|
Exercise 10-1: Classifying Hurricane Damage Imagery with Machine Learning |
This is the companion website for Exercise 10-1: Classifying Hurricane Damage Imagery with Machine Learning from the book Geographic Information Systems (GIS) for Disaster Management (second edition). This page contains video walkthroughs of the exercise, datasets associated with the exercise, and other additional materials.
Follow this link to access Exercise 10-1: Classifying Hurricane Damage Imagery with Machine Learning datasets:
Exercise 10-1: Classifying Hurricane Damage Imagery with Machine Learning
Skill Level: intermediate to advanced.
Estimated time to complete: 1-2 hours
Additional resources needed: an Internet-connected computer, a zip tool, ArcGIS Pro 2.x (with Spatial Analyst and Image Analyst licenses), Exercise 10-1 datasets you can download here.
Purpose: the purpose of this lab will be for you gain experience and knowledge with how to use and evaluate a machine-learning algorithm (Support Vector Machine [SVM]) to classify imagery in a disaster-response context.
Additional notes:
This exercise will only work with Esri’s ArcGIS Pro 2.x. To obtain a 21-day free evaluation of ArcGIS Pro, see https://www.esri.com/en-us/arcgis/products/arcgis-pro/trial
Datasets for this exercise were downloaded from the NOAA Hurricane DORIAN Imagery website. An area near Marsh Harbor on Abaco Island in the Bahamas was selected as a representative example of the massive damage that was incurred on Abaco Island in terms of damaged trees, blue tarp roof cover (used to cover damaged roofs), destroyed building, and intact roofs (or roof tops that have not been damaged). A sample of imagery was extracted from a larger NOAA Hurricane DORIAN imagery tile for ease of computational processing in this lesson. The full tile where the sample imagery was extracted from is also included with the lesson files. A full complete NOAA Hurricane Dorian imagery dataset can be downloaded from the website above and consists of about 10 GB of TIFF and JPEG files. These datasets are an excellent example of post-disaster imagery.
A new classification schema was created in the ArcGIS Training Samples Manager that defined analysis classes for the imagery. This schema included classes such as damaged trees, blue tarp covers, intact roofs, destroyed buildings, and open road ways. These classes are by no means representative of all the types of imagery analysis that could be conducted on the devastating effects of Hurricane Dorian, and they are meant simply to be a starting point that can be expanded upon as a learning exercise and imagery analysis for disaster response. Training sample polygons were then created from the classification schema. The training samples were saved as a polygon shapefile along with the classification schema, both of which have been included with the lesson files that can be downloaded here as a comparison to the datasets that will be created via specific training steps in this lesson.
Learning Objectives: after completing the exercise, you will know:
- how to create training samples to classify imagery with the Support Vector Machine algorithm;
- how to assess the results of a machine-learning algorithm by creating and modifying accuracy assessment points to “ground truth” and compare the results of a machine-learning algorithm against human judgement;
- how to assess the results of a machine-learning algorithm by creating and interpreting a confusion matrix that quantifies the results of comparing a machine-learning algorithm classification with a human classification.
Deliverables: if you are completing this exercise for a class assignment, you can submit a screenshot of your response to the instruction questions as well as answers to the discussion questions.
Overview:
Hurricane Dorian, which occurred in September of 2019, was considered one of the most devastating hurricanes to hit the Bahamas since record-keeping began . Abaco Island, located in the eastern part of the Bahamas, particularly received devastating impacts from the hurricane with over 35 people reported killed because of the hurricane. Satellite imagery was used to conduct analysis of the hurricane’s devastating impacts and shows in stark detail the impacts of Dorian .In this exercise, you will work with actual imagery that was collected from Abaco Island by NOAA as an exercise in learning how to use machine-learning tools to classify satellite imagery in a disaster management context.
For detailed exercise instructions, purchase the book from Routledge Press »
Exercise 10-1: Classifying Hurricane Damage Imagery with Machine Learning
ArcGIS Pro Walkthrough (Part 1 of 2)
ArcGIS Pro Walkthrough (Part 2 of 2)