/hybridmodels

(In progress) scraping remote sensing data from Google Earth Engine and various sources, processing into 32x32 tiles for training a spatiotemporal random forest flood prediction model. Also beta code for CNNs using Tensorflow.

Primary LanguageJupyter Notebook

Hybrid Coastal Flood Models

Work in progress, developing code to:

  1. Scrape coastal flood-related remote-sensing data from Google Earth Engine and other sources
  2. Process into a dataset of 32 x 32 km grid squares
  3. Train machine learning models to predict flood maps from the data

Contents

  • Python script get_data_parallel.py with code to generate dataset.
    Requirements:
    1. CSV current_datasets.csv and event_dates.csv: csvs containing storm name, region name, number of subregions (32x32 grid squares) for each region, storm start date, end date, landfall date (from IBTrACs data), satellite acquisition date of the flood map, and a geojson string of the subregion polygons. These should be in ./hybridmodels/data/csv directory.
    2. a folder with name following the format <storm name>_<region> and containing a flood polygon (flood.gpkg) and area of interest (areaOfInterest.gpkg) polygon in GeoPackage format. These should be in ./hybridmodels/data/<storm name>_<region> directory.
    3. The hybridmodels conda environment (create from hybridmodels.yml contained in envs folder).
    4. Googel Earth Engine Service account with gcloud key saved in hidden folder in the directory (not uploaded here).