O17-2020: Team A- Landslide Prediction
Landslides are a damaging natural hazard because of their difficulty in forecasting, and fast onset that limits emergency responses. We aim to model a spatial variable that helps in predicting the chances of occurrence of such an event. We also propose to develop a multi-platform application aimed to crowdsource the process of data accumulation of any such event and update our model based on the procured data.
Challenge 2- Landslide Prediction
Landslides are a damaging natural hazard because of their difficulty in forecasting, and fast onset that limits emergency responses. We propose to declare a new variable - 'cascade' as a probabilistic parameter of landslide occurrence. We will train the model on data procured from NDMA of India, local newspapers and television reports. We will develop a multi-platform application to collect landslide reports from users with all the relevant geographical information. The data will be modelled using AI techniques such as Deep Neural Networks, Expectation Maximisation, and Maximum Likelihood Estimation. Input variables will consist of a large number of geomorphological and climatological parameters such as soil type, precipitation, slope, topographic curvature, distance from drainage, lithology, which will then be used to extend and regionalise the 'cascade' variable to a high-resolution grid covering India. The 'cascade' variable will help in forecasting future landslide occurrence depending on precipitation levels over a certain grid.
We are a team of motivated junior undergrads and are looking to make use of what knowledge we have for deriving solutions to itching and persistent problems to make the world a better and safer place.
Describe what is the need of this project?
Landslides are on the rise around the world and India is among the most-affected countries, accounting for at least 28% of such events over the last 12 years. Most of which occurs due to heavy rainfall.
Since we plan to cover a very large area to model and train our dataset over, the rainfall data over the entire region may not be readily available and we might need to fill in or extrapolate the missing gaps using a predictive model.
Collecting data (with complete parameters) of the landslide region- crowdsourcing might be challenging as people might not acknowledge all the geomorphological parameters involved while reporting such an incident.
Describe who is affected?
The primary stakeholders that hold our main focus are the people living in areas where such landslide incidents take place, as they are the ones who suffer the damage of property and many times, their lives. We can use this point to further educate and incentive the people of this region to contribute to our application platform, any such event witnessed in the region.
The other stakeholder would be the local governing body, as with a predictive model for such events, they could deploy their resources or needs as per the requirements and execute their strategies accordingly. And if successful, we can incentive them to use the Government framework to carry out the task of reporting the incidents on our application to further strengthen the model.
What are the causes?
I believe the major causes of landslide occurrence are as follows:
Sloped terrains contribute to a major part of rural India and because of this reason, proper slope stability testing is not performed before the construction of houses there. This increases burden on the soil and it becomes more prone to shear failure. Without proper engineering practices, damage on the soil increases and ultimately landslide occurs with even a little rainfall.
Sometimes over excavation at sites weaken the soil above it leading to a greater shear force on the terrain. Proper excavation techniques must be practised, along with the presence of a geo-technical consultant.
When heavy rainfall occurs, soil porosity increases which decreases the shear strength, ultimately leading to shear failure. Hence, it is of utmost importance to forecast heavy rainfall and estimate the shear capacity of the terrain.
What is the evidence? Also who can you interview? What can you find out? What experiment can you run?
I believe this because in the recent years, India had been worst hit by the Kedarnath landslide which took place in 2013 in Uttarakhand state. It was caused by floods in the region due to constant heavy rainfall. This incident alone nearly took lives of about 5000 people. It gives enough motivation to devise a system that can predict landslides some time before their arrival with high accuracy.
A lot of information can be extracted by interviewing the geotechnical engineers who must have performed a detailed analysis of the region post landslide. This will help us to identify the importance of geological parameters and how they govern a landslide.
To dive deeper in understanding why landslide would have occurred, we can take the soil sample of that site and perform various tests such as shear strength test, test to measure porosity, etc. This will give us enough information to conclude the relevant parameters due to which landslide had occurred or landslide occurs.
What is The Big Idea? What is the value proposition?
In the past, there have been several studies which mapped the region with a degree of landslide susceptibility. The following research paper uses a heuristic approach to model the susceptibility of landslides:
Stanley, T., and D. B. Kirschbaum (2017), A heuristic approach to global landslide susceptibility mapping, Nat. Hazards, 1–20, doi:10.1007/s11069-017-2757-y
Our idea is to combine similar study along with AI and crowdsourcing techniques. We aim to collect regional data from citizens though an application and also from local government bodies which have previously worked in the region for infrastructural projects. We will use AI tools to map the landslide susceptibility all over India.
