Harnessing Twitter (X) as a proxy to assess hail impact.
SONAR|HES-SO
- Covaleda Vélez, Carlos Felipe
- Tavares, Erisvelton
- Mumenthaler, Christian (Degree supervisor)
- Genève : Haute école de gestion de Genève
52 p.
English
There are numerous scientific papers on the harnessing of microblogging platforms in the study of natural hazards. However, there are no studies dedicated to hail. This is surprising given the frequency of hailstorms, their impact on the population and agricultural activities, and the severe damage yearly quantified. The present study aims to fill this research gap. Based on a collection of tweets from 2008 to 2023 tagged with "#hail" and a database of hail observations in the United States for the same period, we undertook a model training to predict the impact of hail events from tweets. We were able to confirm a correlation between the posting of tweets and the occurrence of hail events. As a result, we estimate the posting of about 4 to 6 tweets referring to the hail event by increasing the hail size by 1 inch. Its size is equivalent to a quarter-dollar coin.
For this purpose, a method was developed to match tweets with hailstorms using a spatiotemporal buffer. This step was crucial since the meteorological database did not contain hail events but only punctual observations. The outcome of this process consisted of a dataset grouping hail events and tweets in a spatio-temporal dimension, as well as various metrics. This dataset was used to train and test a supervised learning model based on GLM and Random Forest to predict the impact caused by hail.
Although the available data was insufficient to train an accurate prediction model, as the predictions were not as statistically significant, prospects are open for using microblogging networks as a proxy to assess the impact of hail. We consider necessary to obtain more input data to feed an accurate model. It could worth to enrich “magnitude” and “damage” related data, in order to better characterize hail impact.
For this purpose, a method was developed to match tweets with hailstorms using a spatiotemporal buffer. This step was crucial since the meteorological database did not contain hail events but only punctual observations. The outcome of this process consisted of a dataset grouping hail events and tweets in a spatio-temporal dimension, as well as various metrics. This dataset was used to train and test a supervised learning model based on GLM and Random Forest to predict the impact caused by hail.
Although the available data was insufficient to train an accurate prediction model, as the predictions were not as statistically significant, prospects are open for using microblogging networks as a proxy to assess the impact of hail. We consider necessary to obtain more input data to feed an accurate model. It could worth to enrich “magnitude” and “damage” related data, in order to better characterize hail impact.
- Language
-
- English
- Classification
- Information, communication and media sciences
- Notes
-
- Haute école de gestion Genève
- Information documentaire
- hesso:hegge
- Persistent URL
- https://folia.unifr.ch/global/documents/330946
Statistics
Document views: 23
File downloads:
- COVALEDA_TAVARES_projet_recherche_2024.pdf: 45