Simulation of Snowmelt Runoff Using SRM Model and Comparison With Neural Networks ANN and ANFIS (Case Study: Kardeh dam basin)

Document Type : Research Article


1 Ferdowsi University of Mashhad

2 Millan University of Tecknology

3 Natural Resources and Watershed of Khorasan Razavi

4 Zabol university


Introduction: Snowmelt runoff plays an important role in providing water and agricultural resources, especially in mountainous areas. There are different methods to simulate the process of snowmelt. Inter alia, degree-day model, based on temperature-index is more cited. Snowmelt Runoff Model is a conceptual hydrological model to simulate and predict the daily flow of rivers in the mountainous basins on the basis of comparing the accuracy of AVHRR and TM satellite images to determine snow cover in Karun Basin. Additionally, overestimation of snow-covered area decreased with increasing spatial resolution of satellite data.Studies conducted in the Zayandehrood watershed dam, showed that in the calculation of the snow map cover, changes from MODIS satellite imagery, at the time that the image does not exist, using the digital elevation model and regression analysis can provide to estimate the appropriate data from satellites. In the study of snow cover in eastern Turkey, in the mountainous regions of the Euphrates River, data from five meteorological stations and MODIS images were used with a resolution of 500 m. The results showed that satellite images have a good accuracy in estimating snow cover. In a Watershed in northern Pakistan in the period from 2000 to 2006, SRM model was used to estimate the snow cover using MODIS images. The purpose of this study was to evaluate the snowmelt runoff using remote sensing data and SRM model for flow simulation, based on statistical parameters in the Kardeh dam basin.
Materials and Methods: Kardeh dam basin has an area of about 560 square kilometers and is located in the north of Mashhad. This area is in the East of Hezarmasjed – kopehdagh zone that is one of the main basins of Kashafrood. This basin is a mountainous area. About 261 km of the basin is located at above 2000 m. The lowest point of the basin is at the watershed outlet with1300 meters and the highest point in the basin, in the North West part of the basin with 2962 meters above sea level. Kardeh dam was primarily constructed on the Kardehriver for providing drinking and agriculture water demand with an annual volume rate of 21.23 million cubic meters.
Satellite image: To estimate the level of snow cover, the satellite Landsat ETM+ data at path 35-159, rows 34-159 over the period 2001-2002 were used. Surfaces covered with snow were separated bysnow distinction normalized index (NDSI), But due to the lack of training data for image classification (areas with snow and no snow), the k-means unsupervised classification algorithm was used.

Extracting the data from the meteorological and hydrological
Since only a gauging station exists at the Kardeh dam site, the daily discharge data recorded at these stations was used. To extract meteorological parameters such as precipitation and temperature data, the records of the three stations Golmakan, Mashhad and Ghouchan, as the stations closest to the dam basin Kardeh were used. The purpose of this study was to simulate snowmelt runoff using SRM hydrological models and to compare the results with the outputs of the neural network models such as the ANN and the ANFIS model. Flow simulation was carried out using SRM, ANN model with the Multilayer Perceptron with back-propagation algorithm, and Sugeno type ANFIS. To evaluate the performance of the models in addition to the standard statistics such as mean square error or mean absolute percentage error, the regression coefficient measures and the difference in volume were used. The results showed that all three models are almost similar in terms of statistical parameters MSE and R and the differences were negligible.

SRM model: SRM model is a daily hydrological model. This equation is composed of different components including 14 parameters. The input values were calculated based on the equations of degree-day factor. The evaluation of the model was performed with flow subside factor, coefficient and subtracting volume.

Results and Discussion: After determining the study area, the DEM in GIS software was produced and was divided into 4 height classes with 500 meterintervals based on the basin area. Thus, the hypsometric map of the region with slope and aspect maps wasobtained from DEM. The parameters that were entered into the SRM model included area, the average height of DEM and area of slope directions. Weighted average altitude was about 2007 m in the basin. Height classes of 2000-1500 comprise about 47 percent of the total area, with the highest frequency. The main slope happens in the southwestern region(SE). The results show that the model has properly simulated the daily flow hydrograph at the time of the study period. Factor subtracting volume was modeled based on daily discharge hydrograph at a 17-year period. The best x and y values of the simulated hydrograph for watershed dam Kardeh were respectively 0/79 and 0/084 and finally entered into the model. To evaluate the model for the period of 79-80, the subtracting volume was about 0.21 percent, the regression coefficient was 0.91, the calculated runoff volume was 4/876 million cubic meters and calculated discharge was estimated 0.212 cubic meters per seconds, that indicated a very good agreement with observed values. In addition, it was shown that between the parameters introduced into the model, change of the snow runoff coefficient and the coefficient of flow subsidence have the highest sensitivity, and then two parameters were accurately calibrated, to reach more conformity with ground truth. The results showed that the use of images with high spatial resolution, results in relatively good results in determination of snow-covered surfaces. These results were in agreement with other studies. SRM model is relatively successful so that changes in daily flow modeling provide a better quality. The comparison of the mean absolute percent error between the three models of ANFIS than the ANN model by 40 percent compared to15 percent SRM model has reduced the error of the simulation process and the difference in volume between ANN and ANFIS models were better than the SRM model and the value of this parameter for both models are low.


