Better very short term forecasts for old and new needs

点击翻译按钮获取中文摘要

While weather forecasting in general is showing steady progress, the timeliness and quality of warnings for severe summer weather has not markedly improved in the past two decades. Given the current major renewal of Canada's radar infrastructure and the better availability of forecasts from weather prediction models and of other data such as satellite imagery, we want to explore how to improve warnings of summer severe weather threats and understand weather phenomena better in the process. Warnings can be issued either based on the detection of signatures associated with present or future severe weather by radar and other observations or based on predictions by a weather forecasting model. Our past efforts working on the latter have revealed fundamental roadblocks in our current and future ability to use radar data effectively to help determine the current conditions of the atmosphere, a prerequisite to any forecast. Our focus will hence be on how to improve the use of information from measurements to better detect and predict severe weather threats. We also seek to expand this approach to predictions of wind and cloudiness for new applications such as the short-term prediction of wind and solar energy production. Specifically, we want to exploit the increasingly large archives from radar, satellite, weather prediction models, and storm damage occurrence to discover new combinations of signatures that are associated with existing or near-future severe weather threats such as hail, violent winds, torrential rain, and tornadoes, and understand their bases. The search for these new signatures will be helped by techniques rooted in artificial intelligence, though we ultimately seek to learn to recognize threats and understand the bases of that detection, not simply train an algorithm to do the work without having a good idea of what it learned to look for. We not only want to recognize current severe weather, but also patterns associated with severe weather in the near future, as having such knowledge will speed up the issuance of weather warnings. We first intend to develop better pattern recognition techniques using different data sources that also take the quality of the imagery into account to better recognize severe storm patterns we are already familiar with. We then want to use artificial intelligence to find other patterns that are also associated with current or future severe weather threats in different situations. The results of this work will be used by meteorologists and researchers at Environment and Climate Change Canada to develop better algorithms to detect severe weather threats and provide more accurate and timely warnings to improve the safety of Canadians.