Another sub-problem can be to forecast rainfall over a region(instead of relying on Indian Meteorological Department) and then predict landslide with that data.
What is the mechanism of beneficial change?
Our approach will use the past landslide and rainfall data from news reports, government bodies and real time crowdsourced data through an app. We will then train our model with the data and predict landslide occurrence using soil and terrain parameters and forecasted rainfall as input parameters to get a reliable prediction.
We believe that our approach will have an impact because we will collect all the landslide governing parameters for a region either through crowdsourcing or news reports. In the past, individual studies have been conducted like mapping landslide susceptibility region, filtering of parameters that cause landslide in a region, but here we aim to combine these studies and produce a common model with which we will be able to quantify the haphazardness of landslide going to occur in a region with some amount of forecasted rainfall.
What are the key metrics?
In our model, we propose to distribute our area of focus into small unit grids. Then we plan to assign each grid with 2 values.
For the first value, we propose to coin a new variable - 'cascade', from scratch. This value would take into account various geomorphological and climatological parameters such as soil type, slope, topographic curvature, distance from drainage and lithology. And by determining the weighted influence of each of these individual parameters, we would assign a normalized (on a scale of 0.0-1.0) value to each of the grids. Note that the 'cascade' value would remain the same if we assume no major changes in the geomorphological characteristics of the area being analysed, which seems a valid assumption.
As for the second value, we would club grids having a similar range of cascade values and then assess what would be a 'Threshold Rainfall' value for a particular cascade value, any rainfall recorded above this threshold would signal an alert for a probable case of an impending landslide. This 'Threshold Rainfall' value would be determined by deploying our Machine Learning Algorithm on the acquired data set of 'Rainfall Vs Region (in the events of landslide)' from local authorities or governing bodies or NDMA of India.
Who is most likely to be supportive?
We can get the maximum encouragement and inputs from:
1. Local Government Bodies/ Authority of Disaster Management of the region: They have the incentive to help us provide the data because once this project is up and running, it would help these governing bodies to better plan their disaster management techniques and hence reduce any hance of casualties or damages.
2. Other Researchers in the domain: We can approach other researchers working in a similar field be it a Geotechnician or Hydrotechnician, to help us understand their model if we find it fruitful to deploy their research into our project. Moreover, there are many faculties at our college with whom we have face to face touch with and we can take their help, to propel the feasibility of our project.
Key foes? Who is most likely to oppose?
Although, there are no exact 'foes' relevant to the purpose of our project, there might be a few itches down the path due to:
Who has to do what. to make it happen?
who else is in the field?
The following stakeholders have worked or are working in this field:
1. NASA- They have developed a Global Landslide Hazard Assessment model, which assess the hazardousness of a landslide event and predict future landslides on the same region. This model is not effective where there is no landslide observation.
2. Various citizen scientists have conducted post landslide research on areas to identify the cause of a particular landslide. A number of research papers have been published in reputed journals with chunks of information on landslide occurences.
3. Local govt. authorities of landslide prone areas have been researching with the help of geologists to find out what are the primary and secondary factors that govern a landslide.
We will combine those individual researched to produce a model that can be used widely.
What's wrong? Missing? Not working?
My approach is better because, we plan to predict landslide occurence even at a place when there is no previous observation. Our AI model will analyse the geographical parameters such that it will be able to tell a threshold precipitation value, given all the geographical information at that place (which will be collected by us as described before).
Our approach will also involve Indian Meteorological Department in the picture so that we can use the forecasted rainfall data and alert the citizens beforehand, saving their lives from a landslide event.
We will also take into account the secondary observations that may induce a landslide, for example- loosening of soil or sunken roads near a sloped terrain.
The most crucial part of our project will be to identify the information on geographical parameters of a region, which we plan to collect through crowd-sourcing technique(developing a multi platform app), consulting geologists, past researches by local govt. agencies etc.
Once we are able to collect this information, we can start with applying AI techniques, to come up with a model that can be used to alert citizens.
Physical and intellectual resources needed (besides financial resources)
Physical and intellectual resources include the following:
1. Physical interpretation of various geological features of a terrain such as soil type, permeability, porosity etc.
2. Expert geologists to highlight the most important landslide governing factors.
3. A multi platform application developer.
4. Previous knowledge in the field if AI so as to interpret the working of already available AI tools.
5. Previous know-how of modelling a data set so as to use it for prediction through AI tools.
6. Help from local govt. bodies to better understand the behaviour of sloped terrains.
Next steps? Pilots?