Modelling GLOF Susceptibility using Machine Learning
Our project, titled "Modelling GLOF susceptibility using Machine Learning", aims to predict the potential of any glacial lake to cause outburst floods. This will be accomplished using Machine Learning techniques, and a model trained on the glacial lakes in our study area which is the Hunza River Bas
2025-06-28 16:28:36 - Adil Khan
Modelling GLOF Susceptibility using Machine Learning
Project Area of Specialization Artificial IntelligenceProject SummaryOur project, titled "Modelling GLOF susceptibility using Machine Learning", aims to predict the potential of any glacial lake to cause outburst floods. This will be accomplished using Machine Learning techniques, and a model trained on the glacial lakes in our study area which is the Hunza River Basin. A few parameters have been selected which have a direct relation to the potential of outburst of a glacial lake, e.g. area, volume, distance from the glacier etc., and these parameters are used first in Analytical Hierarchy Process, and then in the machine learning model to predict susceptibilities. Thereafter, this model is deployed to a web a application on which a user fulfills a few requirements pertaining to a specific lake and the susceptibility for that glacial lake is calculated, and categorised as High/Medium/Low. Moreover, our project also consists of simulation of the flood extent that can potentially be caused by the most susceptible lakes in our study area of the Hunza River Basin. It also displays the spatiotemporal change of glacial lakes present in the region, between 2016 and 2021. This is an important analysis in the context of climate change. All of these objectives will be brought together and deployed on a web application, as a final result.
Project ObjectivesOur project comprises of four main objectives;
1. To detect potentially susceptible lakes employing machine learning techniques.
2. To identify spatio-temporal change detection of glacial lakes.
3. To assess glacial flood extent for highly susceptible lakes.
4.To develop web application for visualization of GLOF, flood extent, and deployment of the model.
Each of these objectives will be discussed in detail below.
(a) Detecting Potentially Susceptible Lakes
The first and foremost step in this project is identifying glacial lakes which have the potential to cause flooding due to any of the reasons that usually trigger GLOFs. For this objective, our project will employ Machine Learning techniques. However, prior to moving onto Machine Learning models, the prerequisite was the identification and extraction of the parameters we would base the susceptibility upon. After the selection and extraction of these parameters, a multi-criteria decision making method was necessary before the application of Machine Learning. Therefore, Analytical Hierarchy Process (AHP) was used. From this, we can characterise the susceptibility of the lakes as high, medium, or low.
(b) Identifying spatio-temporal change of glacial lakes.
Our project is aligned with two of the Sustainable Development Goals put forward by the United Nations; SDG#11: Sustainable Communities and Cities, and SDG#13: Climate Action. In the context of Climate Action, it is important to identify the spatio-temporal change of glacial lakes. The retreatment or advancement of glacial lakes is a key attribute in predicting the behaviour of those lakes, and in investigating climate change patterns.
(c) Assessing glacial flood extent for highly susceptible lakes.
Once the highly susceptible lakes are detected, it is important to know what could be the flood extent in case of an outburst. To fulfill that objective, HEC-RAS modelling is employed. The reason for using HEC-RAS is because according to our thorough literature review and research, HEC-RAS simulates the flooded area more realistically in terms of the ideal behaviour of flood dynamics.
(d) Developing a web application.
In order to combine all of the objectives together and to bring our developments in a practical mode, a web application is necessary. The web application aims to identify the susceptibility of any lake that the user enters the required parameters for, and then based on the Machine Learning model on the back-end, it characterizes the susceptibility as high/medium/low. Moreover, the web application will also display the flood extent for the most susceptible lakes in our study area, and spatio-temporal change as well.
The project begun with a thorough research and literature review of all of the work that has been done on Glacial Lake Outburst Floods in the past, all over the world. From our literature review we concluded that our project is unique, to the best of our knowledge, in the sense that using Machine Learning to predict GLOF susceptibility has not been accomplished before. Next, we digitised all of the glacial lakes in our study area using Google Earth Pro and Sentinel-2 imagery. These digitised lakes were used in the extraction of parameters which was done using various different methods. Then these parameters were used in Analytical Hierarchy Process and the results classified the lakes into high/medium/low susceptibility. After AHP, those parameters were used to train our Machine Learning model. The model used was Random Forest.
Our next objective is flood modelling. For this, our choice of software was HEC-RAS. This software allows the user to perform one-dimensional steady flow, one and two-dimensional unsteady flow calculations. For flood modelling several factors were taken into account like the slope, breach width, breach height, time of failure etc. The result was a simulation on our web application showing the extent of a flood in case of a breach.
For change detection, once the lakes were digitised, the lakes that were found to be present in both the years were then deployed onto our web interface. On the interface, using a slider tool the user is now able to observe the visual change between the lake as it existed in 2016, and then in 2021. This exercise gives the user a very clear idea of the current scenario of glacial lake rate of change, and the potential a certain lake has to become susceptible at some point in the future.
The need for research and development on reducing the risk of GLOF is absolutely essential, especially in the context of the threat of climate change. UNDP mentions that due to rising temperatures, glaciers in Pakistan’s northern mountain ranges (the Hindu Kush, Himalayas and Karakoram) are melting rapidly and a total of 3,044 glacial lakes have developed in Gilgit-Baltistan (GB) and Khyber Pakhtunkhwa (KP).
Our project aims to serve many factions of society. With our comprehensive approach that covers GLOFs, climate change patterns, and Sustainable Development Goals, we are highly hopeful that our project will benefit many causes and communities. The primary beneficiary will be the communities living in the HRB, and specifically those which are living in the areas highly susceptible to outbursts. Second, we aim to provide assistance to Disaster Management organizations which can use our model to find any susceptible lakes. Furthermore, NGOs which can prepare for relief operations, and awareness programs in case of flood, can also benefit from our project, and lastly, researchers can use our project as basis for future work on GLOF susceptibility or climate action.
All four of our objectives will be brought together and employed on a single web interface which is easy-to-use, beginner friendly, and interactive. The interface aims to show the flood extent simulations for the most susceptible lakes in our dataset, a map showing all of the glacial lakes in Hunza River Basin, and a portal which takes the shapefile of any lake you wish to find out the susceptibility of, and the website uses our model to predict its susceptibility. Moreover it has miscellaneous features like the “Contact Us”, “About Us” tabs, and basic information to familiarize users with the concept of GLOF. Technologies used for the development of the application include Django, Python, Javascript, HTML, CSS, Leaflet, R, Google Charts, API of OpenWeatherMap, TailwindCSS. The Home Page shows a title which summarizes the purpose of our website. If you scroll a bit down further, you will find a simulation and some text explaining the phenomenon of GLOF. And lastly, there are contact details available at the bottom. The Home Page is created using HTML, CSS, Javascript, and Python.
Then we have the GLOF Simulation tab which shows the simulation of all susceptible lakes. This was created using Leaflet. Moreover, each page consists of supporting information on the lake like description, hydrological data used, weather parameters, and a hydrograph which is displayed using Google Charts. Furthermore, the tab on which our model will be deployed will also be using Leaflet, JavaScript, Python, HTML, CSS.
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 900 | |||
| Thesis Printing | Miscellaneous | 3 | 300 | 900 |