2012年1月10日 | EN
The 'ensemble' predicted the Brahmaputra river floods of 2007 and 2008
Many poorer countries rely on traditional weather forecasting techniques, in which an initial set of weather conditions is fed into a single model to predict future adverse events, typically one or two days ahead.
Instead, the 'ensemble' method, developed in the 1990s but requiring a computational infrastructure that eludes most developing countries, uses multiple models, or versions of models, each processing different types of information to reach its conclusions. Ensembles yield an average forecast up to 15 days ahead.
Peter Webster, professor at US-based Georgia Institute of Technology's School of Earth and Atmospheric Sciences, and Jun Jian of Dalian Maritime University, China, wrote last month (13 December) in the Philosophical Transactions of the Royal Society A that south and east Asian countries needed more specific forecasts and a warning time of at least ten days.
This is particularly important, they said, given the likelihood of more frequent and longer-lasting floods from the Brahmaputra and Ganges rivers in the 21st century due to global warming.
Webster's team, along with flood and water experts in Bangladesh, have been using the European Centre for Medium Range Weather Forecasts model, which uses the ensemble approach, to forecast the onset and duration of floods in Bangladesh between 2003 and 2009.
The model accurately forecast that there would be floods in the Brahmaputra River in 2007 and 2008, giving farmers around the river time to change crop practices such as sowing or harvesting, and store food and household items.
Webster said the same system could work well for large monsoonal river basins located around the Tibetan Plateau — the Irrawady, Mekong, Red, Yangtze, Yellow and Indus rivers.
But Ahmadul Hassan, director of Research and Development training at the Center for Environmental and Geographic Information Services, Bangladesh, which studied the ensemble approach told SciDev.Net that it was "excellent" for forecasting only up to four days.
He said comparison with ground data showed that errors crept in beyond this and "more work needed to be done [to improve accuracy] for a ten-day forecast".
Pradeep Mujumdar, a professor in the department of civil engineering at the Indian Institute of Science, said that the ensemble approach was 'robust', but scientists in monsoon regions would need to supplement it with real-time rainfall data to forecast river levels.
Bhupendra Nath Goswami, director of the Indian Institute of Tropical Meteorology (IITM), added that, as both weather and climate have a chaotic component, an ensemble approach is essential to reduce uncertainties during modelling. The Indian Meteorological Department, for example, uses an ensemble of 25 models and model variations for monsoon forecasts, but has yet to link its output to a model of river discharge.
Philosophical Transactions of the Royal Society doi: 10.1098/rsta.2011.0160