Two weeks ago, I had a presentation at TGA 2022 (Taiwan Geoscience Assembly) & Conference on Weather Analysis and Forecasting. The title was “Use Machine Learning to Classify Weather Opinions from WhatsApp Group: A Model Design and Evaluation for Raining Event Prediction”. The term “Weather Opinion” is created for this research since the voice from social media is larger and larger. Not only weather forecasting offices offer weather reports, but also people who are interested in weather forecasting are willing to post their owned weather forecasting online.
Different perspectives to the same weather situation
The posts on social media may have different perspectives to the same weather situation. For example, during Asia monsoon season transition, for the same rainfall events, some think they are caused by stranded fronts, other think they are large storm cells, and the others might think they are from thunderstorms. The different definitions can lead different forecasting, so the classification is important during the early analysis.
Weather Opinion as observations
In some places with less weather research and observation resources, the human decision making results can be a type of observation information. The opinion from same people may have a consistent view to read the weather situations though there might be bias and error. After filtering out the errors, this information can supplement the lack of weather research and observations.
Predict Weather Opinion
This research is trying to use machine learning technology to predict the weather opinion. We didn’t analyze how the human reporters made the decision to classify the weather types, neither analyze the weather situation and tune the weather forecasting model. It just selected features with domain knowledge to apply machine learning for weather type classification.
The writing paper:
Pou-Chang (John) Chen 