This dataset contains input data (20 storm events) for developing a Physics-Informed LSTM (PI LSTM) surrogate model to forecast street-scale nuisance flooding in Norfolk, Virginia, USA. The mass balance (MB) equation is incorporated into the customized loss function (L_total) of the LSTM surrogate model, which combines both data loss (L_data) and physics loss (L_phy). The input features used for the data loss were hourly rainfall, hourly tide level, elevation, Topographic Wetness Index (TWI), and Depth-to-Water (DTW). The input features used for the physics loss were hourly inflow volume, hourly outflow volume, hourly rainfall volume, and hourly pipe flow volume. The model included two target features – flood depth and flood volume. The water depth raster was collected from the TUFLOW model through a coupled 1D/2D simulation for each hour throughout all storm events. The water volume was calculated by summing the hourly water depths over the inundated area within the street segment using the zonal statistics tool of ArcGIS Pro.

There are five files in this resource -

1. The "street_shapefiles.zip" folder includes a shapefile of the street segments (polygons of 50 m length x 7.2 m width) of Norfolk. Alongside, it includes a "D0_R40_S22.csv" file containing 22 flood-prone streets selected from the STORM report.

2. The "vol_wd_22_rh_q.zip" folder includes the input CSV files for the top 20 daily storm events from 2016-2018 for the streets of Norfolk.

3. The "relational_database.zip" folder includes three CSV files for node_data (varied spatially), tide_data (varied temporally), and weather_data (varied spatially and temporally) for efficient data management.

4. The Python script "create_relational_data.py" is used to convert "vol_wd_22_rh_q" CSvs to "relational_database" CSVs.

5. The "variable_description.csv" file describes the input variables used in the "vol_wd_22_rh_q.zip" and "relational_database.zip" CSV files, along with their units.

The Python script of the PI LSTM model is available on GitHub https://github.com/br3xk/A-Physics-Informed-LSTM-Surrogate-Model-for-Real-time-Flood-Forecasting-at-the-Street-scale