科学与发展网络(SciDev.Net)

The groundwater level in the study area has fallen in recent years, and fields have gone dry due to over-extraction

Flickr/eyesore9

[NEW DELHI] Scientists in India have developed a more efficient computing model to predict monthly changes in groundwater, which is depleted in many areas of developing countries because of damaged soils and poor irrigation practices.

A team of scientists based in India's four southern states used the concept of an 'artificial neural network' (ANN), a computing technique that mimics the way millions of neurons — responsive cells in the nervous system — process and analyse the myriad bits of signals they receive.

An ANN contains highly interconnected single units called 'neurones' into which data is fed for analysis. ANNs can analyse complicated or imprecise data, and help detect patterns and trends that are too complex to be noticed by either humans or other computer techniques, say computer experts.

The new ANN-based model can predict monthly groundwater levels with 93 per cent accuracy, and "provide satisfactory predictions with limited groundwater level records", the scientists report in the 10 April issue of Current Science published by the Indian Academy of Sciences, Bangalore.

The scientists tracked water levels in 22 wells in a 53-square-kilometre area of the Maheshwar watershed in Andhra Pradesh state, for six years beginning 2000. Maheshwar receives an average annual rainfall of 573 millimetres, in temperatures up to 44 degrees Celsius. 

They fed in the data into an ANN and found it could predict the groundwater levels well.

People residing nearby extract groundwater for drinking and irrigation from a large water tank in the study area. Groundwater levels in this area have fallen in recent years and fields have gone dry due to over-extraction.

To forecast groundwater levels scientists need enormous amounts of data on weather, underground water levels and the structure of the ground. Although weather data is available routinely, geological and water data requires trained scientific personnel and funds and is comparatively scarce, lead author P D Sreekanth — a scientist at the National Research Centre for Cashew in Puttur — told SciDev.Net. This leads to inaccurate forecast models, he says.

The advantage with an ANN is that it can work with such incomplete data.

Groundwater depletion can also be worsened due to projected changes in temperature and rainfall due to global warming. There is enormous potential for researchers to simulate changes in climate change, land use and land cover and use an ANN to predict groundwater levels under varying scenarios, says Sreekanth.

Link to full report in Current Science [852kB]