Abstract

Early event detection and response can significantly reduce the societal impact of floods. Currently, early warning systems rely on gauges, radar data, models and informal local sources. However, the scope and reliability of these systems are limited. Recently, the use of social media for detecting disasters has shown promising results, especially for earthquakes. Here, we present a new database for detecting floods in real-time on a global scale using Twitter. The method was developed using 88 million tweets, from which we derived over 10.000 flood events (i.e., flooding occurring in a country or first order administrative subdivision) across 176 countries in 11 languages in just over four years. Using strict parameters, validation shows that approximately 90% of the events were correctly detected. In countries where the first official language is included, our algorithm detected 63% of events in NatCatSERVICE disaster database at admin 1 level. Moreover, a large number of flood events not included in NatCatSERVICE are detected. All results are publicly available on www.globalfloodmonitor.org.

Cite as

Bruijn, J.A., Moel, H., Jongman, B. et al. A global database of historic and real-time flood events based on social media. Sci Data 6, 311 (2019) doi:10.1038/s41597-019-0326-9

Links

• Flood Tweet IDs (multilingual)
• Historic flood event database
• Real-time flood event database

How to run

1. Setup
   • Install Python (3.6+) and all modules in requirements.txt.
   • Install PostgreSQL (tested with 12) and PostGIS (tested with 3.0).
   • Set all parameters in config.py. This includes the TWITTER_CONSUMER_KEY, TWITTER_CONSUMER_SECRET, TWITTER_ACCESS_TOKEN, TWITTER_ACCESS_TOKEN_SECRET for which you will need to register as Twitter developer.
2. Preparing data & preprocessing
   • Obtain a high-resolution population raster (e.g., LandScan Global), convert to GeoTIFF (e.g., gdal_translate w001001.adf population.tif -co "COMPRESS=LZW") and place in input/maps/population.tif.
   • Set all parameters in config.py
   • Create elasticsearch index for tweets using create_index.py. This file automatically uses the proper index settings (see input/es_document_index_settings.json).
   • Fill index with tweets (example for reading tweets from jsonlines to database in fill_es.py). This assumes the file input/example.jsonl has a new json-object obtained from the Twitter API on each line.
   • Run preprocessing.py
3. Creating the text classifier
   • Hydrate the labelled data (input/labeled_tweets.xlsx) by running hydrate.py. This creates a new file with additional data obtained from the Twitter API (including the tweets' texts in input/labeld_tweets_hydrated.xlsx). Don't forget to set the Twitter developer tokens in config.py
   • Train the classifier by running train_text_classifier.py. This file exports the trained classifier to input/classifier.
4. Finding time corrections per region
   • In the next step we need to run just the localization algorithm TAGGS so that we can derive the number of localized tweets per hour of the day (see paper). To do so we run the main file main.py, with detection set to false, like so: main.py --detection false
   • Run get_time_correction.py. This will create a new file input/time_correction.json.
5. Run the Global Flood Monitor
   • Finally, run main.py without arguments to run the Global Flood Monitor. The resulting events are stored in the PostgreSQL database.

Contact

Jens de Bruijn -- j.a.debruijn at outlook dot com