Cloud formation and parameterization (Voosen, 2019); T. SCHNEIDER ET AL., GEOPHYSICAL RESEARCH LETTERS 44, 12,396 (2017), ADAPTED BY N. DESAI/SCIENCE
Clouds
The clouds in the story had been associated with many things. It is often seen as soft, fluffy, and harmless object floating around in the sky. It provides many imaginations to the children, world-wide, while also providing life-giving water to the plant. It is difficult, therefore, to imagine clouds as the most difficult and confounding problem in scientific world of weather prediction (Gentine, et al, 2018). Clouds play a significant role in predicting rain falls, and not being able to provide accurate prediction results in inaccurate prediction of the location and amount of rainfall (Voosen, 2019).
Parameterization
To better understand the cloud problem, we need to discuss about parameterization in weather prediction. To process data into manageable set of information, the globe is sectioned off into grids, and each grid is abstracted into a set of numbers called parameters (Voosen, 2019). Smaller grids create high-resolution information while larger grids create coarse-resolution information (Voosen, 2019). It is not too difficult to understand that higher resolution provides more accurate information at the cost of greater number of grids and therefore computation challenge while the lower resolution information provides more feasible computational requirement at the cost of accuracy. The cloud becomes a problem because the cloud formation often happens within a single grid, rendering prediction of location and amount difficult to make due to the abstraction of the information (Voosen, 2019). This problem can be solved by using higher resolution information, but the cost of computation makes it difficult.
Proposed Solution
Professor Gentine and his colleagues decided to take advantage of the accuracy of high-resolution information while using the computationally-feasible coarse-resolution information. The team took advantage of the machine-learning technology that is relatively more inexpensive than observing the high-resolution information. First, they trained a machine-learning model, called Cloud Brain (CBRAIN), to predict cloud-formation using specialized high-resolution information to provide similar level of precision and accuracy. The team then used CBRAIN on the coarse-resolution model to predict cloud formation. This technique provided a powerful method to supplement the current technology by reducing cost and increasing scalability that ultimately increased accuracy. This is an incredible breakthrough as scientists are now equipped with a potential to resolve a challenge that was once through impossible to solve.
References
Evarts, H. (2018, June 19). Machine Learning May Be a Game-Changer for Climate Prediction. Retrieved May 26, 2019, from https://engineering.columbia.edu/press-releases/machine-learning-climate-prediction
Gentine, P., Pritchard, M., Rasp, S., Reinaudi, G., & Yacalis, G. ( 2018). Could machine learning break the convection parameterization deadlock? Geophysical Research Letters, 45, 5742– 5751. https://doi.org/10.1029/2018GL078202
Voosen, P. (2019, May 22). Science insurgents plot a climate model driven by artificial intelligence. Retrieved May 26, 2019, from https://www.sciencemag.org/news/2018/07/science-insurgents-plot-climate-model-driven-artificial-intelligence
AI for Earth In the Microsoft blog called “Researchers turn to AI in a bid to improve weather forecasts” by Roach, the author highlights some of the company's contribution to the research that benefits mankind. Through the AI for Earth initiative, the company funds various challenging research projects that studies Earth. The blog also highlights the social activities, such as hackathons, that illustrates some of the current trends of using machine learning to solve difficult problems. References Roach, J. (2019, May 20). Researchers turn to AI in a bid to improve weather forecasts. Retrieved May 26, 2019, from https://blogs.microsoft.com/ai/ai-subseasonal-weather-forecast/
About In this 2018 paper, “Predicting weather forecast uncertainty with machine learning” by professor Scher and professor Messori, shows an early effort to test the feasibility of the technology as a potential replacement to the popular ensemble weather model approach. The ensemble weather model approach, as the name suggest, is similar to having a panel of experts instead of just one expert; the approach produces information that is a composite of different information generated by multiple models given the same input data. Each model caters to different specialization and has their own strength and weaknesses, and by combining information from each model, the approach attempts to generate more accurate information. As such, while it increases the accuracy and precision, it became computationally very expensive to perform in a timely manner. The researchers’ goal was to provide a competitive solution to the computationally expensive ensemble weather model approach by using machine ...
Background: Data Science Before talking about “AI for Earth observation and numerical weather prediction” by Sid-Ahmed Boukabara, I need to talk a little bit about data science. One of interesting distinction in the data science world is the separation of data and information: the former refers to the raw facts without context while the latter is a processed data to give it a meaning. In another words, raw data is not very useful in that state until it is distilled into information. This distinction formed the foundation of the article by Boukabara, the acting deputy director of the NOAA NESDIS Center for Satellite Applications and Research. Bottleneck Boukabara’s article describes a lop-sided advancement in technology: the human infrastructure to collect data, such as an array of satellites, has far surpassed our ability to process the data in a timely manner. This imbalance is so huge that “only 3-5 percent of satellite observations are actually used in preparing numerical we...
Comments
Post a Comment