regression model

Measuring and recording climate data of gauges are usually used to develop and calibrate hydrological models. Missing hydrological and climate data can cause decreasing models accuracy or disability of developing models. In this study, remaking the missing data of Chelgerd hydrometric gauge located at the upstream of Ghaleshahrokh-Chelgerd sub basin as part of Zayanderud Basin was surveyed. It measures discharge of the first Koohrang tunnel inflow. In order to estimate the missing data, regression support vector machine model was employed and to improve the model performance, Ensemble Kalman Filter (EnKF) was used as data assimilation technique. For evaluating the regression model performance, R, RMSE and Nash-Sutcliff citeria was implemented. The results showed values of 0.83, 3.42, 0.7 for training part and the values of 0.70, 20.38 and 0.25 for test part of the model, for R, RMSE and NS respectively. By using EnKF, the performance of the regression model has been improved and acceptable results were obtained. To modify the EnKF results, the data of Ghaleshahrokh station located at basin outlet as reference station was used alongside a second SVR model. The values of R, RMSE and NS were 0.96, 5.2 and 0.81 for training and 0.76, 6.6 and 0.66 for test stage respectively.

This study is performed to retrieve precipitation amount using Spinning Enhanced Visible and InfraRed Imager (SEVIRI) from Meteosat Second Generation (MSG). According to the relationship between the infrared channels brightness temperature and the microphysical and optical properties of the clouds (such as cloud top temperature, cloud top height, optical thickness, particle size, and particle phase), and the influence of changes in these properties on the determination of precipitation amount and intensity, the relationship between brightness temperature and precipitation have been studied for two stations in Hormozgan province. The performance of an artificial neural network and also several regression models to estimate precipitation has been evaluated. The results showed that the exponential Gaussian process regression model with performing principal component analysis by RMSE of 0.44, POD of 0.96 and the HSS of 0.67 for precipitation threshold 0.1 mm for the less than 10 mm precipitation data set have the best performance. The artificial neural network also presented a RMSE of 1.27 which indicates weaker performance in comparison with the regression model but showed good performance in distinguishing precipitation conditions from non-precipitation conditions (POD of 0.85 and HSS of 0.48). By comparing the correlation between precipitation and brightness temperature of infrared channels (average 0.36), and the correlation between observed and retreived precipitation (0.91 in the regression model and 0.43 in the neural network), it can be concluded that the precipitation products which extracted in this study, show a good correlation between observed and retrieved precipitation.

Pipes failure events in the water distribution networks provide leakage of water. Failures cause the loss of significant fresh water and investments losses. The most important parameters are: material, age, length, diameter and hydraulic pressure. In this paper four statistical methods have used for analyzing pipe incidents, with the goal of estimation of failure probability in future, with finding the most influences parameters in the incidents. The statistical regression models using in this research are linear regression model, exponential regression model, Poisson regression model , and Logistic regression model . For evaluation of the models, the data of a pilot in the first district of the Tehran’s Water and Wastewater Company with more than 48500 consumers, total pipe length of 582702 meter, different materials and diameters were used. The results demonstrated that the logistic model has a better performance than others to predict the future events with a higher probability.