Satellite video remote sensing for flood model validation - Research Data

This repository contains data and scripts for an investigation that used satellite video remote sensing to validate a hydraulic model of the Darling River at Tilpa, Murray-Darling basin, Australia. It encompasses a range of materials, including satellite video footage of a flood (acquired 5 February 2022, 23:12 UTC), data and results from HEC-RAS 2D model simulations of the flood, as well as flood extents derived through the use of deep learning techniques. The Python script used for deep learning is included (for practical considerations, users will need Python 3.x and PyTorch 2.x, alongside other specific libraries, for running the deep learning script, details of which are described in the script itself).


Contact: 
christopher.masafu@glasgow.ac.uk; richard.williams@glasgow.ac.uk; 

This data directory contains six root folders and one readme text file:

HEC-RAS Model Data:
===================

Data to run HEC-RAS 2D simulations in three sub-directories; 

1. Digitial Elevation Model: 
Tilpa_Digital_Elevation_Model.tif
Source: Geoscience Australia (https://elevation.fsdf.org.au/)
Date: 15 Feburary, 2022

2. Geometry: Model geometry files

3. Unsteady Flow Data: Tilpa Gauge Data (Source: https://realtimedata.waternsw.com.au/, Date: 1 March 2022) and Uncertain Streamflow Data


HEC-RAS Model Outputs:
======================

Modelled flood extent rasters for;

1. Model Mobs Flood Extent

2. Model Mqmaxpost Flood Extent


Deep Learning Flood Extents:
============================

Deep-learning derived flood extents used to validate HEC-RAS model predictions in two folders, for the two validation locations (A & B):

 	Deep Learning Flood Extents
	├── A
	│   └── TTA (Test Time Augmentation)
	│   └── No TTA (No Test Time Augmentation)
	├── B
	│   └── TTA (Test Time Augmentation)
	│   └── No TTA (No Test Time Augmentation)
     

Deep Learning Images & Masks: 
======================================================

Images (images) and CVAT annotated masks (masks) used in model training and validation (no augmentation applied)


Satellite Video (Acquired 5 February 2022, 23:12 UTC):
======================================================

Jilin 01 GF03C02 05022022.mp4: RGB satellite video
Ground Sampling Distance: 1.22 m


Segformer Encoder_ U-Net Decoder _ Segmentation Code: 
=====================================================

Python code used for deep-learning semantic segmentation 




*All geospatial data is in the GDA 1994 MGA Zone 55 projection