NASAfe

https://docs.google.com/presentation/d/1R2deIBVpTD8-V17anVi8K5I7vnIftm3RZCUPCWXgg9U/edit?usp=sharing

https://drive.google.com/drive/folders/1Diutv23Uu4Pau4IEvNm7io21kISSA_3O?usp=sharing

Problem statement:

Landslides are an ubiquitous geologic hazard that are more widespread than other geological events in the world. With the recent climate change and rising temperature, the risk of landslides are expected to grow significantly in the next few years.

Landslides pose destructive hazards that threaten both urban and rural populations around the world, which have affected 4.8 millions people worldwide and caused over 18,000 deaths within a span of 20 years. In addition, destruction of infrastructure, damage to properties and loss of natural resources are also part of how the community has been negatively impacted by landslides. In the event where the victims have survived the landslides, the debris may trap or bury the survivors; partial or complete blockage of road also poses difficulties in the evacuation of the local community from the location of the landslides and inability of the rescue team to locate the victims. In fact, the most common cause of death in landslides is due to suffocation in entrapment, in other words, getting out of the landslide area increases your chance of survival.

How can we help?

Our Project offers help in 3 phases:

Phase 1: As preventive measures before the landslides, we utilise the historical satellite imagery and semantic segmentation model to indicate the landslides-prone areas, to perform landslides monitoring, and suggest a few options of safe haven. Using our path planning model, we identify the shortest escape paths to the nearest safety area. 

Phase 2: When landslides do take place, we provide an augmented reality navigation based on the escape paths designed in Phase 1. Crowdsourced data is collected via the users’ cameras to obtain real-time updates on the local situations. This is made available to the other users within the coverage area. In the event where there are blockages that cause a detour from the predefined escape route, the user may explore a new route in real-time. Once the user has completed the new route and marked safe, the server will update the new route to feed the path planning model. 

Phase 3: After the landslides, rescue teams are able to locate the trapped users by using the SOS feature in the application. The authorities could utilise the images captured by the public to identify the level of destruction caused by the landslides.

How does it work?

Our Project NASAfe employs Landslide Hazard Analysis for Situational Awareness (LHASA) model that utilizes Global Landslide Catalog integrating the surface susceptibility and satellite-based precipitation data from the Global Precipitation Measurement (GPM) to provide an indication of potential landslide hazards at the global scale every 30 minutes. This model identifies the occurrences of potential landslide activity in near real-time, enabling our application to be used as a tool for identifying landslide prone areas. This allows the users to prepare themselves to escape from danger. 

Figure 1: Landslide susceptibility map produced from the LHASA model

In addition, NASAfe gathers Landsat Collection 1 satellite images capturing the entire Earth’s surface at a 30-meter resolution about once every 16 days, generated from the Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS). On top of that, high-resolution multispectral images from the Sentinel-2A satellite with a global 5-day revisit frequency are gathered in this project in order to monitor and assess the soil and water cover, land cover change, as well as humanitarian and landslide disaster risk. At each update of the above-mentioned satellite images, the latest data is fed into our self-developed path planning model of NASAfe through a spatiotemporal Graph Neural Network (GNN) to find the shortest escape route for the users to a safe area before or during the occurrence of a landslide.

Figure 2: Sentinel-2A satellite image of the landslide prone area

Due to landslides occurring just in the blink of an eye, crowdsourced spatiotemporal data is collected automatically via the user's camera during their navigation to the safe place. The central database includes the GPS location, roadcuts, slope gradient, elevation, and direction, and is then integrated with the built-in data of above-mentioned satellites images to provide the most recent post landslide data to the other users. Extending from the real-time crowdsourced data and satellite images, the GNN model is able to promptly predict and recommend the shortest and safest new route to the users from time to time. In the event where a new route is to be attempted, the users can also record and mark their waypoints, enabling other users to follow the trail.

NASAfe also provides users with the SOS features in which they could make an emergency call to seek for immediate assistance if themselves or other individuals in the neighbourhood are found injured, trapped, or lost during the evacuation. From there, the SOS notifications would be sent to the authorities for the fastest rescue work. When the users have safely escaped from the landslide area, they would also be able to mark themselves safe in the app.  

Benefits of the project: 

Preventive: 

• Hazard management by indicating potential landslide areas and designing emergency and evacuation routes;
• Alert users in advance prior to the occurrence of landslides to ease evacuation missions for local communities; and
• Alert the neighboring community by triggering warnings and keep them informed on the potential hazards and indicators of landslides.

Prescriptive: 

• Identify trapped or injured victims through the SOS feature to direct rescue teams to their locations;
• Navigate neighboring communities and landslide victims to proceed to immediate evacuation of affected communities; and
• Minimise casualties resulting from landslides events.

Collaborative:

• Using crowdsourced data to collect real-time information and update information to pave the best possible escape route for the users; and
• Use real-time maps to assist in determining the affected areas of the landslides and aid the recovery assessment of post-landslides events.

Tools, coding languages and software used to develop the project:

- Java

- JavaScript 

- Python 

- Tensorflow

- Android Studio

- Firebase

- Figma

A. Landslide Hazard Analysis for Situational Awareness (LHASA) model

The LHASA model is the default prediction model in the NASAfe app, giving near real-time estimations on potential landslide occurrence location.

B. Global Landslide Catalog

This catalog is incorporated in the LHASA model for making estimations.

C. Landsat 8 and Sentinel-2A satellite images

Images of land surface conditions from both satellites are combined to provide information like debris area, road blockage and general landscape change in order to facilitate the model in crafting up-to-date evacuation routes.

