Predictability of monthly streamflow discharge using remotesensing precipitation data by data driven models


Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, İnşaat Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2017

Öğrenci: MEHMET ALİ ÇOLAK

Danışman: MUSTAFA TUĞRUL YILMAZ

Özet:

Predictability of stream flow has been the focus of many studies involving water resources management and hydroelectric energy production. Many hydrologic models have been developed to predict future and current streamflow at various time lags and locations. However, these physically-based models require reanalyzed future data sets (particularly precipitation forcing data) to predict future streamflow. Alternatively, data driven models can also provide predictions without the need of future projections by relying on the strong seasonality and autocorrelation that exist in the streamflow data. In this study, a data driven approach has been taken to predict monthly streamflow data sets utilizing precipitation data sets and using various linear and non-linear methods. Streamflow predictions of Coruh Basin have been performed using both the Tropical Rainfall Measurement Mission (TRMM) and the ground-based station precipitation (MGM) data sets between years 2000 – 2011. Predictions are validated using independent streamflow measurements acquired from General Directorate of State Hydraulic Works (DSI). A Simple Linear Regression Model (SLR), a Multiple Linear Regression Model (MLR), an Artificial Neural Network Model (MLP), and two Copula Models (Normal Copula and Frank Copula) are constructed and their predictions are cross-compared with the climatology- and persistence-based predictions. To further investigate the source of the predictive skills of these methods, separate predictions are made using the standardized anomaly components of data sets [after climatology (long year monthly mean) components are removed and standardized by dividing by the standard deviation of the data] and complete data sets (normal/non-standardized data sets retaining both anomaly and climatology components). Results show the best predictions are obtained from the climatology-based predictions of the stations for the complete data sets while persistence-based predictions are also strong. Predictions using standardized anomaly data sets are improved when long-term climatology values added. These climatology added predictions show above 0.90 correlations, showing heavy majority of the predictive skill and the relation between the precipitation and the streamflow data sets are due to the strong seasonality impacting both variables.