Automated Detection of Hazards
Shelter. – Organized Habitation Planning Software
Summary
Shelter. is an AI based prediction and analytic decision making app that helps national decision makers distribute habitation and slums. Unplanned habitation is an underrated destroyer leaving 4.8 million people homeless annually and thousands dead. Shelter will focus on landslides, river erosion, floods and earthquakes. Here we have created an algorithm on landslide susceptibility prediction, analysis and visualization that can be automatically put against datasets like population density to determine intersecting regions and make responsible decisions based on it. The algorithm was based on instances, precipitations, temperature, triggers etc. and is convertible into an API for Shelter.
How We Addressed This Challenge
What we aim to build:
Our activity surrounds Shelter. an application that aids in proper long term urbanization, slum distribution and community establishment by monitoring potential hazards on the fields of landslides, floods, earthquakes and river erosion. The application will:
- Visualize hazard and habitation intersections.
- Create simplistic representation of complex datasets
- Advices on potential changes to be done in establishments distribution
What we built:
As all 4 disaster prediction and analysis algorithm follow same development procedure, we developed one complete segment i.e. LTP (Landslides Tracker and Predictor) powered with advAIsor. We analyzed and built learning model on instances from NASA's Global Landslide Catalogue, precipitation from NASA's Global Precipitation Measurement and Tropical Rainfall Measuring Mission, temperature and contents from NASA's Technical Report Server and so on. We also worked on NASA's observations on Mongolia's wind and soil statistics and the observations of 1700 landslides. We built a latitude-longitude based prediction and tracking algorithm that is convertible to API and we visualized it in a 3D model.
Why we built it:
Currently about 150 million or 2% population is homeless causing disrupt in economy. 37% of homeless individuals reported to have lost their households in disasters of which the most notable is landslide in Europe and America and river erosion in Asia. That's 55 million people with shattered rights. In our best case scenario even if the government successfully rehabilitate them, there remains equal risk of being prey to landslides, erosion or floods. That is why we have chosen to tackle these situations with advanced machine learning and satellite data. We believe that through algorithmic power, we can ensure safer shelter distribution and thus arrived the name 'Shelter.'
Main features of LTP:
- Landslide susceptibility prediction with an estimated over 90% accuracy
- Near Real Time Analysis and Trigger Monitoring
- Accessibility to lay secondary data over the algorithm
- Simplistic visualization of holistic product
How We Developed This Project
Motivation:
Our country Bangladesh just faced simultaneous floods and river erosion causing hundreds of people to lose their households and end up in streets. Issues like landslides are also quite high down to southern regions. While we dug deep in the situations, we found that unplanned community distribution is the bane causing these devastating effects. While we took a look at Global Landslide Catalogue by NASA, a passion struck to use such data to build an algorithm to help in the proper distribution of habitations. Perhaps in our foreseeable future these data can be the basis of saving millions of lives and houses.
Steps of development:
1) We started with brainstorming and creating a simple Ux/UI on 75 disasters.
2) As data for each would set a massive reduction on accuracy we decided to work on major 4
3) We decided to show one algorithm and chose landslide
4) Our initial expectation was that we would collect data from the resources at https://github.com/nasa/spaceapps-phenomena_detection/ but we couldn't find relevant information.
5) Then we depended on GLC and analyzed every triggers and reportings
6) We worked on 1700 datasets from 2007-2016 for python based analysis
7) To increase our accuracy we used resources like GPM working on precipitations
8) For sub-setting we used Giovanni and labelled many data from Image Labeller by NASA
9) We used raw python coding with some libraries to provide these 1700 samples into a supervised machine learning algorithm.
10) We overlay the data on static datasets.
Tools used:
1) Numpy library
2) Panda library
3) TensorFlow
4) Global Landslide Catalogue
5) Giovanni
6) Image Labeller
7) Jupyter Notebook
8) Microsoft Excel
How We Used Space Agency Data in This Project
Landslide Instances: https://data.nasa.gov/Earth-Science/Global-Landslide-Catalog/h9d8-neg4
Landslide Fatalities:https://maps.nccs.nasa.gov/arcgis/apps/webappviewer/index.html?id=824ea5864ec8423fb985b33ee6bc05b7
Landslide Trigger Calculations: https://www.nasa.gov/mission_pages/GPM/main/index.html and https://trmm.gsfc.nasa.gov/
Sub-setting data: https://giovanni.gsfc.nasa.gov/giovanni/
Labelling data: https://labeler.nasa-impact.net/
Project Demo
The demo is available here https://youtu.be/361i9nGeXf8
Project Code
Data & Resources
1) https://www.nasa.gov/mission_pages/GPM/main/index.html
2)https://giovanni.gsfc.nasa.gov/giovanni/
3)https://data.nasa.gov/Earth-Science/Global-Landslide-Catalog-Export/dd9e-wu2v
4) https://drive.google.com/file/d/1f5Hd8umBntvGHYK0LBSG68xLqMuiY2bg/view?usp=sharing
(The excel file was exported via NASA's Landslide Archives)