Space Apps experience:

No physical meetups, no interactive sessions, no physical pitch sessions… These are just some of the perks of a virtual hackathon in 2021 amidst the COVID-19 pandemic. But team NASAfe definitely had an interesting and fun ride with SpaceApps as we went through each phase of a virtual hackathon just like everyone else. From forming a team of people with a diverse set of skills to realizing the fact that every single solution or idea has been attempted at least once, the team still saw many opportunities in making the best use out of the resources given. Instead of brainstorming sessions, we had ideation sessions. Instead of nerve-wrecking presentations, we told a story. Instead of coding an entire app, we attempted to tackle a problem creatively. Instead of forcing a new feature into the solution, we put ourselves in the shoes of the users. Instead of stressing out, we, most importantly, had fun! Make that the power of FIVE, NASA!

Inspiration for selecting this challenge: 

In our country Malaysia, heavy rainfall has caused landslide disasters in different states although it is not a precipitous country (mountains and hills are less than 25% of the terrain). The disasters caused loss of lives, injuries, destruction of infrastructure, damage to properties, and loss of natural resources (Rahman & Mapjabil, 2017). 

For example, alone in one day on August 18, 2021, at least four landslides had been detected on the slopes of Gunung Jerai, Kedah based on the initial observation by the Disaster Task Force of Department of Mineral and Geoscience, Malaysia (JMG). However, a total of four lives were still lost following the water surge phenomenon with the mountain terrain at an angle of more than 25 degrees. 

In the above scenario, if the warnings and prompt evacuation plan as well as escape routes were shared to the nearby villagers once the potential landslide incidents were detected by the Task Force, the four deaths and other destruction could have been avoided. 

Such similar incidents due to the lack of prompt warnings and escape routes have occurred worldwide every year. This has inspired us to develop this NASAfe prototype that combines satellite images and crowdsourced data to provide the prompt alerts and real-time escape guide to the users to mitigate the risks, potential human losses, and cost of impacts of landslides. The authorities could also utilize the data for emergency evacuation and rescue work. 

Approach to develop this project:

The present digital era has made every person dependent on smartphones, proven by the surging use of mobile applications with an increasing level of acceptance by users. Therefore as compared to widely available data on the internet, most people prefer the use of mobile applications as they are easily accessible and user-friendly.

As a core approach to Landslide Disaster Risk Reduction, we aim to develop an application loaded with the LHASA model and crowdsourced data to plan evacuation routes during the occurrence of a landslide.

The escape route is planned using a graph neural network (GNN) to identify a safe and unaffected location situated away from the landslides area. The GNN model is also used to identify potential shelters and flat areas such as schools and hospitals, in order to allow victims to seek immediate help. 

Crowdsourced data collection will be overlaid in real-time camera feed in the format of augmented reality (AR) to guide directions and serve as an update to the existing escape paths. This will then ensure a more accurate and up-to-date navigation for the users at all times. 

Challenges faced during the process:

- Difficulties in figuring out the required data from the space agency due to lack of science knowledge on space as our team is all from other fields of study.

- We wanted to insert a universal design feature so that the application is feasible and suitable for individuals with special needs too. For example, voice-based guidance and location indication systems could be beneficial to users with visual-impaired profile so that alerts could be sent immediately to their respective emergency contact when they report they are injured or trapped. However, due to time-constraint, we were only able to create a prototype without the feature.

Resolving setbacks and challenges:

- We started our team discussions one week earlier and spent hours digesting the provided resources as best as we could.   

Our concerns / future development:

- Low spatial and temporal resolution of image data from the satellites would render inaccurate predictions on landslide prone areas.

- Accuracy of path planning depends on the quantity and quality of crowdsourced data and crowdsourced data is not immediately available.

- The prototype could be improved by adding a universal design feature so that it could also benefit users with special needs. 

- We could verify our existing mobile application data with the input from the local government.

- There might be limitations of technology and lack of mobile device resources in under-development countries or areas, the local government could look into the issues so that the tool is more feasible by the public.

Earth Engine Data Catalog - Landsat Collections: https://developers.google.com/earth-engine/datasets/catalog/landsat

Earth Engine Data Catalog - Sentinel Collections: https://developers.google.com/earth-engine/datasets/catalog/sentinel

Google Earth Engine: https://code.earthengine.google.com/

Landslides - http://www.fao.org/emergencies/emergency-types/landslides/en/

Landslides - https://www.who.int/health-topics/landslides#tab=tab_1

Landslides Disaster in Malaysia - An overview: (PDF) Landslides Disaster in Malaysia: an Overview (researchgate.net)

Landslides Toolkit - https://earthdata.nasa.gov/learn/toolkits/disasters-toolkit/landslides-toolkit 

LHASA - https://github.com/nasa/lhasa

Mineral and Geoscience Dept detects four landslides in Gunung Jerai: Mineral and Geoscience Dept detects four landslides in Gunung Jerai | Malaysia | Malay Mail

NASA’s Open Data Portal - Global Landslide Catalog: https://data.nasa.gov/Earth-Science/Global-Landslide-Catalog/h9d8-neg4/data

Satellite-Based Assessment of Rainfall-Triggered Landslide Hazard for Situational Awareness - Satellite‐Based Assessment of Rainfall‐Triggered Landslide Hazard for Situational Awareness - Kirschbaum - 2018 - Earth's Future - Wiley Online Library

USGS - Landsat Collections - Landsat Collection 1 (usgs.gov)

#landslide #crowdsourcing #augmented_reality #AR #ARnavigation #escape_guide #escape_route #evacuation #satellite_images #prediction

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