Astra

Huntsville, AL | Identifying risk with science + communities

Modeling Landslide Risk in the Philippines

High-Level Project Summary

Through the Space Apps Challenge, we developed a way to analyze geological data and find the corresponding risk of landslides in that area. In the process, we also created a data pipeline we can apply to any other area of the world. The use of machine learning means it can also adapt to any other area of the world. The website we made provides an easy-to-use visual, describing which areas of the Philippines are susceptible to landslides. This can be extremely helpful to local governments as it can identify if an area needs an evacuation plan in the case of such an event. Moreover, our app provides geological data about each specific area, so if needed, it can be extracted.

Link to Project "Demo"

https://docs.google.com/presentation/d/1pzAGRTku6jL3mjOgX8KGzmhh4RS1ILRD0niHXmaFkOg/edit?usp=sharing

Link to Final Project

https://github.com/katsukixyz/astra

Detailed Project Description

Landslides are common global phenomena that can have major impacts on local economies. Recently in the US alone, Landslides cause an estimated $2 - 4 billion annual cost from infrastructure, homes, contamination, etc., and have a death toll of thousands worldwide per year. Regions susceptible to landslides are characterized by erosion, weakened soil, and earthquakes, and while we know these factors increase the likelihood of a landslide, it is still challenging to predict and prevent landslides. In this project, we identified several parameters of interest and developed a model from historical data of landslides in the Phillippines. Finally, we generated a landslide risk map using our model.

Landslides are not only damaging but also unpredictable, making them difficult to prevent and avoid. In this project, we analyzed data of previous landslide incidents and developed a tool that maps the risk of landslides over the Philippines area. Our model uses Logistic Regression on 4 variables (slope, aspect, distance to geological boundary, distance to faults) and is able to easily predict the risk of a future landslide occurrence. With further development of tools like ours on a global scale, we hope that landslides will become more predictable and as a result less damaging.

Initially, we wanted to understand the environment that creates high risk for landslides. Thus we looked into a lot of papers that analyzed the geographical factors that lead to landslides. This led to our use of the 4 variables, slope, aspect, distance to geological boundary, distance to faults (aspect is the cardinal direction the slope faces). After settling on these variables, we found datasets for each of these 4 variables and used machine learning to develop our model to determine risk.

Landslides have a great financial cost and death toll globally, and similarly to events such as earthquakes, they are spontaneous and sometimes hugely damaging. We wanted to create a simple tool that visualizes the risk zones of landslides to make these events more predictable.

First, we began looking at papers on landslide analysis to determine what kind of methods they used to find risk. After reading the papers, we found the most important factors that contribute to landslides: slope (degree of incline), aspect (what direction the slope is facing), nearest distance to fault, nearest distance to geological boundary and drainage density.

Second, we decided to focus our scope to a specific area or country to decrease the amount of data we have to work with and specialize our analysis. We ended up working with the Philippines since it is a country that generally suffers from landslides.

Next, we searched for datasets with these values and found all but drainage density. Drainage density was a rather uncommon data point. We compiled all of the data into a single file with the corresponding latitude and longitude. The research paper we found used a Logistic Regression model and performed well, so we decided to use it as well.

We obtained data points for landslide occurrences and then chose a similar amount of data points from places rather far from landslides for unlikely landslide locations. We trained the Logistic Regression model on these data points, then applied it to a grid of points contained in the Philippines to generate risk across the country.

For the front-end web application, we used ReactJS to construct it and hooked it up to an ExpressJS backend that was connected to a PostgresSQL Database. The frontend retrieved the data from the database and displayed it using various component libraries like Mapbox. For right now we serve it as a static site by having the database be client-side and rendered through github pages.

Space Agency Data

We used data about the slope and aspect of the world from NASA as one of the factors in our model. We also used a lot of data from the USGS about fault lines, historical landslide data, and more about the geologic makeup of the globe and specifically the Philippines. We needed to get data to build our model and were very pleasantly surprised that we were able to find so much of it through NASA and other US Government agencies' data portals.

https://www.americangeosciences.org/critical-issues/faq/how-much-do-landslides-cost-terms-monetary-losses
https://www.usgs.gov/faqs/how-many-deaths-result-landslides-each-year?qt-news_science_products=0#qt-news_science_products

Hackathon Journey

Our Space Apps experience was interesting in that we came across challenges where we didn’t expect any, and found some problems to be easier than we had expected. One such challenge that gave us a lot of problems was data collection and data processing. It was difficult to find comprehensive datasets for overlapping areas, and it was difficult to process this data into a useful format that allowed us to use it in our model. From these problems, we learned different methods to optimize data/distance analysis and to allow different datasets to correlate with each other.

We decided to choose this challenge since lack of landslide risk analysis is a problem for countries and this leads to infrastructure built on unstable areas.

Our approach to tackling this challenge was to first find and process relevant data, use this data to create our model, and then use our model to calculate risk for the Philippines and create a map that visualized this risk.

Our team resolved setbacks and challenges by having all team members collaborate on major issues that would be solved through more people (ex: finding data) and splitting up tasks for problems that could be approached at the same time (ex: frontend and backend).