Application of Machine Learning in Order to Improve Accuracy of Climate Prediction
Published Date | 2018 September 12
Researchers from University of California-Irvine, are shifting to the use of data science to achieve better cumulus calculating results for weather patternsUnited States: The research team of University of California have explored the application of deep machine learning to provide an objective, data-driven and efficient alternative that could be rapidly implemented into mainstream climate predictions. The application uses computer algorithm to mimic learning and thinking abilities of human mind wherein a deep neural network is trained to predict the results of thousands of small, two-dimensional, cloud-resolving models as they interacted with planetary-scale weather patterns in a fictitious ocean world. This will have a positive impact on the technological innovation in climate prediction model using machine learning. The Standard climate prediction model used for weather forecast is based on cloud physics that use simple numerical algorithms. It relies on imperfect assumption such as it indicates drizzle instead of more realistic rainfall and entirely missing other common weather patterns. The research team of University of California, thus, planned to explore the use of data science in climate prediction. They started the application by training a deep neural network as mentioned above. This newly taught program has dubbed "The Cloud Brain," that functioned freely in the climate model researchers and can lead to stable and accurate multiyear simulations including realistic precipitation extremes and tropical waves. As per Stephan Rasp, an LMU doctoral student in meteorology from the University, the neural network learned to approximately represent the fundamental physical constraints on the way clouds move heat and vapour around without being explicitly told to do so, and the work was done with a fraction of the processing power and time needed by the original cloud-modelling approach. The neural network is found to learn the simulation of cloud physics model within three months. Thus, it is expected to simulate a hundred days of global atmosphere, if trained with realistic geography and in trickier model-setups. The research followed the application of machine learning in the field physics, biology and chemistry. The application of machine learning on climate science is heavily centred on large data sets, especially these days when new types of global models are beginning to resolve actual clouds and turbulence. According to BlueWeave Consulting Research, the clouds that play major role in the Earth's climate by transporting moisture and heat, absorbing and reflecting the sun's rays and by trapping infrared heat rays to produce precipitation. However, these clouds can be as small as a few hundred meters, much smaller than a standard climate model grid resolution of 50 to 100 kilometres. Thus, simulating them appropriately will require an enormous amount of computer power and time. The use of deep machine learning can resolve this problem because it has the ability to develop a neural network that can predict realistic precipitation extremes and heat waves, providing more accurate weather forecast. Thus, the application of deep machine learning has huge market potential in order to change the field of technology innovation.