While natural disasters are plaguing our planet from centuries, in India, floods are the major disasters which causes massive loss to life and property. Due to the heavy southwest monsoon rains and distended river banks, almost all of India is flood-prone. Extreme precipitation events such as flash floods and torrential rains have become common in central and south India from past few decades. To mitigate the effects of floods, both structural and non-structural measures can be employed, such as dams, dykes, channelization, flood proofing of properties, land-use regulation and flood warning schemes. Earlier advanced warning can be achieved through mathematical modelling of various data which results in the occurrence of floods. We are intending to tackle the problem of floods with a two-way approach.
Prediction - Flash floods are generally caused due to heavy rainfall and overflow of rivers. Hence, we would build a model to predict the probability of flood occurrence based on weather data, rainfall forecast and river water levels, relative elevation of land masses and few other factors. We are looking at Artificial Intelligence techniques such as the Artificial Neural Networks & Support Vector Machines to build the prediction model. The ANN has been found suitable for modelling the rainfall-runoff process in a wide variety of catchments under specific circumstances. Rainfall-Runoff modelling is a mathematical model describing relations of a rainfall catchment area, drainage basin or watershed with rainfall parameters. It basically produces a surface runoff hydrograph in response to a rainfall event which will be used in flood forecast. We intend to research about Recurrent Neural Networks, of Elman and Jordan forms specifically and compare it with the basic Feed Forward Networks and Multilayer Perceptrons since it has been found to give better results according to latest research from Institute for Water Education. If everything goes well, we will be implementing using the RNN algorithm.
Management - Communication and connectivity in the affected areas is highly important and with those, relief provision is much more effective. During floods and other natural disasters, there will be a lot of damage to infrastructure. Network towers will cease to work. But, many cell phones may still be working. Some may be present in neighbouring unaffected areas and some phones in the affected areas may still work considering that few are waterproof. At this time, when network towers cannot be used, we propose the use of a special routing algorithm, Petal Ant Routing for Mobile Ad-Hoc networks among the working cell phones. This does not require any infrastructure. We would build a model where the nodes(phones) are enclosed in an elliptical area and a modified version of Petal-Ant Routing is used to traverse and connect the source node to the destination node. Using this we can establish communication and make relief provision a lot easier. Once communication is established with a quick response centre, it becomes a high priority task to figure out where exactly people need urgent help and prioritize the search and rescue operation. For this process, we would like to implement an 'Active Listening' code which will listen to the calls and convert it to text, which can be further processed to identify the location of the victims.