Slidetect

https://drive.google.com/file/d/1TcWUmnwW6n0b8ZmE-H-iJHHKWLBNF9Mj/view?usp=sharing

https://github.com/mona-14/SliDetect

Landslides happen more often worldwide than any other geological event, causing an estimated 18000 deaths in the past years. These landslides have affected many rural communities by compromising their social and economic systems. The current risk assessment maps of landslides are not effective and do not allow actions to try to mitigate the risk at the local level. Thus, to tackle this issue, we have done preliminary research that indicated that there are three crucial environmental factors contributing to the occurrence of landslides worldwide, including precipitation rates, elevation, and the time of the year. Built on the basis of this preliminary research, we will develop a machine learning model, a convolutional neural network specifically, to predict the future occurrences of landslides. Training datasets built from the three environmental factors tracing the 30 days leading up to a landslide and the occurrence of landslides will be inserted into the convolutional neural network (CNN). These training datasets will be created from open-source, reliable NASA resources. After the CNN is deployed and trained, it will be integrated into a user-friendly, easy-to-access tool in the form of a website to allow local communities and governments to track the risk of landslides in their regions. Even more, this tool will be tested by using multiple test datasets from historical landslides as input to our model. The overall accuracy of our tool will be calculated accordingly. The final end product will be a website that is accessible to all rural communities.

When the user chooses their region, NASA’s reliable data will be downloaded from their servers and input into the CNN, calculating the risk for landslides in a specific region with the highest accuracy possible. To help assess the risk of landslides in rural areas, we did research into the environmental factors that contribute to the occurrence of landslides. We found that in a given region, precipitation rates, the elevation of land, and the time of the year strongly affect landslide susceptibility. This research represented the basis of our solution. Data on the three environmental factors used to assess landslides risk will be collected from open-source, reliable resources including NASA WorldView, NASA Earth Data, and other space agencies and local governments’ resources. The data will be collected as images depicting the levels of each environmental factor in a given rural area at a given time of the year. A data set that correlates past landslides and the changing environmental factors leading up to them will be used as the training set to train and evaluate a machine learning (ML) model that will assess the risk of future landslides. To develop the machine learning model, a deep learning (DL) framework will be deployed. A convolutional neural network (CNN) model will be utilized to detect the changes in the levels of environmental factors in the image data. The CNN architecture is a multi-layered architecture, where the first layers detect simple patterns and the following layers detect more complex patterns. As previously mentioned, once the CNN framework is deployed, it will be trained to detect different levels of landslide risk using the past landslides data set we will build. Each layer of the CNN model will contain a filter corresponding to one of the environmental factors we are evaluating (precipitation, elevation, etc.).

The input layer will receive the data from the NASA resources and process it in the hidden layers of the CNN. The output will be calculated from the last layer in the artificial neural network. It will be divided into three possible risk levels: HIGH (red), MEDIUM (orange), and LOW (yellow). This CNN model will be integrated into a user-friendly website to communicate the detected risk level to local communities and governments. A user will choose a rural region they want. Then, data from the NASA resources will be extracted and entered as input in the DL model. The trained model will analyze it and the output, the level of landslide risk, will be displayed as a coloration of that region on a map, according to the colors corresponding to each risk level. Furthermore, this tool will be tested by using multiple test datasets from historical landslides as input to our model. The overall accuracy of our tool will be calculated accordingly. 

In our project, we have used multiple resources including the NASA resources which are provided on the website of the challenge. Our team used the data provided in the NASA land resources https://gpm.nasa.gov/landslides/resources.html we used the data from the different publications provided. We also used the data provided by the NASA landslide model https://gpm.nasa.gov/applications/landslides#modelingandreportinglandslides to get to know the features used in the model. The USGS model https://www.usgs.gov/natural-hazards/landslide-hazards/data-tools also gave us different features on different models used by different agencies and helped us to collect different data. 

Earth data was a beneficial website where we got to know different features of the earth https://worldview.earthdata.nasa.gov/ and besides, NASA Earth Data https://earthdata.nasa.gov/ provided us with different data used in the project. 

Besides, https://storm.pps.eosdis.nasa.gov/storm/: It was used to access the precipitation rates to create our training data set.

also, https://worldview.earthdata.nasa.gov/: NASA WorldView was utilized to obtain the elevation and precipitation data. And https://landslides.nasa.gov/viewer: It was used to pull past landslides occurrences across the world, and used as training data.

Our experience in NASA Space Apps was an extraordinary experience. We collaborated during the 48 hours of the hackathon to develop a solution to the challenge. We communicated using online applications and were able to complete the project on time.

During the hackathon, we learned many hard and soft skills. For example, we learned how to develop a machine learning algorithm, use python to embed weather APIs, and integrate this algorithm into a website. In addition, we learned how to work under pressure as the 48 hours of the hackathon were jam-packed with tasks to finish our project on time.

Our team was inspired to choose the challenge “Identifying risk with science + communities” because we learned that many communities around the world suffer from sudden landslides that cause death and injuries of many people as well as many losses. That’s why we chose this challenge to try to help these communities by predicting landslides before they occur.

Our approach to developing this project was to check what has already been done regarding the problem of landslides. After that, we used the space data provided by NASA to find correlations between these data. Then, we developed a convolutional neural network to predict future landslides from this data and embed it into a website.

Of course, our journey was not free of challenges and setbacks. For instance, we faced a problem when we had to use a weather forecast API to help us predict future landslides. We found out that this API was closed 5 years ago. However, we discovered that very similar data was provided from another space agency and it worked.

Last but not least, we would like to thank everyone who helped us in completing this project including all the organizers of NASA Space Apps Cairo. We would love to represent our gratitude for our team mentor, Lamia Hasan, who supported us throughout the hackathon.

• Tensorflow
• Adobe photoshop
• Microsoft Excel
• Fluid UI
• Adobe XD
• Anvil
• Python 3
• Visual Studio Code
• Spyder
• R
• Highland, L. M., & Bobrowsky, P. (2008). The Landslide Handbook—A Guide to Understanding Landslides. In The Landslide Handbook—A Guide to Understanding Landslides (USGS Numbered Series No. 1325; Circular, Vol. 1325, p. 147). U.S. Geological Survey. https://doi.org/10.3133/cir1325
• Kirschbaum, D., Kapnick, S. B., Stanley, T., & Pascale, S. (2020). Changes in Extreme Precipitation and Landslides Over High Mountain Asia. Geophysical Research Letters, 47(4), e2019GL085347. https://doi.org/10.1029/2019GL085347
• Kirschbaum, D., & Stanley, T. (2018). Satellite-Based Assessment of Rainfall-Triggered Landslide Hazard for Situational Awareness. Earth’s Future, 6(3), 505–523. https://doi.org/10.1002/2017EF000715
• Wang, Y., Fang, Z., & Hong, H. (2019). Comparison of convolutional neural networks for landslide susceptibility mapping in Yanshan County, China. Science of The Total Environment, 666, 975–993. https://doi.org/10.1016/j.scitotenv.2019.02.263
• World Health Organization. (n.d.). Retrieved September 24, 2021, from https://www.who.int/westernpacific/health-topics/landslides
• Mind’je, R., Li, L., Nsengiyumva, J. B., Mupenzi, C., Nyesheja, E. M., Kayumba, P. M., … Hakorimana, E. (2019). Landslide susceptibility and influencing factors analysis in Rwanda. Environment, Development and Sustainability, 22(8), 7985–8012. doi:10.1007/s10668-019-00557-4 

#landslide, #software, #data, #machine learning, #convolutional neural network, #deep learning, #user friendly

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