121st Data Assimilation and Prediction Science Seminar

Title: "Exploring Live Cameras for Actionable Weather Prediction: Recent Studies with Chiba University’s BAECAST Network"
Abstract: Numerical weather prediction has advanced substantially through improvements in observing technologies, data assimilation methods, and numerical weather prediction (NWP) models. More recently, artificial intelligence (AI) has increasingly been used to develop surrogates for NWP models and to advance nonlinear and non-Gaussian data assimilation. These developments are expected to enable more effective use of large volumes of observational data, including satellite observations. To further improve weather forecast accuracy, it is important not only to refine data assimilation methods and NWP models but also to explore new sources of observations. Our research focuses on the vast amounts of data generated by live-streaming cameras already deployed in everyday settings. Many of these cameras capture aspects of weather conditions, and harnessing this information could contribute to further improvements in weather prediction. This talk will present several studies on weather forecasting and diagnostics using live-streaming cameras installed at Chiba University. I will introduce BAECAST, an approach to visual weather forecasting, and discuss which meteorological variables can be retrieved from live-camera imagery. I will also present our efforts to transform the nature of weather forecasts using live-camera data using a conditional diffusion model.