I recently came across a tool/twitter bot that uses machine learning to process social media data and detect early signs of wildfires (https://twitter.com/CalFireSafe). This is amazing, given that people are often on social media during natural disasters. The tool can identify patterns in social media data that are indicative of wildfires, such as posts about smoke or flames, or posts from people who are evacuating their homes. Didn’t come across on anything similar before… Would like to know your thoughts on this.
1 Like
my gut thought is that by the time someone is evacuating the people who need to know already know. Same with “I see smoke” posts and tweets… hopefully they’ve called that in before posting… lol.
1 Like
i think the advantage of this bot is its ability to collect and analyze data simultaneously, enabling fast alerts, so the goal the is not just to repeat what people are already saying but to provide crucial early warnings. by detecting patterns indicative of potential wildfires, these alerts can help inform communities before a fire gains momentum or before an official evacuation order is issued…
1 Like
There is value to this i think in a geospatial component as well but we need fusion/intel
Centers to process this data to better inform ops.
2 Likes
that aspect sounds interesting.
you know what? I think I’m spoiled here in CA because we have the cameras and a huge chunk of departments using wildcad and nifc so we really do know about fires fast whereas in other states it’s not the same. so I should modify or rescind my earlier comment.
FireWhat? tried doing something similar to this about ten years ago, showing geotagged photos from #FireName hashtagged posts on a web map. It seemed like it was a lot of blurry smoke column photos and not much useful intel. Garbage in, garbage out. One big problem with Fire Twitter is a lot of people all retweeting the same basic post, so you get buried in a pile of the same stale or marginally useful info.
This is 3 out of 4 tweets on my timeline right now.
5 Likes
Yeah, that’s pretty interesting! I actually checked out their detection model documentation, and I must say, they’ve made some significant advancements, like excluding retweets from the data processing. This change helps prevent the system from getting bogged down with repetitive or not-so-helpful info.