Small-grid urban flood prediction model using Twitter data and population GPS data - an example of the 2019 Nagano city flood

Abstract

In this study, a small-grid urban flood prediction model integrating Twitter data and population GPS data was constructed using the 2019 Nagano City flood as an example, and the validity of these two data for the model was determined. Using natural language processing techniques, Twitter data was filtered to extract real-time information relevant to flooding. At the same time, geographic information processing techniques were applied to analyze the population GPS data and obtain the distribution of the local population. Based on these two types of data, we combined with terrain, land use, traffic and infrastructure data related to flooding, a real-time flood prediction model was constructed using the random forest algorithm with a basic unit of a 70 m × 70 m grid. An analysis of the model accuracy showed that, the model that included both GPS and Twitter data showed an improvement in prediction accuracy of about 8% compared to flood prediction models that do not have these data sources. This indicated that the integrated use of Twitter and GPS data allowed us for a more accurate representation of the dynamic characteristics of flood disasters, thereby improving the performance of real-time flood prediction models and increasing real-time awareness of flood events. This approach provided effective flood monitoring methods for disaster management authorities.

Keywords

• Flood
• Prediction model
• Twitter data
• GPS data
• Random forests