculty of Civil and Water Resources Engineering
→
Structural Engineering
→
Thesis
→
View Item

dc.contributor.author	Adimitew, Zemenu
dc.date.accessioned	2020-03-19T10:58:45Z
dc.date.available	2020-03-19T10:58:45Z
dc.date.issued	2020-03-19
dc.identifier.uri	http://hdl.handle.net/123456789/10748
dc.description.abstract	The consequence of a long and complete rainfall record is very important for working a hydrological analysis effectively. Generally, the data series in these records may contain miss for various reasons. Among these reason instrumental failure, absence of observer, loss of data and shortage of budget. The objective of this study is to analyses the different methods available for infilling miss rainfall data records and propose the best method. Five different techniques are studied to ascertain their suitability in the twelve gaging station. The methods studied for this paper are the Arithmetic AverageMethod, Normal Ratio Method, Inverse Distance Weighting Method, Multiple Regression Method and Artificial Neural Network MethodDaily rainfall data of thirty-eight gauging stations in the Lake Tana basin are collected.These stations allocated to themajor four contributing rivers in the Lake Tana basin like Gilgel Abay, Gumara, Ribb, and Megech due to this gaging station reduced to twentyseven and by using high Correlation Coefficient and low missing proportion rate the gaging station compressed to twelve gaging stations. These twelve gaging stations are used for this study. The data generated for the target stations are compared with real observations made, based on error statistics, Error Standard Deviation (STDE),Root Mean Square Error (RMSE) and Correlation Coefficient (CC). The results of this study showed that for target stations that have two or more neighbouring stations so the target stations between actual value and estimation with a high correlation coefficient, small Error Standard Deviation, and small Root Mean Square Error use thus error stastics preferred the neural network method give good estimations. The target stations that have relatively low correlation coefficients, high missing proportion rate with the neighbouring stations are selected when the neighbouring station less than one. Multiple linear regression method outperformed the second best method among others. To obtain accurate results from neural network method after many trains by changing the number nodes. The best weight value gets during train with the highest R2 or Correlation coefficient. Keywords	en_US
dc.language.iso	en	en_US
dc.subject	Structural Engineering	en_US
dc.title	Infilling Missing Rainfall Data in the Lake Tana Basin	en_US
dc.type	Thesis	en_US