dc.description.abstract |
Predicting runoff and analyzing its variations under future climates play a vital role in
water security, water resource management and the sustainable development of catchment. The
rate of runoff is required for the design of drains, canals and other channels and for the prediction
of water level in the streams and rivers. Runoff prediction, as a nonlinear and complex process is
essential for designing canals, water management and planning, flood control and predicting soil
erosion. To model such non-linear systems, Artificial Neural Networks (ANNs) are effective
tools. The ANN functions as a data-mining tool, in which the input and output data sets are fed to
the software and trained before validating the model.
As per the study objectives, the required meteorological data were acquired from
meteorological observatories that contained six input variables viz. rainfall, relative humidity,
maximum and minimum temperature, wind speed and evaporation. Runoff data from river gauge
station working under The Central Water Commission located at Kumbidi was used for the study
.Statistical data analysis was carried out to identify important climatic variables for developing
ANN model.
In this study the MATLAB software was used for ANN model development
having neural network tool with different network architectures, transfer function, learning
functions etc. The networks were developed by using feed forward back propagation with tan
sigmoid transfer function. Trial and error procedures were attempted with different
combinations of ANN architectures and tan sigmoid transfer function to arrive at two modeling
strategies. The acquired data was partitioned into training, testing and validation data set.
Training and testing data set were used for model development purpose while validation data
set was used for model evaluation. The modeling strategy of ANN model having back
propagation learning algorithm, tan sigmoid transfer function and model input strategy having
rainfall as input by assigning number of layers as 5,10,15,20,25,30 and 40 gave two best
models. The best models –Model 3 and Model 6, exhibited better results with coefficient of
correlation(R), coefficient of determination(R 2 ) and Root Mean Square Error (RMSE). SWAT
simulated runoff data were collected and a comparison of ANN predicted runoff was done with
the SWAT predicted runoff before which the comparison of SWAT was done with the
observed runoff. It was found that SWAT simulated runoff was more similar to observed runoff
than ANN predicted runoff .However, the ANN predicted values were comparable with the
SWAT model. |
en_US |