Description:

Nowcasting and Short-Term Forecasting Applications Leveraging Artificial Intelligence, part 1:

Stakeholders across the public and private sectors are increasingly demanding short-term forecasts and nowcasts for weather events spanning a wide range of impacts, from near-term precipitation assessments to multi-hour severe weather risk outlooks. Traditional modeling approaches leveraging pure numerical weather prediction can struggle in this application owing to their computational costs, limitations in their underlying parameterizations, and difficulties assimilating rapidly updating observations or non-traditional/crowd-sourced observations. However, novel methods using techniques such as deep learning are increasingly showing success over these traditional approaches with applications ranging from:

1. Probabilistic severe weather forecasts based on ensembles of convection-permitting mesoscale simulations,
2. Quantitative precipitation nowcasting based on radar, satellite, and other remote sensing datasets,
3. Forecasting development of convection, icing, or other aviation hazards with the aid of numerical model and satellite remote sensing data.

This session highlights recent successes in employing AI to create specialized or novel nowcasts and short-term forecasts with applications across the meteorology community. Presenters are additionally encouraged to discuss challenges regarding communicating these new types of forecast products to stakeholders who may be encountering AI applications for the first time, as well as the limitations and trade-offs that these forecast products may have compared to their traditional counterparts.

Who Determines the Forecast? - The Changing Mix Between Human and Technology:

The quality of NWP and post-processed forecasts continues to improve.  At the same time there is increasing demand for meteorologists to provide decision support which allows the community to derive greater benefit from weather forecast information.  Given these two trends, many weather organisations are considering how they can streamline the forecast production process to rely more on automated forecast information and require less intervention from meteorologists, thus freeing them to add value in other ways.

Submissions are called on to explore different perspectives on this topic, such as:

• what verification reveals about where the meteorologist can add the most value in the weather enterprise
• tools and approaches for optimizing the human/machine mix in forecast production
• risk and responsibility: to what extent do automated forecasts need to be supervised?
• challenges posed by extremes, warnings and high-impact weather
• emerging opportunities and new roles for meteorologists opened up by technological advances
• cultural factors affecting acceptance of forecast production with less human intervention
• lessons from recent developments in the private and public sectors and in the international community.
• ML can be used to make better forecasts, but do those forecasts lead to better decisions?  
• As meteorologists adopt more and more ML, how do we prevent their skills/intuition from atrophying?

• How do meteorologists know when to trust an ML algorithm vs. when to override it with their own expertise?  

• Is it important to show meteorologists how/why an ML algorithm arrives at a certain decision or prediction?  etc.