EASy - Hazard Detection| Automated Detection of Hazards
Automated Detection of Hazards
Machine Learning to Monitor Coastal Erosion
Summary
In our project, we attempted to build an algorithm to interpret multispectral images from satellites to determine the coast-ocean boundary. By analysing the change in this boundary over time, the algorithm would tell where coastlines were retreating due to erosion and display this on an interface for researchers and policy-makers. Overlays that we intended to include were mappings of coastal defences, soil composition, and soil moisture.
How We Addressed This Challenge
We developed an interface for the program, but most of it was hard-coded due to difficulties we faced in implementing machine learning. The program looks at the NIR and green spectral frequencies of pixels and calculates each one's Normalised Difference Water Index (NDWI). NDWI values vary from -1 to 1, with values closer to 1 being more likely to be water.
In future, the algorithm could build on many satellites images to mitigate the effects of noise and cloud cover and gain a probability for a given pixel being ocean or land. Over time, the changing of pixels from land to water will show coastal erosion and allow it to be monitored in an accessible and easy way across entire coastlines. This can also be used to judge the effectiveness of coastal defences and assess which areas are most at risk due to soil composition or location. It will help policy makers to determine where and how to distribute resources.
How We Developed This Project
Our team chose this challenge because of the unique nature of the hazard of coastal erosion and its effect on the UK, where we study. We felt that it was often overlooked by remote sensing, and must be considered considering its potential for financial loss and the investment in prevention.
Our approach consisted of first brainstorming which hazard we wanted to address, and then dividing the deliverables into smaller tasks that individual team members worked on.
Java was the language used. IntelliJ was the IDE used to create the program and Java Swing was used to create the interface.
The main problem we faced was dealing with the amount of data available! It was difficult to know where to start and which agency to take our MSI from. We then faced the issue of interpreting the MSI data, which made it difficult to develop a full machine learning algorithm in the time allotted.
Our achievements include determining a way to measure coastline erosion from MSI, as well as a working interface!
How We Used Space Agency Data in This Project
We used NASA's worldview to identify areas of interest for training the algorithm, then collected data from ESA's Sentinel 2 to obtain values for multispectral indices.
Project Demo
https://github.com/JessicaMcclure97/NASA_Sapce_App_2020/blob/main/Coastal%20Erosion%20Hazard.pptx
Data & Resources
NASA Worldview
ESA Sentinel 2 Data
https://www.sciencedirect.com/science/article/abs/pii/S0034425719301531