Anagnor

https://bit.ly/anagnor_project_demo

https://github.com/anagnor/Anagnor

Anagnor is a publicly available system to provide the users a predictive analysis of occurrence of landslides all across the world. It's a predictive model that maps and analyses land features, global precipitation patterns, uneven heating of land and tectonic activities to predict the occurrence of landslides around the globe. An easy-to-use integrated system allows anyone to seamlessly access data from the past, the present, and the future. Anagnor also allows authorities and civilians to take advance action, helping minimise the damage caused by landslides around the world. It also acts as a tool to spread awareness and leads a path to further modification and worldwide usage.

Unlike conventional landslide detection systems which mainly focus on rainfall and surface-water erosion, Anagnor focuses on several other factors that can also lead to major landslides like uneven heating of land, floods and earthquakes. The system can not only explain data to professional mitigation authorities but also to local people about such an imminent catastrophe.

Landslides are caused when land erodes due to its soil structure, water carrying capacity and external factors like rainfall, floods, uneven heating. Erosion causing landslides generally occurs from a higher to lower slope. Anagnor integrates the data from global precipitation and thermal soil mapping datasets to find the probabability of a landslide occurring.

The predictive model takes several other factors and analyzes it using 3D Convolutional Neural Networks. All four datasets are used in NetCDF format. NetCDF format is an array/matrix-like format, where the two dimensions are latitude and longitude. This is easily readable in Python and can be shown on the system that is comprehensive enough to the user.

Collaborating with locals, Anagnor also provides a cost effective underground device that needs to be installed in the underground water table to predict the tectonic activities in that area. ISince tectonic activities is another major cause for landslides, Anagnor’s device helps the system get a more accurate prediction of the occurence of a landslide. However a technology that can predict tectonic activities in the present world has not really been developed, Anagnor brings a concept of achieving so by measuring and comparing concentrations of radon and its daughter products.

This is done by taking radon-dissolved underground water into a degassing chamber to extract radon that will be then transported to a Lucas scintillation chamber through a tesla valve. The tesla valve just ensures that there's no reduction in fluid pressure and velocity while going from degassing chamber to the scintillation cells. In the scintillation chamber, Alpha particles emitted, will collide with the surface made up of Zinc sulphide to generate light. Such light signals can be measured through a photomultiplier tube which transfers data about probable earthquake that might cause a landslide.

Anagnor has a very user-friendly user interface that helps user to easily access data. This helps to alert the locals and the authorities in a particular location to evacuate the landslide prone area and take necessary steps to mitigate/prevent landslides. Thus saving billions of dollars worth property and thousands of lives.

The only reason behind this project is to save lives and property that is damaged every year by landslides occurring all over the world. A predictive model like Anagnor helps the civilians know when a landslide might occur and thus evacuate the place in time / take steps to prevent such a grand calamity.

GISTEMP 4.0, Python, React were used to develop the model and the system. The machine learning and data preprocessing models used Pandas, Numpy, Pytorch, Torchvision and NetCdf4 module. The hardware was designed on blender while all the UI design and motion graphics were made on Figma and Adobe After Effects.

Anagnor uses data from rainfall patterns, surface temperature, underground radon concentrations and land feature of that location give a probability of a landslide occurring in that location using 3D convolutional Neural Networks.

All four datasets are used in NetCDF format. NetCDF format is an array/matrix-like format, where the two dimensions are latitude and longitude. This is easily readable in Python and can be shown on the system that is comprehensive enough to the user.

The following datasets were taken from NASA earth observatory site -

1) NASA’s Tropical Rainfall Measuring Mission (TRMM) Dataset - Precipitation

2) NASA's GISS Surface Temperature (GISTEMP) for surface temperature feature

3) NASA's NLDAS for elevation and slope data features

We used this as we saw that Surface Temperature and Global Precipitation are one of the major causes for landslides. Surface Elevation and slope gradients were required to increase accuracy by determining whether increase in rainfall and disparities in thermal data is indicating towards a landslide prone area or not. This is because Landslides only occur in sloped gradients from a higher to lower slope.

In the last 48 hours, the entire team developed many new skills. They explored and researched different approaches and nasa data to make machine learning models, etc. The team chose this challenge because of an ardour for machine learning and data visualisation. Team members were all STEM Enthusiasts who were passionate about scientific research and coming up with brilliant ideas to tackle problems which further can save lives and benefit the society as whole.

We viewed different datasets and models that could serve our purpose. We even had to change our approach for making a predictive model when we were unable to make our system distinguish between a coastal region and a hilly region (since landslides don't occur on beaches).

Another major challenge we faced was finding the global precipitation dataset from NASA where datasets were in form of images that needed computer vision algorithm to map it according to coordinates of longitude and latitude. Although time consuming, the team did accomplish in completing this task.

In the end, we would like to express our gratitude towards our teachers Mr. Ajithkumar K.G and Mr. Mukesh Kumar who constantly mentored us throughout the development process of Anagnor. We are also grateful to NASA SpaceApps team to give us this opportunity to participate in such a grand event that helped us to hone our skills and learn a lot in return as well.

Datasets for total rainfall patterns across the globe : NASA. (n.d.). Global Precipitation Dataset. NASA Earth Observatory. Retrieved October 3, 2021, from https://earthobservatory.nasa.gov/global-maps/TRMM_3B43M

Datasets for Thermal readings of land across the globe : NASA. (n.d.-b). Surface Temperature. GISS Surface Temperature. https://data.giss.nasa.gov/gistemp/

Datasets for Land Features of a location : NASA. (n.d.-a). Elevation and Slope. NLDAS Elevation Datasets and Illustrations | LDAS. Retrieved October 3, 2021, from https://ldas.gsfc.nasa.gov/nldas/elevation

Radon Detection Concepts and Research : Sethy, N., Jha, V., Ravi, P., & Tripathi, R. (2014). A simple method for calibration of Lucas scintillation cell counting system for measurement of 226Ra and 222Rn. Journal of Radiation Research and Applied Sciences, 7(4), 475–477. https://doi.org/10.1016/j.jrras.2014.08.002

#hardware, #blender, #figma, #motiongraphics, #machinelearning #datavisualisation, #python, #art #designing, #science

This project has been submitted for consideration during the Judging process.