DEVELOPMENT AND PERFORMANCE EVALUATION OF SOFT COMPUTING MODELS FOR RUNOFF ESTIMATION
Loading...
Date
2018-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Indira Gandhi Krishi Vishwavidyalaya, Raipur
Abstract
Management of water resources in a proper manner is a big challenge nowadays. The scarcity of water in some areas and pollution causes the necessity of it. This can be done by estimating the accurate amount of water availability in any watershed area. The rainfall-runoff modeling helps to estimate the actual available flowing water which goes to the water sources like ponds, lakes, rivers etc . Many physical based hydrological models are here but they are very complex due to using empirical equations and consuming time and requires many skilled and trained person. In the present era, the rainfall pattern is very complex and having more spatial variation than past. So it requires the models which can be made in less time, with less inputs and requires less manpower. The soft computing techniques are answer of it. In this study, two soft computing tools (Genetic Algorithm and Artificial Neural Network) were used to model the relation between rainfall and runoff with only three data sets viz., rainfall, runoff and CN. The area considered for the study is Kelo macro-watershed of Mahanadi basin, Raigarh, Chhattisgarh. The daily rainfall and gauge – discharge data for past 12 years (2002 to 2013) were used. Weighted rainfall for the study area was estimated by constructing the Thiessen polygons. More than 90% of the rain falls in the active period (AP) of months 1st July – 31st October. Out of 12 years, 9 years data was used for calibration/training of the model while remaining 3 years data was used for model verification. The developed GA model and ANN models have been analysed on basis of various performance indices, i.e., Mean Absolute Deviation (MAD), Root Mean Square Error (RMSE), Coefficient of Correlation (CC), and Coefficient of Efficiency (CE).
GA model was developed with only two input variables (rainfall and CN) where CN was optimizes in each generation to get the more accurate result and at 6th generation, the value occurs which get the most accurate result for estimating the runoff. On comparison with observed runoff values, it gives the best result with maximum CC and CE values as 97.61% and 94.37% during training and highest CC and CE value as 93.22% and 90.87% and minimum MAD and RMSE value as 4.05 and 4.65 during training and minimum MAD and RMSE value as 4.76 and 5.02 during testing.
ANN models were developed with two combinations of input viz., (i) current day rainfall, previous day rainfalland previous day discharge, and (ii) current day rainfall and previous day discharge. Upon comparison, we see that the model M1 performed better than the other models of ANN. It showed the highest CC and CE value as 87.81% and 84.61% during training and highest CC and CE value as 83.35% and 81.80% and minimum MAD and RMSE value as 7.72 and 8.11 during training and minimum MAD and RMSE value as 8.06 and 8.55 during testing of the model. So from MAD, RMSE, CC & CE point of view, GA and ANN both are performing better and GA is performing best for rainfall-runoff modeling.
Scatter plots between observed and simulated GA values showed that the most of the values lie near 45° line and it is clear from the plots that the model is optimizing the values in testing and training period. Scatter plots between observed and predicted ANN values showed that most of the values lie near 45° line and it is clear from the plots that the model is underestimating the higher values in training while lower values are almost perfectly matched with the observed runoff values in testing period. GA Model is describing runoff slightly better as compared to ANN model for the study area. This might be due to the incorporation of CN value as input. Thus, it can be concluded that use of GA models are slightly better choice than ANN models for rainfall – runoff modelingof the study area.
Description
DEVELOPMENT AND PERFORMANCE EVALUATION OF SOFT COMPUTING MODELS FOR RUNOFF ESTIMATION
Keywords
null