By using a technique inspired by the working of the brain, scientists have been able to accurately predict streamflow at dam location three to five days in advance. Such methods give a reliable estimate of the quantity of water available for the next few days, and hence help better manage the precious resource. Streamflow is a measure of how much water flows in a river or a stream, and at what rate.
Researchers from the Indian Institute of Science applied three different statistical techniques to forecast streamflow for short lead times. Out of them the technique called 'Field Forward Neural Network' (FFNN) gave the best forecasts. The researchers presented their findings at the International Conference on Water Resources, Coastal and Ocean Engineering (ICWRCOE).
FFNN is based on a broader statistical technique called the Artificial Neural Network (ANN), a statistical method inspired by the working of the brain. ANN works well when a large number of factors affect the outcome. Researchers also used Radial Basis Neural Network (RBNN), yet another method based on ANN, and Auto Regressive Integrated Moving Average (ARIMA). Each technique employs a different way of manipulating data.
The researchers used data driven methods, as opposed to a physical based approach. Hence, they didn’t have to know of all the complex physical processes and their inter-relationships that govern the streamflow. They instead used the large data sets available to construct their model.
They applied the statistical procedures to the water flow data collected at a location near the Hirakud reservoir, built across the Mahanadi river in Odisha, and compared the predictions of each method with the measured streamflow data. Their analysis showed that streamflow predicted by FFNN was closest to the measured data.
Mahanadi is a 858 km long major river flowing through the states of Odisha and Chhattisgarh in India. It is the major source of water for more than 80 surrounding cities and villages including Raipur, Sambhalpur, Cuttack etc. Obviously, any step that leads to better management of Mahanadi river is a boon to the region that depends on it.
Currently, among other physical based approaches, methods like Support Vector Machine (SVM), Adaptive Neuro Fuzzy Interference System (ANFIS) or a conjunction of these methods are employed for yearly, monthly and even weekly forecasting. The current research however, apart from making daily forecasts, has proven a lot more accurate at forecasting and simple to use.
“The water flow from a dam is used for a lot of purposes, like human consumption, agriculture, industries etc, while also maintaining a minimum flow to conserve the ecosystem. If we can forecast the amount of water coming in, we can use the water in the reservoir efficiently according to our priorities and avoid flooding and draining of the reservoir” says Alok Pandey, a research student at IISc and an author of the paper.
About the authors:
V V Srinivas is an Associate Professor at the Department of Civil Engineering, IISc.
Alok Pandey is a research student at the Department of Civil Engineering, IISc.