1- Akbari M., Ranaee E., Mirzakhan H., and Dargahi A. 2016. Snowmelt runoff simulation using SRM hydrological model (Case Study: Kardeh dam basin). Watre Engineering Journal, 9(30): 63-73.
2- Ahmadzadeh Gharah K., Mirlatifi S.M., and Mohammadi K. 2010. Comparison of Artificial Intelligence Systems (ANN & ANFIS) for Reference Evapotranspiration Estimation in the Extreme Arid Regions of Iran.Journal of Water and Soil, 24(4): 679-689.
3- Fathabadi A., Salajegheh A., and Mahdavi M. 2009. River flow forecasting using neuro-fuzzy methods and time series models. Iran-Watershed Management Science & Engineering, 2(5): 21-30
4- Fatahi A., Delavar M., and Ghasemi A. 2011. Simulation of snowmelt runoff in mountainous basin using SRM model, (Case Study: Bazaft basin). Journal of Geographical Sciences and Applied Research, 20(23): 129-141.
5- Firat M., and Gungor M. 2007. River flow estimation using adaptive neuro-fuzzy inference system. Mathematics and Computers in Simulation, 75: 87-96.
6- Ghafari G.A., and Vafakhah M. 2013. Simulation of Rainfall Runoff Process using Artificial Neural Network and Adaptive Neuro Fuzzy Interface System (Case Study: Hajighoshan Watershed). Journal of Watershed Management Research. 4(8):120-136.
7- (visited 20 Jul 2010).
8- Kashefipur M. 2002. Use of ANN in river engineering. 6th International Seminar on River Engineering, shahid Chamran University, 6: 126-133.
9- Li X., and Williams M.W. 2008. Snowmelt runoff modelling in an arid mountain watershed, Tarim Basin, China. Hydrological Processes, 22:3931-3940.
10- Majid Rezai B., Feyzolahpour M., and Sadrafshary S. 2013. Using Neural Fuzzy Inference System to Estimate Sediment Load and Comparison with MLR and SRC Models in Ghranghu River Basin. Journal of Physical Geography. 45(2): 77-90.
11- Menhaj M. 2002. Artificial neural networks. Amirkabir university press.
12- Nayak P.C., Sudheer K.P., Rangan D.P., and Ramasastri K.S. 2005. Short-term flood forecasting with a neurofuzzy model. Water Resources Research. 41: 2517-2530.
13- Nezam Kayami KH., Nezam Khayami C., and Nezam Khayami K. 2010. Estimation of suspended sediment load in rivers using Artificial Neural Network and Adaptive Neuro Fuzzy Interface System (Case study: River Gharehsou in Ardebil). Iranian Hydraulic Conference. IRAN, Tehran,
14- Nurani V., Kaynejad M.A., and Malekani L. 2009. Application of neuro- fuzzy inference system in rainfall-runoff modeling. Journal of Civil of Engineering & Environment, 39(4): 75-81.
15- Porhemmat J., and Saghafian B. 2007. Evaluation of Spatial Resolution of Satellite Data on Snow Cover Estimates. Journal of the Iranian Natural Res, 60(2): 383-395.
16- -Rango A., and Martinec J. 1995. Revisiting the degree day Method for Snowmelt Computations. JAWRA Journal of the American Water Resources Association, 31: 657-669.
17- Ross T.J. 1995. Fuzzy logic with engineering application. McGraw Hill Inc.
18- Seidel K., and Martinec J. 2004. Hydrological Applications of Satellite Snow cover Mapping in the Swiss Alps. Paper presented at the Proceedings of EARSeL-LISSIG-Workshop Observing our Cryosphere from Space.
19- Tokar A.S., and Markus M. 2000. Precipitation runoff modeling using artificial neural network and conceptual models. Journal of Hydrologic Engineering, 5: 156- 161.