Forecasting river flow using neural intelligence models and LARS-WG models (Case study: Kashkan Watershed)

Introduction
Due to water resources limitations in the country, future decisions will be based on future status. In the present study, to investigate the effect of climate change on hydrological components (minimum temperature, maximum temperature, precipitation, and sunshine hours), the information of Kamalvand water gauging station from the headwaters of Kashkan River located in Khorramabad City and Khorramabad meteorological stations were used. For this purpose, taking into account the climate information by LARS-WG software, their values were predicted in the future under three scenarios including A1B, A2, and B1 in three time periods of 34 years from 2011 to 2113. Considering the limited water resources, making management decisions will require knowing the future state of water resources. This phenomenon can cause considerable damage in vulnerable areas. Therefore, as water and its related issues are among the main concerns of mankind in the coming periods, it is necessary to evaluate the occurrence of climate change and the extent of its impact on water resources. According to the importance of knowing the amount of river flow in hydrological studies and water resources management, and the lack of information about the changes in the amount of river flow in the coming years, this study was designed and implemented to predict the daily flow of the Kashkan River in Khorramabad City in the coming years. To achieve this purpose, with the application of the atmospheric general circulation model and various intelligent neural models, the prediction of river flow with high accuracy under different climate change scenarios was examined.
 
Material and Methods
This watershed forms an important part of the water-rich tributaries of the Karkhe River and covers about one-third of the land of Lorestan Province. In this study, the data of the Kamalvand River gauge station from the headwaters of the Keshkan River located in Khorram Abad have been used. In this study, relying on the ability of the artificial neural network, the application of this method was evaluated along with two hybrid models including neuro-fuzzy (CANFIS) and neuro-genetics (ANN-GA). The water crisis can be considered one of the challenges facing different regions of the country in the coming years. Managing water resources and dealing with the water shortage crisis requires knowing the state of hydrological components in the coming years. For this purpose, in this study, the status of meteorological parameters and the amount of river flow in the coming years were investigated. To achieve this goal, the capability and application of the LARS-WG model in forecasting meteorological parameters and intelligent neural models were used in river flow forecasting.
 
Results and Discussion
Based on the obtained results, the trend of increasing minimum and maximum temperature and evaporation and transpiration was predicted until 2113. Regarding the parameters of precipitation and solar radiation, a decrease in precipitation and an increase in radiation were predicted from 2080 to 2113. Comparing the performance of intelligent neural models in predicting river flow showed the superiority of the neural-fuzzy model over artificial neural and neuro-genetic models. river flow prediction with the neuro-fuzzy model until 2113 under scenarios A1B, A2, and B1 indicated that the lowest amount of river flow will be observed in scenario A1B and the highest amount will be observed in scenario B1. The temporal changes of the river flow during different seasons showed that the river flow will increase in spring, autumn, and winter. In general, according to the changes in meteorological parameters and the observed values ​​of the river flow, the description of the changes in the river flow in scenario A1B was closer to reality. This makes it necessary to properly manage the river flow, especially in the summer season of 2080-2113.
 
Results
The results indicated that minimum and maximum temperatures and evapotranspiration during the next years will increase. In scenario A2, the precipitation changes trend was decreasing and solar radiation was increasing, however in other scenarios trend of increasing and decreasing. Then the discharge amount under different scenarios was calculated. The forecasting discharge values by intelligent models showed that the CANFIS model had more accuracy than the ANN and ANN-GA. The results of the optimized structure of CANFIS illustrated that the minimum discharge value in the future will occur in scenario A1B and the maximum discharge amount will be recorded in scenario B1. The evaluation of the seasonal trend showed that the flow rate increased in spring, autumn, and winter compared to the base period by 20.60, 17.31, and 9.27%, respectively. The lowest river flow in summer will occur under the A1B scenario during 2080-2113. rivers are one of the most important effective factors in the geomorphological processes of the earth and the hydrological cycle. Effective factors in very diverse hydrological processes and their applications in designed models are very difficult and the existence of high uncertainties and strong nonlinearity of the data complicates the issue. Long-term records of hydrological data show the temporal changes in discharge caused by climate change and vegetation changes.