Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
ConvLSTM: Mobile Wildfire Prediction
This talk demonstrates using ConvLSTM models to predict wildfire spread from crowd-sourced mobile location data, including data preprocessing, model details, and dynamic heatmap visualizations.
Wildfires are a growing threat, with devastating consequences for communities and ecosystems. Inspired by recent events in LA, I’m working on a system that uses ConvLSTM (Convolutional Long Short-Term Memory) to predict wildfire spread based on crowd-sourced mobile device location data.
The demo will showcase a live walkthrough of the model in action. I’ll start by showing how location pings and time-series data from nearby users are preprocessed into spatial-temporal inputs. Then, I’ll dive into the ConvLSTM code, explaining how it processes this data to generate wildfire likelihood heatmaps over time. To bring the predictions to life, I’ll visualize the output heatmaps as dynamic animations, representing how a wildfire might spread in real-time.
I’ll also briefly highlight how external factors like wind direction, temperature, and vegetation type can be incorporated using public APIs and how they influence predictions. This isn’t a polished product, but a tinkerer’s attempt to explore how accessible data and open-source AI tools can solve real-world problems!
Kaggle script implements a predictive algorithm for wildfire risk assessment.